/

University of Southern California Dissertations and Theses

/

Intraocular and extraocular cameras for retinal prostheses: effects of foveation by means of visual prosthesis simulation

(USC Thesis Other)

Intraocular and extraocular cameras for retinal prostheses: effects of foveation by means of visual prosthesis simulation

PDF

Transcript (if available)

Content
INTRAOCULAR AND EXTRAOCULAR CAMERAS FOR RETINAL PROSTHESES:
EFFECTS OF FOVEATI ON BY MEANS OF VISUAL PROSTH E SIS SIMULATION

by

Benjamin P atrick McI ntosh

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCH OOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for t he D egree
DOCTOR OF PHILOSOPHY
(ELECTRICAL ENGINEERING)

May 2015

Copyright 2 015 B enjamin Patrick McI n tosh
ii

ACKNOWLEDGEMENTS
A great many people supported me throughout the process of this
research a nd t hroughout the writing proces s of t his thesis, a n d I am forever
grateful. First and foremost, thank you to my advisor, colleag ue, mentor, and
friend, Dr. Armand R. Tanguay, Jr. for all of the support throu ghout the years.
None o f this r esea rch would have b een p ossible without his guidance and wisdom
when t hings did not quite work a s expected, and his humor and s pirit when
things got downright difficult.
I a lso woul d li ke to gi v e a ve ry speci a l tha n k you to Dr. P a tri ck J. Nasiatka,
who provided m uch need ed a d v ice at k ey p oints along the way and who I
consider on e of my best friends that I have gained sinc e attend ing the Univ ers i ty
of Southern California. Thanks to his easy‐going attitude and breadth of
knowledge, the long nights and hard work became that much easier. Thank you
also to my other colleagues within the Optical Materials and De v i ces Laboratory
with whom I have had the honor of working, including Dr. Michelle C . Hau e r ,
Noelle R. B. Stiles, Furkan E. Sahin, Michael Caputo, Bobby Gro om, Jenna C r i s p ,
and Richelle Smith all of whom co ntributed to my success in som e major way.
I would also like to thank all of my colleagues and mentors wit hin the
Biomimetic M icroElectr onic S ystem s Engineer i ng R esea rch Center, e s p ecially D r.
James D. Weiland and Dr. Mark S. Humayun, who are also members of m y Ph.D.
committee. I owe in fin i te g r a t i tude t o both o f t h em f or l ea din g such a n amazing
iii

research o r g aniz at ion within U SC, and for making t his int e rd isc iplinary a nd
extremely rewarding research possible. Thank you also to my ot her committee
members, D r. B . Keith J e nkins and Dr. Alexander (Sandy) A. S awchuk. Thank you
to Dr. Bosco Tjan for the idea of visual search in evaluation o f the visual p rosthesis
simulator, and thank you to Diane Demetras for all of the guida nce throughout
the graduation process.
Lastly, I would like t o t h ank my friends and family for all of the support
throughout the years. Thank you to my mother and father, from whom I have
learned the perfect mix of w ork ethic and humility, which has n o doubt served
me w ell thr o ughout m y schooling a n d research. F inally, I want to thank my wife,
Stephanie, f or t he u nbelievable a m o unt of m otivat ion a n d suppor t that she h as
given me d uring the la st few y ea rs of this res earch.
iv

TABLE OF CONTENTS
Acknowled g ements i i
List of Tables v i
List of Figu res ix
Abbreviatio n s xxii
Abstract xxiv
Chapter 1 Introductio n 1
1.1 Background and Problem Statement 1
1.2 Intr aocular Retinal Pro stheses 4
1.3 Effects of Head and Eye Pointed Cameras 8
1.4 Intraocular Camera 1 4
1.5 Effect s of P i x ellatio n 1 6
1.6 Summary and Organizat ion of the Thesis 20
Chapter 2 Visual Prosthesis Simulator 27
2.1 Introductio n 2 7
2.2 Prior Art 28
2.3 Hardware Implementation 30
2.4 Software Implementation 35
2.5 Benchmarking a nd Lat ency 4 6
2.6 Visual Prosthesis Simulator as a n Eye‐Track e d Extraocular Camer a 55
2.7 Summary and Other A pplications 6 4
Chapter 3 Visual Psyc h ophysics E xperiments w ith the V i s u al Pros thesis
Simulator 70
3.1 Introductio n 7 0
3.2 Liter a ture R eview o f Human Subject Test i ng 7 2
3.3 Visual Sear c h 77
3.3.1 Introductio n 7 7
3.3.2 Methods 77
3.3.3 Results 84
3.3.4 Discussion and Conclu s ions 9 1
3.4 Functional Reach and Grasp 9 5
3.4.1 Introductio n 9 5
3.4.2 Methods 95
3.4.3 Results 99
3.4.4 Discussion and Conclu s ions 105
3.5 Navigation and Mobility 107
3.5.1 Introductio n 107
v

3.5.2 Methods 107
3.5.3 Results 115
3.5.4 Discussion and Conclu s ions 118
3.6 Summary 119
Chapter 4 Hermeticity and Clocki ng of Implantable Bio m edical Circui t s
and Dev i ces 125
4.1 Introductio n 125
4.2 Prior Art 127
4.2.1 Posit i ve Mo b ile Ion Con t amina t ion Detectio n 127
4.2.2 CMOS Clock s 129
4.3 First Gener a tion Hermetic Co ating Test Chips 131
4.3.1 Hermetic C oatin g Tes t Chip 1 (HCTC‐1) 131
4.3.1.1 Performanc e Characteristics of CMOS Clock Circuits 137
4.3.1.2 Per f ormanc e Char acter i st ics of PMIC Sensors 147
4.3.1.3 Results Derived from Other HCTC‐1 Devices 152
4.4 Second Gen erat ion Her m etic Coa ting Tes t Chips 155
4.4.1 Hermetic C oatin g Tes t Chip 2 (HCTC‐2) 156
4.4.2 CMOS Clock Test Chip (CCT C) 161
4.5 Third Generatio n Her metic Coa t ing Tes t Chips 171
4.5.1 Hermetic C oatin g Tes t Chip 3 (HCTC‐3) 171
4.5.2 Hermetic C oatin g Tes t Chip 4 (HCTC‐4) 195
4.6 Summary o f Key Findings 200
Chapter 5 Fourth Generat ion He r m etic Coa ting Tes t Chips 205
5.1 Introductio n 205
5.2 Hermetic C oatin g Tes t Chip 5 (HCTC‐5) 206
5.3 Hermetic C oatin g Tes t Chip 6 (HCTC‐6) 215
5.4 Hermetic C oatin g Tes t Chip 7 (HCTC‐7) 225
5.5 Summary o f Key Findings 230
Chapter 6 Advanc ed I ntraocular C amer a Lens Designs 236
6.1 Modelling of the Biological Corne a for Intraocular Camera Testi ng 236
6.2 Next Gener a tion I ntr a o c ular Camera Lens Des igns 241
Chapter 7 Summary and Futur e Resea r ch Directions 248
7.1 Summary 248
7.2 Future Research Direc t ions 252
Bibliography 258
Appendx A Hermetic C oatin g Tes t Chips Cad e nce Layouts and MOSIS
SPICE Dev ice Simulation Par amet ers 270
vi

LIST OF TABLES
Table 1. Summary o f key pro s thetic v ision simulat i on r eferences
including c o ndit ions s imulated a nd p arameters r e lev a nt t o
this thesis research. 29
Table 2. Technical c omparison of Z 800 and zSight H MD p erformanc e
character i st ics. 3 5
Table 3. System s pecifica tions of t he V isual Prosthes is S imulator P C
used in ben c hmarking e xper iment s. 48
Table 4. Comparison o f minimum, a verage, and maximum proce ssing
times in m il liseconds fo r variou s code s egments of t he v isual
prosthesis s imulator. 49
Table 5. Comparison o f minimum, a verage, and maximum proce ssing
times in milliseconds fo r three d i fferent s y st em latencies. 54
Table 6. Comparison o f simulation r equire me n t s f or t h re e di ff ere n t
applications o f the visual p rosthesis simulator, a nd t he
relax e d r e quirement s f or a n eye‐t r acked extr a o cular camera
system. 58
Table 7. Summary of data collected during the visual search
exper i ment c omparin g t he h ead ‐ pointed a n d ey e‐pointed
camera modes for each experiment. The p ‐v alues displayed
in parentheses compare corresponding points for the head‐
pointed and ey e‐pointed dat a sets, w ith gr een values l ess t h an
or e qual t o a threshold of 0 .05 indica ting sig nific a nt
differ ences. 8 7
Table 8. Results of t he M ATLAB ANOCOV A tool, which compares t he
slopes and intercepts of the linear regressions of the head‐
pointed a n d eye‐point e d visual s earch time t o find t he t arget
data. V al ues in g r een i ndic a t e sta t ist i c a lly significant
differ ences with a thr es hold of 0.05. 9 2
Table 9. Aver age time t o grasp the ta rget o bject in t he f unction a l depth
perceptio n e xper iment s. T he stan dard er ror of t he m ea n is
shown in p aren theses b elow each aver age tim e . 102
Table 10. Percenta ge i mprovement a fter b aseline subtraction for
various functional depth percept io n exper i mental c ond i t i ons.
Any differences below a threshold p ‐value o f 0.05 hav e b een
highlighted with yellow, and replaced with values of zero
otherwis e. 104
vii

Table 11. Summary o f key dev i ces that w ere test ed o n h e rmet ic c oatin g
test chip HCTC‐1 including size, location, and presence of a
passivation o xide cut fo r each device. 133
Table 12. Simulated frequency a s w ell as m easured frequency, d uty
cycle, p eak‐to‐peak voltage, a nd dissipated power for six
differ ent CMOS clock generat ion circuits on HCTC‐1. 142
Table 13. Comparison of simulated waveform outputs for four different
capacitive l oads on the clock gen er ation c i rcuits on HCTC‐1. 14 6
Table 14. Change in the measured drain current and floating gate
voltage as H CTC‐1 chips are soak ed in a saline solution for 29
hours cumulatively. 152
Table 15. Summary o f key dev i ces that w ere test ed o n h e rmet ic c oatin g
test c hip H C TC‐2, incl uding siz e , location, a n d presence o f a
passivation o xide cut fo r each device. 158
Table 16. Summary of key devices that were tested on CMOS clock test
chip CCTC, including size, location, and presence of a
passivation o xide cut fo r each device. 164
Table 17. Simulated frequencies for the six var i ants o f the ring‐
oscillator‐based CMOS c lock o scillators, as i mplemented i n a
0.5 µm CM O S process and operated at 5 V. 168
Table 18. Summary of key PMIC sensors that were tested on hermetic
coating tes t c hip HCTC‐3, in cluding size, location, and
presenc e of a passiv a tion oxi de cu t for each device. 176
Table 19. Summary o f key dev i ces that w ere test ed o n h e rmet ic c oatin g
test chip HCTC‐4 including size, location, and presence of a
passivation o xide cut fo r each device. 197
Table 20. Resistance m easurements of k ey i nter dig i ta ted electrodes f or
two differ ent moisture sources. 199
Table 21. Summary o f key dev i ces that w ere test ed o n h e rmet ic c oatin g
test c hip H C TC‐5, incl uding siz e , location, a n d presence o f a
passivation o xide cut fo r each device. 208
Table 22. Summary of key PMIC devices that were tested on hermetic
coating tes t c hip HCTC‐6, in cluding size, location, and
presenc e of a passiv a tion oxi de cu t for each device. 217

viii

Table 23. Summary of key PMIC devices that were tested on hermetic
coating tes t c hip HCTC‐7, in cluding size, location, and
presence of a passivation oxide cut for each device. All six
PMIC sensors were mirrored to comprise an additional set of
six sensors on the bottom half of the chip. 227
Table 24. Summary of RMS spot sizes for various field angles when
comparing the original i ntraocular c amera design u s i ng t he
Liou and Brennan eye model and the OmniVision OV6920
image s e ns or, the s a me c amera design i magin g i n a i r (no
cornea/aqueous humor model), and the same camera design
using the co rnea r eplac e ment mod el. 240

ix

LIST OF FIGURES
Figure 1. Schematic diagram of the human eye and retina showing key
anatomical s tructures and cellular layers o f t h e ret i na; from
[10], after [12]. 2
Figure 2. Component s o f a retina l prosthesis i ncluding a v ideo c amera
and a micro s timulator array ; from [9]. 5
Figure 3. (Left) I llustrat i on o f an e ye w ith implanted ep iretinal
microstimulator array; from [10]. Note that the diagram of
the ey e has been r o t at ed c ounter clockwise b y 90°, a n d the
ultra‐flexible ribbon cable rout ed in the opposite hemispher e
relat i ve t o its usual placement, for clarity of illustration.
(Right) Fundus p hotograph of a n implant e d 60 el ectr ode
array as part of the Argus II devic e ; from [9]. 7
Figure 4. Intr aocular ret i nal p r osthesis w ith ex ter n ally m ounted
(extraocula r) c amera. T he diag r am o f the eye has been
rotated by 90° for clarity o f illustr a tion ; from [10]. 8
Figure 5. Simulation o f the effects of f oveation o n a subject with
Retinitis Pigmentosa t hat has been i mplanted w ith a retinal
prosthesis. ( Top Left) Image from t he s cene c amera with t he
centr a l field of v iew o utlined in r ed, and th e direct ion of g aze
outlined i n green. ( Top Right) T he v isual fiel d with n o visual
prosthesis. (Bottom Left) Head‐pointed case in which the
image from the central FOV appears in the direction of gaze.
(Bottom Right) Eye‐pointed case in which the region of the
ca m e ra F O V th a t is aligned with the direct io n of gaz e ap pears
in the d irec tion of g a ze. 10
Figure 6. Simulation o f the effec t s of f oveation o n a su bject with A ge‐
Related Macular Degeneration t hat has been i mplanted w ith a
retinal pros thesis. ( To p Left) Ima g e from t he s cene c am era
with the central field of view outlined in red, and the directi on
of g aze outlined i n green. ( Top Right) T he v isual field with n o
visual p rosthesis. (Bottom Left ) H e ad‐point ed c ase in w hich
the imag e from t he c entr al f ield of v iew a ppears in t he
direct ion o f g aze. ( Bot t om R ight) Eye‐pointed case i n w h ich
the r e gion o f the c a mera f ield o f v iew t hat is aligned w it h the
direct ion of gaze appea r s in th e dir ection o f ga ze. 12

x

Figure 7. Retin a l prosthesis w ith int r aoc u lar camera, showing the
support electronics enclosed in a hermeticall y sealed housing
and an u ltra‐flexible ribbo n cabled c onnected t o the
microstimulator array. 1 5
Figure 8. Retin a l pro sthesis w it h int r aocular camera; enla rged v iew.
The IOC is c omposed of a h ermet i cally s ealed housing, s ilica
window, custom polymer lens, a nd C MOS imag e senso r . 15
Figure 9. An o rig i nal image at a r esolution of 512 512 pixels ( Left),
down‐sampled and pixellated to a level of 16 16 pixels
(Cent e r) a n d t hen con v ert e d to a n array of c ir cular electr odes
with a 50% duty cycle (Right). 1 8
Figure 10. A gr idded v e rsion of t he o rig i n a l image from F igure 9, b u t w ith
addit i onal G aussian b lur to sim ulate el ectric f ield sprea ding
from the electrodes into the retinal tissue. Gaussian blurs
with sta nda r d deviat ions set t o 1 8 percent (L eft), 31 per c ent
(Center), and 62 percent (Right) are shown here. 20
Figure 11. First g e n e ratio n v is ual prosthesis s imulator h ard w are
overview showing two PCs, with one running custom VPS
software and the other runni ng eye tracking so ftware. 31
Figure 12. The first generation v isual prosthesis s imulator Z 800 h e ad‐
mounted display (HM D ) hardware s howing a n eye‐tracking
camera, a scene camera, and ca bling mounted to the HMD. 33
Figure 13. Sensics zSight h ead mounted display showing the scene
camera mounted to the front (Left) and the eye tracking
camera with nearly on‐axis v iew o f the pupil (Right). 34
Figure 14. Flow c hart o f the visu al p rosthesis simulator code s howing
the algorithm used t o form i mages for the AM D and RP m odes
as well as t he head‐pointed and ey e‐pointed c a mera mod es. 37
Figure 15. Images c ap tured from t he scene c amera before ( Left) and
after (Right) dewarping, w it h the dewarping parameter, f , set
to 1 .39. In t his case, th e effec t iv e focal length i s exp r ess e d as
a multiple of the sensor’s h alf‐diag onal distance. 3 9
Figure 16. A demonst r atio n of t he v isual p r osthesis s imulator i mage
processing o utput showing a resolution o f 6 10 (similar t o
the Ar gus I I dev ice) w ith va rious stages o f G a ussian b lur of
none ( First Row), 30% G aussian b lur (Second Row), and 60%
Gaussian blur (Third Row). The original image is shown for
comparison (Bottom). 41
xi

Figure 17. A demonst r atio n of t he v isual p r osthesis s imulator i mage
processing o utput showing a resolution o f 12 20 ( Left
Column) and a resolution of 24 40 (Right Column). Va rious
stages o f G a ussian b lur are sh own of n one ( F irst R ow), 30%
Gaussian b lur (Second Row), and 60% G aussian b lur (Third
Row). The original im age is shown for comparison (Bottom). 42
Figure 18. The curren t g enerat io n visual p rosthesis s i mulator with
exper i ment er m onitor s for th e graphical user i nt erfa ces,
head‐mounted display view, and s c ene camer a view. 4 4
Figure 19. Image capt ure of t he v isual prosthesis s imulator g raph ical
user int er fa ce. 45
Figure 20. Swim l ane diag ram s h owing the sequence o f ev en ts t hat
produces the lowest possible system latency. The external
even ts h appen just b efore (1) the cameras capture images,
and (2) the correct g az e coordinat e s ar e calculated a nd m ade
available ju st b efor e th ey a re n eed e d to e xt rac t t he r eg ion of
int e res t . 51
Figure 21. Excerpt of the Excel file used to calculate worst case and best
case latency scenarios. Each pair of letters represents an
output from a previous step that can be used as an input in a
subsequent processing s tep. 5 3
Figure 22. Plots of the range of sub‐normal minimum, normal minimum,
and maxim u m values i n dB f or f loatin g point numbers with
fractional significands as a function of the number of exponent
bits. 63
Figure 23. An image that has been blurred using the OMAP processor
and Cimg library by means of Gaussian convolution (Left),
anisot ropic filterin g (Center), and a moving average box f ilter
(Right). 6 5
Figure 24. Plots of e xecution tim e as a f un ctio n of k er nel or f ilt e r siz e for
three differ e nt b lurrin g a lgorit hms using the Cimg l ibrary
running on a LogicPD O M AP dev elopment boar d . 67
Figure 25. Subject performing t h e v isual search e xper iment, s howing
projected letters i n the background a nd t he v is ual prosthesis
simulation and MATLAB software running and displayed on a
set of computer monito rs in th e for e ground. 78

xii

Figure 26. Screen capture of the automated visual search data collection
MATLAB application showing the placements of letters on the
screen (Left) as well as the data collected for an experiment
(Right). The response times are in units of sec o nds. 7 9
Figure 27. Example of simulated AMD condition with an intraocular
retinal pros thesis d uring a visual s earch exp e riment. ( Top)
Image capt ured b y the scene camera showing t he e xt en t of t he
HMD field of view (Yellow Box) and where the subject is
currently looking (Red Box). ( Bottom) View displayed on the
HMD that the user sees, in this case simulating AMD with a
central scotoma scaled t o an a pproximatel y 28° h orizontal
field of v iew in o rder t o extend b eyond the extent o f the
central visual prosthesis simulation. Note that the image
portion within the red box and the displayed percept are
offset by a slight amount, likely due to the drift in calibrati on
over the length of the experiment as these images were
extr acted after exper i ment completion. 8 3
Figure 28. Plots of t ime to f ind the t a rg et i n a v i sual search with
simulated Retinitis Pigmento s a , comparing exper i ments
using a limited field of v iew b ut f ull resolution d isplay ( Top)
and a 24 40 pixellated display (Bottom). Curves shown are
for head‐point ed ( Red), eye‐point e d (Blue), full field of v iew
HMD (Magenta), and nor mal vision (Green). 8 5
Figure 29. Plots of t ime to f ind the t a rg et i n a v i sual search with
simulated Age‐Related Macula r Degeneration, comparing
exper i ment s using a limited fiel d of v iew but full resolution
display (Top) and a 24 40 pixellated display (Bottom).
Curves shown are for head‐pointed (Red), eye‐pointed (Blue),
full field of v iew HMD (Magen t a), a n d normal v ision (Gr e en). 8 6
Figure 30. Summary of slopes (Top) and intercepts (Bottom) of
regr ession l ines f it t o ea ch p lot of r esponse times as a f unct i on
of the number of distractors. 88
Figure 31. Plots of t ime to f ind the t a rg et i n a v i sual search with
simulated Age‐Related Macular Degeneration, in the target‐
absent case, and comparing experiments using a limited field
of view but full resolution display (Top) and a 24 40
pixellated display (Bottom). C urves shown are for head‐
pointed (Red), e y e ‐point ed ( Blue), f ull field of v iew H MD
(Magenta), and normal vision (Green). 90
xiii

Figure 32. Plots comparin g head‐ p ointed ( Left) a n d eye‐ pointed (Right)
modes for RP w hen p e rforming a target‐absent trial. Plots
from t op t o bottom are horizontal g aze coordinate, vertical
gaze coordinate, horizontal head motion (yaw), and vertical
head motion (pitch). 9 4
Figure 33. Subject performing the reach and grasp task using the visual
prosthesis simulator, with t he t a r get object v isible i n the
background b ehind the cen t er o ccluder o bject. The VPS
software can be seen in the foregro u nd. 96
Figure 34. Schematic diagram of the reach and grasp experimental setup.
The subject is s ea ted as c lose t o the table as p ossible, a nd t h e
visual p rosthesis simulator scen e c a mera e xtends slightly i nto
the reach and grasp area. T h e t arg e t o b ject i s placed
randomly in one of the four locations. 9 7
Figure 35. Screen capture of the visual p rosthesis simu lator, s howing
one reach a n d grasp tr ial in t he e ye‐pointed, AMD, 24 40
resolution, and blurred configura t ion. T he e xp erimenter v iew
is shown (Top) as well as the subject’s view in the HMD
(Bottom). 98
Figure 36. Bar chart c o mparing t h e time t o grasp a t a r g et o bject for the
pixellated mode ( red) t o the pixellated and blurred mode
(blue) for each of the 24 40 resolution experimental
parameter combinatio ns. 101
Figure 37. Bar chart c o mparing t h e time t o grasp a t a r g et o bject for the
pixellated mode ( red) t o the pixellated and blurred mode
(blue) for each of the 12 20 resolution experimental
parameter combinatio ns. 101
Figure 38. Bar chart c o mparing t h e time t o grasp for t h e eye‐pointed
mode (blue) to the head‐pointed mode (red) for each of the
24 40 res o lution exp erimental p a rameter c o mbinations. 103
Figure 39. Bar chart c o mparing t h e time t o grasp for t h e eye‐pointed
mode (blue) to the head‐pointed mode (red) for each of the
12 20 res o lution exp erimental p a rameter c o mbinations. 103
Figure 40. Bar chart co mparing the number o f obstacle h it s for th e head‐
pointed mo de ( red) t o the ey e‐pointed mode ( blue) for each
of t he 24 40 resolution e xperimental parameter
combinatio ns. 105
Figure 41. MATLAB GUI for course generation and data collection for the
navigat i on a nd mobilit y exper i ment. 109
xiv

Figure 42. Histogram plot o f the difficulty s cores for 2,000 rando m ly
generated c o urses. 110
Figure 43. Two experimental courses (“Course A” and “Course B”) and
one tr ain i n g c ourse (“Tra i ning”) desig ned for the na vig a tio n
and mobility e xp eriment. F iducial marks and color coded
obstacles aid in the e ase of course assembly. 111
Figure 44. Three exa m ples o f front faca des of o bstacles u sed in t he
navigation and mobility experiment. From left to right: a
bookshelf, a filin g cabin e t, an d a multi‐level set of drawers. 112
Figure 45. Photograph o f a subject per f or ming t he n avigat ion and
mobility e x p erimen t. The bookshelf obstacle c an b e seen i n
the foregro u nd t o the left, and the pillar obstacle c an b e seen
in the back g round occluding the t a rget locat io n. 113
Figure 46. Bar chart of t hr ee c ases o f preferr e d walkin g speed
measurements i n the case o f a no rmal subject w alking i n a
straight l ine along th e obstacle c ourse with n o obstacles
(Left), a normal subject n av igat ing the tr aining c ours e
(Cent e r), a n d a subject wear ing the VPS HMD with f ull
resolution ( Right). 116
Figure 47. Bar chart showing the percenta ge i mprovement i n effec t ive
preferred walking sp eed w hen in t he e ye‐p ointed m ode as
compared w ith the head‐pointed m ode, a nd c omparing t he
full resolution ( but limited fi eld of view) mode to the 24 40
resolution mode. 117
Figure 48. Example layout of an E PROM device with a fl oating gate. The
red floating gate is initially charged to a negative potential by
tunnel injec t ion of e lect rons, and is t hen gr adua lly n eutraliz e d
in the pr e sence of posit ive mobile ions. 128
Figure 49. The smallest k nown q uartz crystal oscillator package (Left) i s
the Nihon NX1612AA, a nd i s shown compared t o a millimeter
ruler. T he c rystal o scil lator size o f 1.6 1.2 0.3 mm is s t ill
too large for many m icro‐implanta b le device a pplications. A
MEMS oscillator (Right) can be made small, but is difficult to
fabrica t e a n d usually o perates best i n the kilohertz r a nge;
from [11]. 130
Figure 50. The Caden c e layout ( Top) a nd a n optic a l microgr a ph
(Bottom) of Hermetic Coating Tes t Chip 1 (HCTC‐1). 132
Figure 51. An illustration of the generalized design of the PMOS EPROM
to be used a s a positiv e m obil e ion contamina t ion sensor. 134
xv

Figure 52. Photograph o f the HCTC‐1 c hip pa ckaged i n a ceramic 40‐ pin
dual‐inline package (DIP) with t he c over r emoved. W ire
bonds (thin gold w ir es ) can be s een connect i ng t he b onding
pads on the perimet er o f the ch ip t o the packa g ing pins. 136
Figure 53. Electrical s chematic o f the simulated a n d fabricat ed 25 MHz
CMOS c lock g enerato r w ith a 5‐ stage ring o scillator and a
divid e ‐by‐two output s tage c on sisting of a D ‐flip‐flop w ith
buffering inverters on either side of the D‐flip‐flop. 137
Figure 54. Electrical s chematic o f the simulated a n d fabricat ed 25 MHz
CMOS c lock g ener ator w ith a 7‐stage ring o scil lator and n o D ‐
flip‐flop output stage. Buffering inverters are included at the
output to supply sufficient curre nt to a capacitive load. 138
Figure 55. Schematic diagram of the delay cell used in the ring‐oscillator ‐
based clock design o n the CMOS c lock t est chip ( CCTC),
showing an inverter with added load capacitance to decrease
the fr equen c y of oscillation. 139
Figure 56. Simulated output w aveform of t he C MOS ring o scillator with
a 5 V power supply and a 0.5 pF load. 140
Figure 57. Cadence lay o ut o f the CMOS c lock g enerat ion circuit showing
a 5‐stage ring oscillato r (Left) coupled with a standard cell D‐
flip‐flop out p ut stage ( Right). 141
Figure 58. Screen c ap ture o f the LabV iew G UI a pplication for the
measurement of clock jitter. The waveform is shown (Top),
in a ddition t o period d eviation measurements over time with
respect to t he a v e rag e p eriod (Bottom Left) and a histogram
of all perio d deviatio n m easurements (Bottom Right). 144
Figure 59. The two electrical m ea surement s etups used f or t he p os itive
mobile i on c ontaminatio n e x p erimen ts, including the
charging a n d m easurin g o f the EP ROM (Left) a s well as t he
extr actio n o f the PFET c alibration curve (Right). In both
cases, the drain is at the top and the source is at the botto m of
the trans i stor. The bulk substr ate and N‐well are both tied to
ground. 148
Figure 60. Plot of the measured drain current in milliamps as a function
of t he a pplied gat e ‐source voltag e for a PFET o n HCTC‐1, with
V DS = –1 V. This plot is measured from the PFET device with a
non‐floating gate, and can be used to estimate the floating gat e
voltage of a s imilarly constructed E PROM. 149
xvi

Figure 61. Photograph o f the saline s oak exp e riment u nderway with the
firs t hermet ic coatin g test chip (HCTC‐1). 151
Figure 62. Cross sectio n of t he O N Semiconductor C5 C M O S process layer
thickness estimation s tructure with relative thicknesses
measured using a Dektak profilometer. 154
Figure 63. The Caden c e layout ( Top) a nd a n optic a l microgr a ph
(Bottom) of Hermetic Coating Tes t Chip 2 (HCTC‐2). 157
Figure 64. Optical mic r ograph o f two as‐f abricated EPR O M PMIC s ensors
with corresponding calibration PFETs on the second
generat i on h ermetic co atin g t e st c hip (HCTC‐ 2), showing one
with a supervening passivation oxide cut region (Left), and
one with a fully intact p a ssiva tion o xide r eg ion (Right). 159
Figure 65. The Cad e nc e layout ( Left) and a n o ptical m icrograph (Right)
of the CMO S Clock Test Chip (CCT C). 163
Figure 66. Schematic diagram of t he r ing‐ oscillator‐based clock design
containing a 7 ‐sta ge r ing oscilla to r, w ith inverters and d e lay
cells designed to slow down the frequency of oscillation. The
25 MHz output i s passed through a two‐stage D‐flip‐flop to
halve the fr equency tw ice, a nd o ut put taps a re p rovided for
each frequency div isio n step. 165
Figure 67. Allowable frequency v a ria t ions f o r b oth square w a v e input
signals (Blue Bars) and sine w ave input signals (Red B ars)
centered a t a nominal input freq uency of 6 . 14 MHz, a s a
function o f the peak‐to ‐ peak e xter nal clock signal a mplitude
(V pp ) for the OmniVision OV69 20 CMOS imag e senso r. 167
Figure 68. Screen c ap ture o f the LabV iew G UI a pplication for the
measurement of c lock j itter for the CMOS C lock Tes t Chip. T he
waveform i s shown (Top) in a d ditio n to p eriod devia t ion
measurements over time with respect to the average period
(Bottom L e ft) and a histogra m of a ll period d eviation
measurements (Bottom Right). 169
Figure 69. (Left) E xperimental setup co nt aining a M OSIS‐fabricated
CMOS digital clock based on a multi‐stage ring oscillator
design. The IOC lens and image sensor are held in a mount to
the right of the power supply on the optical table. (Right)
Acquired image from an OmniVision OV6920 image sensor
array coupled with an IOC custom polymer lens optical
system, with the OV6920 image sensor array driven by the
CMOS dig it al clock. 170
xvii

Figure 70. The Caden c e layout ( Top) a nd a n optic a l microgr a ph
(Bottom) of Hermetic Coating Tes t Chip 3 (HCTC‐3). 172
Figure 71. Screen c ap ture o f the LabV iew G UI a pplication for the
measurement of c lock j itter for the t h ird hermet ic c oatin g t est
chip (HCTC‐3). The waveform is shown (Top) in addition to
period d eviation m eas u rements o v er t ime w i th r espect t o the
aver age period ( Bottom Left) and a histogram of a ll p eriod
deviatio n measurements (Bottom Right). 174
Figure 72. Cross sec t io n illustrat i o n o f the nin e P MI C sens or e xper im ents
on the third hermetic coating test chip (HCTC‐3). The top and
middle groups have passivation oxide cuts, while the bottom
group has t h e passiv a t i on l ay er i n t act. The n u mbers above
each s ensor show t he e stimated t hickness o f o xide c overing
the sensor floating gate. The floating gate metallization laye rs
are Metal 1 (dark blue), Metal 2 (maroon), and Metal 3 (light
blue). V ias wer e u sed to i ntercon n ec t the v a rious
metallizat i o n levels for a giv e n PMI C senso r flo a tin g gat e. 178
Figure 73. Photomicrograph of t he n ine P M IC sensor s o n the third
hermetic c o a tin g t est c h ip ( HCTC‐ 3 ) that w er e incorpora t ed
to m easur e t he p ositiv e mo bile i on c ontamina t i on
sensitiv it ies o f EPROM PMIC s ens o rs w ith v a rious float i n g
gate depths (i.e. metallization layers) and sizes. PMIC sensors
1 through 9 from Table 18 are arranged left‐to‐right and top‐
to‐bottom. The larg est floating gate shown in t he middle row
a n d r i gh t m ost column i s 1 2 by 3 6 mi cr on s. T he t op t wo r ows
have a passivation oxide cut and the bottom row has the
passivation o xide i ntact. 179
Figure 74. Scanning electron microscope (SEM) micrograph of an
exposed HCTC‐3 M et al 3 f loating g a te. T his “ m edium” s ized
PMIC senso r is l ocated i n the cent ral exper i ment al r egio n of
Figure 73, just to the left of its companion PFET calibration
device. 180
Figure 75. Photograph o f two custom h igh temperature PCBs w ith HCTC‐
3 chips mounted in h igh tempera t ure zero‐inser tion‐fo r ce
sockets in t he Binder F D ‐53 precis ion oven. 182
Figure 76. Photograph of an HCTC‐3 chip in a high temperature zero‐
insertio n‐fo rce socket. 182
Figure 77. Screen capture of the PMIC testing data control and
acquisition LabView pr o gram. 183
xviii

Figure 78. Photomicrograph of a n HCTC‐3 c hip aft e r cont aminat ion with
a saline dro p that was s ubsequently dried. 184
Figure 79. Plot of the normalized drain current as a function of time for
three contaminated PMIC sensors on HCTC‐3, all with an
oxide passivation cut. Cases including that of a Metal 2
floating g ate (exposed t o air), a Metal 1 floating g ate, a nd a
poly floating gate are shown. 185
Figure 80. Plot of the number of accumulated charges as a function of
time for the case of contaminated PMIC sensors with a Metal
1 floating gate (Blue) and a poly floating gate (Red), under
conditio ns o f accelerated lifetime t esting at 180 °C. 187
Figure 81. Plot of the number of accumulated charges as a function of
time for the case of contaminated (Dashed) and
uncontaminated (Solid) PMIC sensors with a Metal 1 floating
gate (Blue) and a poly floating g a te (Red). 188
Figure 82. Plot of the normalized drain current as a function of time for
contamina t ed ( Dashed) and uncontamin at ed ( Solid) P MIC
devices on HCTC‐3, comparing the decay of the drain current
for a number of floa tin g gate con f i g urations. 189
Figure 83. Photomicrograph of t h e P MIC lateral diffusion e xp erim ent
test area, sh owing one PMIC sensor at the corner and another
at the center of the test region. Both PMIC sensors are
fabrica t ed w ith Met a l 3 gat e s. 191
Figure 84. (Top) SEM micrograph of one area of the outer rim o f the t est
area, showing a metal trace passing between the Metal 2 layer
and Met a l 3 layer ox ide int e rface. ( Bottom) A n illustration t o
help visualize this geometry. 192
Figure 85. Plot of normalized drain current as a function of time for two
sets of PMIC sensors at the corner (Red) and at the center
(Blue) o f the lateral diffusi on t est region o f HCTC‐3.
Uncont aminat ed c urv e s a r e sho w n as solid l ines, whil e the
contaminated curves are shown as dashed lines. 193
Figure 86. The Caden c e layout ( Top) a nd a n optic a l microgr a ph
(Bottom) of Hermetic Coating Tes t Chip 4 (HCTC‐4). 196
Figure 87. Photomicrograph showing th e fabrica t ed i nt erdigitated
electrodes w ith 20 µm f ing e r lengt h s. 198
Figure 88. The Caden c e layout ( Top) a nd a n optic a l microgr a ph
(Bottom) of Hermetic Coating Tes t Chip 5 (HCTC‐5). 207
xix

Figure 89. Photomicrograph showing a close‐up v iew o f the PMIC
sensors with v arious g ate sizes (Top), t he w edge‐shaped
oxide cut (Cen ter), an d the set of i nt erdigit a ted elect r odes
(Right) included in the HCTC‐5 des ign. 210
Figure 90. Plot o f the aver age normalized dr a in c urr e nt a s a funct i o n o f
time f or f our different HCTC‐5 c hips, with e rror b a rs
indica ting ± 1 stand a rd deviatio n at selected t imes. 211
Figure 91. HCTC‐5 chip prepared for coating with an aluminum foil wrap
(Top) and a macro photograph o f the chip a fter c oating
(Bottom). 213
Figure 92. Plot of the accumulated number of positive ions as a function
of t im e fo r both u ncontamina ted a n d cont aminat ed u nco a ted
chips, a nd a c ontamin a ted chip w ith a multilayer d iel ectric
diamond‐like coating (DLC). 214
Figure 93. The Caden c e layout ( Top) a nd a n optic a l microgr a ph
(Bottom) of Hermetic Coating Tes t Chip 6 (HCTC‐6). 216
Figure 94. Plot of the normalized drain current as a function of time for
one uncont aminat ed a nd o n e c on tamina ted chip, each w ith
four PMIC sensors tested simultaneously. 219
Figure 95. Plot of the normalized drain current as a function of time for
an uncontaminated HCTC‐6 chip with four PMIC sensors of
differ ent g a te an d chan n el ar e as. 220
Figure 96. Plot of the normalized drain current as a function of time for
a contaminated H CTC‐6 chip w ith four P MIC sensor s of
differ ent g a te an d chan n el ar e as. 224
Figure 97. The Caden c e layout ( Top) a nd a n optic a l microgr a ph
(Bottom) of Hermetic Coating Tes t Chip 7 (HCTC‐7). 226
Figure 98. Plot of the normalized drain current as a function of time for
a contamin ated H CTC‐ 7 chip c ontain ing PMI C sen sors w ith
varying am ounts of protection ag ains t ion co ntaminat ion. 228
Figure 99. Plot of the normalized drain current as a function of time for
both c ontaminat e d and uncontamina t ed H CTC‐7 chips
containing P MIC s e nso r s with v ar ying a mounts o f prot ec tion
against io n contamina t ion. 229

xx

Figure 100. PFET I‐V calibration curves for a number of uncontaminated
HCTC‐6 a nd H CTC‐7 chips, s everal o f which ha ve u ndergo ne a
pre‐cleanin g p rocess before a h er metic coating was app l ied.
In sev eral c ases, the p a ssivat i on o x i de c ut a pp aren tly cau ses a
signific ant alter i ng of t h e PFET device charact e ris t ics. 233
Figure 101. The fourth g ener at ion int r aocular camera r ay diag r am,
from [3]. 236
Figure 102. The fourth g enerat ion intr aocular camera c oupled w ith the
OV692 0 image senso r a rray, a nd t he e quivalent biological
cornea model composed of two commercially available
lenses. 238
Figure 103. MTF plots of the original intraocular camera design with the
Liou and Brennan eye model and the OmniVision OV6920
image s e ns or ( Top) c ompared wit h t he m odified d esig n u sing
the biological c ornea model (Bottom). O nly minor var i at ions
can be observed when compari ng the two sets of plots. 239
Figure 104. Compariso n o f the fourth g eneratio n int r aocular camera
imaged i n air w i th out the cornea r epl a cement m odel
(Top Row), and again imaged in air with the cornea
replacemen t model (Bottom). C OD E V s i mulated ima g es a re
shown (Left Column), a s well as e xperimentally c aptured
images (Rig h t Column). 241
Figure 105. One pot e nt ial next‐genera tion i ntraocular c amera imaging
system l ens design t o b e u sed with the OV6920 image sensor
array. The lens ray diagram (Top), a simulated image
(Bottom, L e f t), and a sp ot d iagr am ( Bottom, R i g ht) are shown.
The field of v iew i n the simulate d image is ± 2 0 °, w ith on a xis
spot sizes of approximately 14 µm. This design retains the
window placed in front of the lens to maintain an air gap in
front o f the l ens. 243
Figure 106. A potent ial next‐g e ner a tion b ioim plantable imaging sys t em
lens d es ign based on t h e O V6920 i mage s enso r. T he l ens ray
diagram (Top), a simulated image (Bottom, Left), and a spot
diagr a m (Bottom, R ight) are shown. The fiel d of v iew i n the
simulated image is ±20°, with o n axis spot sizes of
approximately 2 µm. T his design h as n o window i n front of
the lens, but the ind e x differ ence i s large eno u gh t o allow for
good imaging. 244

xxi

Figure 107. Example schematic diagr a m of a l ow b attery v oltag e
detection circuit including a c o mparator t hat can be
repurposed a s a low gate v oltage d etect i on c ir cuit i ntegra ted
with PMI C s e nsors. Fro m Inters il ISL21440 data sheet. 254
Figure 108. Photograph of an HCTC‐4 chip w it h a P e ltier c ooler m ounted
to the underside of the ceramic package, including an
aluminum heat sink a nd thermis tor for feedb a ck control. 255

xxii

ABBREVIATIONS
AMD Age‐Relat e d Macular Degen e ra tion
RP R etin it is Pigmentosa
IRP Intr aocular Retinal Pro sthesis
IOC Intraocular Camera
EOC Extraocular Camera
VPS Visual Prosthesis Simulator
HMD Head‐Mounted Display
CPU Cen t ral P r o c essing U n i t
DSP Digital Signal Processor
VPU Visual Proc e ssing Unit
LGN Lateral Geniculate Nucleus
VOR Vestibulo‐O c ular Reflex
CMOS C omplementar y Met al‐Oxide‐Sem iconductor
VLSI V ery La rge Scale Integr ated
OLED O rganic Light Emitting Diode
PMIC P osit ive Mo bile Ion Contamination
HCTC H ermetic Coatin g Tes t Chip
CCT C CMOS Clock Test Chip
MTF Modulation Transfer Function
PMMA P oly(Methyl Methacrylate)
xxiii

IOL Intraocular Lens
DOE Diffractive Optical Element
SVDOE Strat i fied V olume Diffractiv e Optic al Element
WDR Wide Dynamic Range
FOV Field of View
FPS Frames per S econd
EPROM Electrically Programmable Read‐Only Memory
EEPROM Electrically Erasable Pr ogrammable Read‐Onl y Memory
BMES ERC Biomimetic M icroElectr on ic S ystem s Engineer i ng R esea rc h Center
PMOS P ‐Type Met a l‐Oxide‐Semiconductor
NMOS N ‐Type Met a l‐Oxide‐Semiconductor
FET Field‐Effect Trans istor
MEMS M icroelectr omechanica l Systems
GPU Graphics P rocessing U nit
DLL D ynamic‐Linked Library
GUI Graphical User Interfac e
CUDA C ompute Unified Devic e Archit ect u re
IRB Institution a l Review Bo a rd
MCM Multi‐Chip Module; Multichip Module
FPGA F ield‐Progr a mmable Gate A rra y
ARM Advanc ed Reduced Inst r uction Set C omputer Machine
RGC Retinal Ganglion Cells
xxiv

ABSTRACT
Blindness due to A ge‐Related M ac ular D egen er atio n and Ret i nitis
Pigmen tosa i s unfortun ately both w idespr ead and larg ely incurab le. A dvanc e s in
visual prostheses that can restore functional vision in those a fflict ed b y thes e
diseases h ave evolved rapidly from new areas of researc h in oph thalmology a nd
bi ome di cal e n gin ee ri n g . T hi s the s is i s f ocuse d on f urther adva ncing the st ate‐of‐
the‐art of b oth visual prosthese s, and implantable biomedical d ev ices.
A nov e l real‐time syst em w ith a high p erform ance h ead‐mounted d isplay
is described t hat ena b les enhanced r ealist ic simulation of i ntr ao cular retinal
prostheses. A s et o f v i sual p sychophysics e x p erimen ts i s prese nted u sing t h e
visual p rosthesis simulator that q uantify, i n several ways, the b enefit o f fovea t ion
afford ed b y an e ye‐pointed c amer a (such as a n eye‐tr ack e d ext r a ocular c amera
or an implantable intraocular camera) as compared with a head‐p oin t ed c amer a.
A visual s ea rch exper i ment d emonstra tes a s i gnifica n t improveme nt i n the time
to l ocate a targ et o n a screen w hen using an e ye‐point ed c amera. A reach and
grasp exp e r i ment d emonstra tes a 20% t o 70 % improvement in tim e to grasp an
object w hen using a n e ye‐point ed c amera, w ith the improvemen t m aximized
when the percept is blurred. A navigation and mobility experim ent shows a 10%
faster walking speed and a 50% better ability to avoid obstacle s when u sing a n
eye‐point e d camera.
xxv

Improvements t o implantable biom edical d evices a re a lso describ ed,
including the design and testing of VLSI‐integrable positive mo bile i on
contamination senso r s and humidity s ensors t hat can validate t h e hermeticity of
biomedical device p a ckages e ncaps u lated by h ermet i c coatings, and can provide
early war n ing of l eaks o r contaminatio n that m ay j eopardize the implant. The
positiv e m obile ion contamina t io n sensors are shown to b e sensi tiv e t o externally
applied contamination. A model i s proposed to describe sensiti v i ty as a function
of dev ice geometry, an d ver i fied experimentally. Guidelines ar e pro v ided o n the
use of spare CMOS oxide and metal layers to maximize the hermet icity of a n
implantable microchip. I n ad dit i o n , results a r e pres en ted on t he d esig n and
testing of small form factor, very low power, integrated CMOS c lock g enerat ion
circuits that are stable enough to drive commercial image sensor array s, and
therefore c a n be incor porated in an intraocu lar camera for reti nal p r ostheses.
1

CHAPTER 1
INTRODUCTION
1.1 Background and Problem Statement
Age‐Related Macular Degeneration (AMD) is the leading cause of late
onset blindness in people over the age of fifty, and remains in curable [1, 2].
Retin i t i s Pigmentosa ( RP) is a not h er i rr ever sible ret i na l disea se t hat causes
blindness worldwide [3, 4]. B oth of t hese diseases dam a ge p hot osensitiv e c ells
(rods and c o nes) i n th e retina a n d m iraculously leave other ele ctrically a ctive
retinal ganglion c ells ( RGCs) la rgely in tact, thereby a llowing for electron ic
intr aocular ret i nal pro s theses t o emerge a s a viable t r e atment option [5–10].
Modeling, unders tand ing, a nd i mproving t hes e i ntr a ocular r et ina l prostheses i s
the crux of t h is thesis w o rk.
AMD usuall y begins w ith a lessen ed v isual acuity in the macular reg io n of
vision ( approximat ely 10° f ield of v iew) that i s often accompan ied b y distortio n
and may pr ogress out w ard, a ltho ugh peripheral v is ion is t ypically not affected.
This d iseas e a ffects o v e r 20 million p eople worldwide, w ith projections of up to
75 million b y the year 2030 [1, 11]. T here a re c urrently n o known preventativ e
measures f or A MD, an d trea tmen t options ar e gen e rally either ri sky or l argely
ineffectiv e. On e t r ea tment consist s o f intr avitr e al i njectio n s of Avastin, in which
some visual acuity improvement can be seen [36]. RP, on the ot her hand, tends
to b egin w ith a loss of p eripher a l visio n t hat progresses unt i l the point of
2

complete blindness, and affects nearly 2 million people worldwi de [3]. T o date
there a r e n o p reven t at ive measur es, but because RP i s largely a g enetic d isorder,
there are p r omising results from g ene therap y. H owever , at t he time of writing,
there ar e o v er 45 iden tified g en es w ith hundr e ds of m utations t hat m a y give r is e
to RP, so that a complete geneti c tr eatment solution may be decades away [3, 4].

Figure 1. Schematic diagram of the human eye and retina showing key anatomical structure s
a n d cellula r layers of the retina ; f r om [10], a ft er [12].
Because both A MD a nd R P are conditio ns t hat largely affect o nly the rods
and cones of t he r etina (shown i n Figure 1 ), v arious a posteriori treatment
approaches h ave been e xplored in r ec ent years tha t i nclude r eti nal cell
transpla nta t ion [1 3], s t em c ell therapy [1 4], and prosth etic v i sion devices t ha t
can el ectric ally stimulate t he o ptic n erv e [1 5 ], visual c ortex [16– 18], o r r emaining
viable retinal ganglion cells [5–9]. Each of these treatments comes with i ts o w n
set of b enefits and do wnsides. B oth retinal cell transplantati on a nd s tem cell
3

therapy have not yet been largely successful, although recent r esults i n
transpla nta t ion o f p h o toreceptor p recursors have b een e ncouragi ng [ 13].
However, m uch like g ene thera p y for prev ent i on, these treatment s are most
likely many years away from being acceptable a nd rel iable treat ment o ptions.
In t he n earer term, visu al p rost heses have b een developed and t est e d that
work b y stimulating viable e lectrically a ctive cells i n the ret ina, o ptic n erve, or
brain with a n array of m icroelectrodes, w h ich in t urn creat e s p hosphenes
(perception s o f light elicited b y dir e ct s imulation of t he v isual system) [5, 6, 19,
20]. Th e se system s ca n b e u sed t o conver t a n im a g e cap tu red b y a v ideo camer a
into a n ima g e perceived by t he p at ien t with t h e v isual pro sthes is v ia p hosphenes.
The major disadva n t a ge o f optic nerve prosthes es i s the fact t h at t he o rganiz atio n
of nerve fibers leading to the visual cortex is not well known, and as such
contacting the full set of nerve fibers in order to provide a ret inotopic p ercept i s
difficult. A nother disa dvant age is t he r ela t iv e in accessibility of the optic nerve
itself. D ir ect visual c o r tex stimulation has a l so b een sh o wn to be effective in
producing phosphenes, although o ften t he pr o spect of brain surg ery is daunt ing ,
and the implantat i on o f a dev i ce i nto the br ain is b oth unusual ly d ifficult a nd
invas i v e .
Other fo rms of v isual prostheses i nclude dev ices t hat leverag e the cross‐
modal plasticity of the human brain, whereby one sense can be e ffect iv ely
“mapped” o nto a d i ffer ent s e ns e. For example, e ffort s ar e unde rway t o re‐map
the brain t o i nterpret t ouch o r sound into v isual stim uli, w hic h can then b e
4

interpreted as images by the patient [21, 22]. While there are i ndic ations t ha t
this type of non‐surgical approach is possible, the extent to w hich v is ion can be
restored and the effect on the e xisting sense th at is being lever aged h as yet to be
fully deter mined.
1.2 Intraocular Retinal Prostheses
Of the various treatment options for RP and AMD presented in th e
previous s ection, int r aocular ret i nal prosth eses h av e been a n active area o f
research due to their relative ease of implantation, as well as t he s uccess of
h u m a n stu dies u si ng th i s ap p r oa ch . Typ ical l y , intr aocu l a r reti nal pros theses are
composed o f a small vid e o camera p laced eith e r outsid e t h e eye and mounted o n
the head (extraocular) or implanted within the eye (intraocular) to act as the
image capt ure componen t of t he system t o r eplace t he damaged n a tural
photoreceptors o f the eye, a s shown in F igure 2 [9]. T he i mage s or v ideo stream s
captured by the camera are then routed to a visual processing unit (VPU) t o
convert the pixel values t o electrical s timulation. T he s ignal s that are generated
by t he V PU are spec ific ( in b ot h amplitude and wav e form) to t he type of
microstimulator, t he t y p es o f cells t hat a r e being st imulated, an d the proximity of
the elect r o d es t o the c e lls. L astly, t he e lec t ric a l stimulatio n signals are routed t o
a microstimulator array, which consists of a set of microelectr odes t hat deliver
the stimula t ion to t he R GCs by w ay o f ind u ced electr ic c urren t s [8, 9]. This
microstimulator chip i s often pla c ed i n a hermetically s ealed p ac kage t hat is
5

mounted to t he e xterio r of t he s clera and attac h ed t o the eye globe with a cuff. In
the case of the extraoc u lar camera, wireless induction is u sed to t ransmit the
signals and power from the VPU t o the microstimulator array, m u ch l ike a
cochlear implant.
The placement of the microstimulator array in proximity to the retina i s
of particular importance i n order to f orm precise phosphenes f o r the pa ti en t;
however, various placements also have surgical tradeoffs. The microstimulator
array can generally be implanted in one of three locations: (1) directly under the
retinal surface closest to t he phot oreceptor c ell layer (subretinal), (2) on top of
the reti nal surface closest to t he RGC l ayer ( epireti n al), or (3) within or just
outside of the sclera (suprachoroidal or extraocular). The epi retinal and
subretinal approaches are i llustrated in Figure 2.
Fi gure 2. Comp onent s o f a re ti na l prost h esis i ncludi ng a vi d eo camera and a microstimulator
a rra y; f ro m [9].
6

The subretinal p rosthesis placem ent is t yp ical ly a ccomplished s urgically
by intentionally detaching a portion of the retina, lifting the r etinal r egion, a nd
inserting a thin m icro electrode arr a y betw een the r e tina a nd t he choroid. I n the
epiretinal a pproach, a microelectr o de a r r ay i s ins e rt ed t hrough a small incision
in the sclera and positioned on top of the inner surface of the retina. The array
can be held in place with a barbed tack or a biocompatible adhe sive. A
suprachoroidal array can similarly be tacked or glued onto the e xte ri or scle r al
surface, or implanted into a f lap just under th e sclera.
The most successful p rosthesis to date is b ased o n the epiretin al
prosthesis invented by Dr. Mark S. Humayun, et al . and commercialized as t he
Argus II retinal prosthesis (Second Sight Medical Products, Syl mar, C alifo r nia)
[23–26]. The current generation of this device allows for a to tal of 6 0 phosphenes
to b e created using a m i crostimulator ar ray th at i s tacked i nto p lace directly on
the retinal surface, a s shown in F igure 3. A dvantages of t his approach o ver the
others i nclude t he relat ive eas e o f s u rgery, g oo d proximit y cou pling to the RGCs,
and eas e o f removal for device u pg rade o r r e placement. H owever, di sa dva n t a ge s
include the possibility o f retinal detachment a t the tack l ocati on a n d di f f i cult ie s
routing control and po wer connec tions w i thin the ey e .
7

Figure 3. (Left) Illustration of an eye with implanted epiretin al m i c ros t i m ula t or a rr ay; from
[10]. N ot e t h at t he d iag r am o f t he eye has been rota t ed c ount e rcl o ck wise b y 90°, a nd t he ul t ra ‐
flexible ribbon ca ble rout ed i n t he opposit e hemisphe r e rela t i v e to i ts u sual p l a c e me nt, f o r
clarity of illustration. (Right) Fundus photograph of an impla nt e d 60 elect rod e a rray a s p art
of t he Arg us II d evice; from [9].
Even with only 60 discrete stimulation sites, subjects are able to perform
a variety of tasks significantly better with the retinal prosthesis on, including
locating objects, discrimination of direction of motion, discri mination of visual
grat ing o r i e ntat ion, a n d f i n e contr ol of hand movements [23, 2 5 ]. Research i s
currently u nderw a y to i ncrease th e number of electrodes f rom 60 t o 1,000, w hich
could enable t he capability of more advanc ed v isual functions s uch as f acial
recognition and read ing. A s such, one recurring f ocu s i n this t hesis i s th e
simulation a nd u nders t and i ng of r e ti nal prost hesis microstimula tor array s w ith
250 to 1 , 0 00 e lements, e ven befor e t he a c t ual implementations o f such d evices
are r e aliz ed .
8

1.3 Effects of Head and Eye Pointed Cameras
The Argus II devic e p r e sen t ed i n the prev ious section, a s w ell as ot h er
visual prostheses, typically make use of an externally mounted extr aocular
camera sometimes mounted on sunglasses as shown in Figure 4. T he a dvantages
of such an extraocular camera include easy hardware replacement or upgrades,
and easy w ired a cc ess to t he V PU. A lthough this s ystem has bee n used i n p a tient
trials t o great success, o ne d isadvantage of s uch a system i s t hat b o th f oveation
ability a n d t h e natural c o upling between head and ey e motions a re lo st [23, 25].

Fi gure 4 . Intra ocul a r re ti nal pr os the s i s w i t h e x te rn ally m oun t ed (extr a o c ular ) camer a . The
diagram o f t h e eye has been ro t at ed by 9 0 ° for cl arity of illus t r a t i o n; from [10].
In order to use such systems effectively, patients must learn t o scan t he
scene with l arge h ead movements while att e mpting t o keep t heir eyes f ixed
forward as b est as p ossible. T his task c a n b e quite d i ffic ult due to the human
visual s yst e m’s vestib ulo‐ocular r eflex (VOR) that n at urally m a kes one’s ey e
9

move opposite the direction of any fast head motion in order to keep the eye’s
gaze d irec tion f ixed i n s p ace. I n o n e study, i t w a s shown th at the VOR of subjects
implanted with the Argus II device was still intact after four years of i mplantatio n,
and that t h e discordan ce b etween t he h ead directio n and eye dir ection c aused a
distorted p e rception o f localizatio n o f objects [37]. I n this thes is, the “head‐
pointed” c a m era mode m eans t hat the camera i tself is f irmly aff ixed t o the head,
so any head motion will change the direction that the camera is pointed.
However, the intraocular retinal prosthesis microstimulator arr ay i s affix e d t o
the retina, and therefore moves with the eye. The effect on th e percept i on o f
where objects in t he s c e ne a r e l oca t ed i s illustr a ted in F igure 5 for the case of RP.
The entire 56° 74° f ield o f view ( FOV) o f the scene camera i s shown at the top
left, with the head direction indicated by the solid red box an d t h e ey e gaz e
direction indicated by the solid green box.
The VPU extracts the center 12° 20° region of the scene camer a image
and stimulates t he r et ina in a r eg ion of t he m acula, r esulting in t h e simulated
image show n in F igure 5 Bottom Left. The field of v iew o f t h e s i mulated r e gio n
in t his exa m ple has been c hosen to b e large enough t o comfortably cover the
macular region o f vision approx imately 13° , a nd t he v ert i cal fiel d of v iew i s
adjusted to mainta in t he aspect ratio of the ele ctrode array t ypically 2:1. If the
head p osition were h eld constant i n this c ase, a ny m ovement of the eye would
change the apparent direction location in space of the percei v e d image,
although the image contents woul d not change a s the microstimul ator a rray is
10

affixed to a given location on the retina. Imagine now that th e direct ion that t he
camera is pointed could be controlled by the eye’s gaze directi o n , hereinafter
ca lle d t h e “e ye ‐poi n t e d ” ca me ra mode . T he e ff e ct of such a n ey e‐poin ted camera
system i s s h own in F ig ure 5 (Bott o m Right). I n this c ase, b eca use th e region o f
the scene camera image extracted by the VPU is always aligned w ith the dir e ction
of g az e, w h e n th e micr ostimulator arr a y is a ct ivat ed t h e p ercep t appears wher e
it should be in space.

Fi gure 5 . Si mulati on o f the e ffe c t s of f ove a ti on o n a sub j e c t w i t h R e ti ni ti s Pi gme n to sa t ha t h a s
been i mpla nted w it h a re t i na l prost hesis. ( Top L e ft ) Imag e f ro m t h e scene came ra w i t h t h e
centr a l field o f v iew outline d i n red, a n d t he d irec tion o f g a z e outli n e d i n gre e n . ( T o p Ri gh t)
The visual field with no visual pr os the s i s . ( Bo tt om L e f t ) H e a d‐p ointe d c a s e in w hich t he i m a g e
from the central FOV appears in the direction of gaze. (Bottom R ig ht) Eye‐point e d case i n
whi c h the re g i on o f the ca me ra F OV t h a t i s aligned w ith the directi o n of g aze ap pears in t he
direction o f g a z e.

11

Consider now a subject with AMD who has the same intraocular re tinal
prosthesis i mplanted, and operat ed i n either the h e ad‐ p ointed or eye‐point e d
mode, just a s described for the case o f RP. T he r esults o f a s imilar s im ulation are
shown in F igure 6. A s u bject with A MD i s lik e ly t o hav e t heir perip h eral v ision
intact, but has lost their central (macular) vision. This is d emonstrated in t his
example by a darkened shroud added to the scene camera image, w hich
repres ent s a c en tral s c o toma t hat is a pprox imately 30° i n the v isu a l field, a n d
overfills the macular vision region by a factor of two. With the intraocular retinal
prosthesis implanted, t h e regio n of centr a l vision is now p a rti ally restored.
The scene c a mera i mag e f or t he A MD c ase is e ssen t ially i dent ica l to t hat
for the RP c ase, b ut w hen using a head‐pointed c amera system i t w ill now be
apparen t t o the subject that t hey perceiv e t w o vers ion s o f the same obje ct a t
differ ent locations, a s shown in F igure 6 (Bottom Left). On e p erc e pt i s due to
natural vis i on i n the peripheral r egio n, a nd t he o ther p ercept is a rtific ially
generated b y the retinal prosthesis. Switching to an eye‐point ed c am era system
would relieve this duplicatio n o f in formatio n and resultin g con fusion i n location
b y ensu r ing th at th e p erce p t is al ways al i gn ed with th e cu rren t gaze direct ion as
demonstrated in Figure 6 (Bottom Right).

12

Fi gure 6 . Si mulati on o f the e f fe ct s of f ovea tion o n a subject w it h Ag e‐Re la t e d Ma cul a r
D e ge ne rati on t ha t h a s b e e n i mplan t e d w i t h a re ti nal pr os the s i s . ( To p L e f t) I ma ge f ro m the
scene ca mera w ith the centr a l field of v iew ou tlined i n red, a nd t he direc t io n of g a z e o u t l ined
in green. (Top Right) The visual field with no visual prosthesis. (Bottom Left) Head‐pointed
case in which the image from the central field of view appears in the direction of gaze. (Bottom
R i g ht) Eye‐point ed c as e in whic h t he reg io n o f t he ca m era f i eld o f view t hat is aligned wit h t he
direction o f g a z e appe ars in t he direct i on of ga ze.
Interestingl y, for the c ase of AMD in which residual perip heral v is ion still
exis ts, in it ial experiments hav e demonst rated that p lacing i nco rrect o r
discordan t i nformat i on i n the pros t h etic v is ion area l eads t o a dditional c onfusion
and disorientation, as the percept does not agree with the surr ounding visual
informa t ion . I n fact, su bjects r epor t that t hey t e nd t o igno re t he p rosthetic vision
and simply revert to using prima ril y periphera l vision, as s ubj ects wit h AMD but
without an intraocular retinal prosthesis must do. However, when the eye‐
13

pointed ca mera mode is used, bot h task perfo rmance and comfort a r e improved,
as will be described in deta il in Chapter 3.
Such a n eye‐pointed ca mera s yst e m could be r ealiz e d in a n umber of ways.
First, the patient may wear an eye‐tracking device that continu ally m onitors the
movements of the eye and extracts the current gaze direction as it changes in
time. T hen , t his eye‐pointing d ir ection c ould b e used t o mecha nically r eorient
the direction of the camera by the use of servo motors in conju ncti on w i t h
mounting the camera on a gimbal mount. The key challenges asso ciated w ith this
ap p r oa ch are to im p l e m ent th e m otion syst e m with a h igh enou gh response rate
and low enough latency to follow the natural motions of the eye i n real time, a nd
to m inimiz e the poten t ial lack o f cosmetic a pp eal. U s i ng t his same eye direction
information, another approach (as in an eye‐tracked extraocular c amera) c ould
involve the use of a h ea d‐mounte d wide‐angle scene c amera in c o njunction w i th
a digital signal processor (DSP) to extract the sub‐region of the image that
corresponds to the direction of gaze, rather than extracting th e cen t er r egion of
the camera image. The visual prosthesis simulator presented in Chapter 2 is
based on t his concept of a n ey e‐pointed camer a c reat ed b y the c oupli n g of a n eye
tracker wit h a wide‐an gle scene c a mera.
A third app r oach t o acc o mplish t he r ecoupling of h ead and eye m otions is
to implant the camera inside the eye, at which point it becomes an intraocular
camera (IOC) as oppos ed to an extraocular camera (EOC) . The fo llowing s ection
14

describes o n e such c amera that h as b een p roposed to r estore f ov eation f or a
retinal pros thesis, or a ny visual pr osthesis tha t uses a ca m era fro nt‐ e nd.
1.4 Intraocular Camera
An intr a ocular camera (IOC) for r e tinal pros theses has been proposed for
implantat i o n i n the ey e itsel f , thereby provid ing a mea n s for n atur al f oveat i on.
The principal componen ts o f a n i ntr a ocular r et inal p rosthesis i ncluding t he
intraocular camera a re i llustrated in F igure 7 [ 10, 2 7–34]. I n general, an
intr aocular camera m ust implemen t functions of i mage f ormation with a
resolution and quality suitable for display on the retinal pros thesis, as w ell as
image acquisit ion a n d signal c ond i tion ing as a ppropriat e for th e g i ven electrod e
array. The proposed in t raocular ca m era des i g n meets these requ irements whil e
still mainta ining a form f actor small enough t o be s urgically i mplanted i nsid e the
crystallin e l ens sac fol l owing pha c oemulsification of t he b iolo gical crystalline
lens.
The ima g es c aptured by t he i nt raoc ular c amera can subseq uently be s ent
wirelessly to an external processing unit, or can be processed en tirely i nsid e the
eye, a s indicated by t he h ermetic a lly s ealed electron ics housin g in F igure 7. A
custom a spherical lens i s design ed t o work i n conjunction with the biological
cornea to form an image at the image sensor array at the back o f the intr aocular
camera housing. The IOC housing itself is h ermetically sealed with a f used s ilic a
window a t t h e anterior e nd, and is b iocompatible w ith an i mplan ted lifetime g oal
15

of at least 30 years. An enlarged view of the IOC camera and its elements is shown
in Figure 8.

Fi gure 7 . Re ti nal pr os t h e s i s w i t h in tra ocul a r c a mera, showin g t he suppo rt e lect ro nics
enclosed in a hermet ica lly sea led housing a n d a n ul t ra ‐ f lexible ribbon ca bled c o nnect ed t o t he
microstimul a tor array.

Fi gure 8 . Re ti nal pr os th e s i s w i t h i n tra ocular ca me ra ; e n large d view. The IOC is com posed of
a hermeti c ally sealed housin g, s ilica windo w , cus t om p olymer le ns, an d CMO S im a ge sensor.
16

In addition to the complementary metal‐oxide‐semic o nductor (CMO S)
image s e ns or a rra y, a d ditio nal CM OS c hips m ay b e included w ith i n the housing
for control or sensing functions. In order to meet surgic a l co nstraints, the entire
camera must be ultraminiature in its physical dimensions, and t o meet b iological
constraints, the power dissipated by the camera must be minimized to reduce the
risk o f thermal damage t o the re tinal tissue. Because of t hese s trict constraints,
the imagin g performa nce of t he i ntraocula r c amera is l ikely nev er t o be
competitive with consumer grade cameras, such as those found in c ellular
phones. H owever, the camera m ust only f orm images w ith quality and resolution
suitable for current and envisioned microstimulator arrays. Fu rthermore, k ey
optical ima g ing and image process i ng t echniq ues can be e mployed to ensure that
the best p ossible phosphene‐base d percept is deliver ed t o the s ubjec t i n the case
of l ow p ix ellation. S ev eral o f thes e basic p i xellation e ffec ts are discussed in the
following s ection, which then f orm the basis for var i ous image processing
algorithms i mplemented i n the visual p rosthesis simulator (pres ent e d in C hapt er
2), and ar e the impetus for many o f the exper i ments pres ent e d i n Ch apter 3.
1.5 Effects of Pixellation
Stimulation of p hosphenes at t he r etina using a microstimulator a rray
poses sign ificant challenges w hen attempt i ng t o recr eat e a h igh r esol ution imag e
of a scene. First, the microelectrode array has a much lower r esolution tha n a
typical consumer display. The first clinical trials with the Argus I intraocular
17

retinal prosthesis w er e conducted with a n arr a y comprising o nly 16 electrodes
arranged i n a 4 4 grid, which is a pproximately 1,000 times fe wer pixels i n each
linear dimension compared with the overall resolution of the re ti n a . A s expe cte d ,
the function al v ision r e s t ored t o pat ien ts during this i nit i al t r ial was quite limited.
However, after some training, su bjects w ere able t o differentia te c ommon
household objects such a s a cup and a plate, a nd w ere able t o c orrectly i dentify
bars at different o r ienta t ions [2 3 , 2 5 ].
With t he n ext generatio n A rgus II device, the el ectrode ar r a y n ow c onsists
of 60 discrete electrodes in a 6 10 rectangular grid [9, 25]. Although a marked
improvement in r esolution over t he A rgus I , st udies i nd ic ate th a t a pproximat ely
625 electro des a re n eeded to r esto re s ufficiently functio n al v i sion f or e veryday
life [10, 26, 35]. To this end, it is important to ensure that t he b est possibl e
repres ent a t i on b e g i ven to t he su b ject, espec i ally w ith t h e con siderably lower
number o f electrodes i n the curr ent implement a tio n o f Argus II. Efforts ar e
underway to better understand what patients actually see with s uch prostheses,
but a rudimentar y im age proc ess i ng s cheme can be e nv ision e d to understand
what the pa t ients may see, a nd to explore opt i mal display schem es.
Consider a retinal prosthesis consisting of a 16 16 square gr id for a total
of o nly 256 discrete p icture e le ments. T his can be r epresented visually by
segmenting, or pixellating, an image into 16 16 patches of li ght, a s shown in
Figure 9 (Center) based on the original image (Left). To furth er c omplicate th e
image, the electrodes are circular and cannot be touching in or der to m aintain
18

electric al i s o lation. T his can be v is ualized by m aking eac h p a tch of l ight s maller
and circular, with s om e spacing in b etween t he e lectrodes, a s r ep resented b y
Figure 9 (Right), henceforth referred to as a process of “gridding”. By simply
comparing the gr idded image to t h e o rig i n a l visually, one can s ee t ha t the imag e
quality is severely d e gr aded to the point that t he image m ay be unrec ognizable.
F i g u re 9. An o rig i na l i mag e at a re s olut ion of 5 1 2 512 pixel s (Left), down‐sampled and
pixella t e d to a l evel o f 16 16 pixels ( Cent er) a n d then conve rt ed t o a n a rra y of c irc ular
elect rode s with a 50% dut y cycle (Rig ht ).
One s t rateg y f or i mproving t he o v e rall qualit y and int e r p retati on o f this
image is t o filter o ut t he h igh‐ spatia l‐frequency conten t tha t was added due to t he
sharp edges of the electrodes. Another strategy could be to sp read ( blur) the
percept from e ach electrode into t he r egio n of e ach neig hboring electrode in an
attempt to r emove the empty space between electrod es. F ortunat ely, b oth of
these str a t e gies c an b e employed w ith pr oper desig n of t he e l e c trodes a nd
proximity c o upling of t h e elec tric fields and currents produced by the electrod es
within t he r etinal tissu e.
This e lect ric field a n d c u rrent s pr eading f rom the electro d es a s a function
of the intimacy of proximity coupling to the retinal surface ma y be v isualized by
19

a simple and well‐known Gaussian blur function applied to the g ridded image,
whereby each pixel is blurred by Gaussian convolution and sprea d into t he
neighborin g region a r o und each e lectrode. A G aussian b lur function can be
defined by a G aussian k ernel as d escribed i n E q uation 1 , and th en c on volved w ith
the image.
22
2
2
2
1
(, )
2




 



x y
Gx y e

(1)
In E quation 1, x i s the d i sta n ce f ro m the imag e orig in i n t h e horizont al direc t ion,
y i s the distance f rom the image origin i n th e vert ical direct i o n, a nd σ is the
stand a rd deviat i on o f t h e Gaussian distr ibution. T ypic ally σ i s expressed in u nits
of the number of pixels, but in this thesis, σ will be expressed as the percentage
of the standard deviation size compared with the size of each p ix el lated r e gio n .
For example, t he o rig i nal image in F igure 9 is 512 512 pixels . A fter down‐
sampling to 16 16 p i xels, each block is 32 pixels across, m ai ntai ni ng an ove r all
image siz e o f 512 512 pixels. A pplying a G aussian b lur with a stand a rd
deviatio n of 16 pixels w ould be expressed in sh o rthand as a 50% Gaussian blur.
The effect of such a Gaussian blur on the gridded image from Figure 9 is
shown in F igure 10 w it h three different l evels of b lur applied. Counterintuitively,
the more blur that is added, the more “in focus” the image becomes. In this
example, quite a large amount of blur (> 50%) is needed to overcome the gridding
effect and reproduce a remarka bly recogniz able image [10].
20

F i g u re 10. A g ridded version of t h e orig ina l ima g e f rom F i g u re 9, b ut with addi tional G aus s ian
blur t o simul a te e lectric field spre adin g fr om t he e lectr odes i n t o the re ti nal ti s s ue . Gaus s i a n
blurs wit h s t a nda r d devia t i o ns s et t o 18 percent (L eft ) , 31 percent ( Cent er), a nd 62 percent
(Righ t ) are sh own here.
The simulations pres ented in t his thesis r epr e sen t t his idealiz ed i mag e
processing procedure, and are not necess arily perfec tly matched to what is seen
by p atients using the curren t A rgus II dev i ce. H owever, an alyz ing thes e id ealiz e d
perceptio n s of p ixellat i on a nd b lur is m ean t t o driv e the desig n for future
intr aocular retinal pro s theses, a n d to i nd icat e just h ow c apable such a visual
prosthesis c an be at i ts f ull potential.
1.6 Summary and Organization of the Thesis
The curren t c hapter p resen t ed a n overview o f some o f the m a jor
successes and improvement opportun ities for visual p rostheses, in p a r ticular for
intr aocular retinal prostheses w ith extr aocular and int r aoc u lar c ameras. S everal
unique c hallenges w ere pr esen t e d in t erm s o f both p hysical a n d imag ing
constraints, particularly in the limit of low pixellation. The n ext chap ter explores
a number o f the imag e processing techniques that c an b e impleme nt ed i n a real‐
21

time s ys tem to s imulate such v isual prosthes es, in o rder t o bet ter experience a nd
understand what subjects with vi sual prostheses may be c apable of perceivin g.
A novel ha rdwar e a n d softwar e implement a tion o f a visual p rosth esis
simulator will be p res e nt ed t hat can not o n ly s imulate pr osthetic v is ion, b ut c an
do s o for both R etinitis P igmen t osa and Ag e‐Related Mac u lar Degen erat ion, a n d
in b oth eye‐pointed and head‐pointed c amera modes, a s described e a r lier i n this
chapter. Chapter 3 then presents a set of visual psychophysics e xperiments u sing
the visual p rosthesis simulator in order to quantify, in severa l ways, the benefit
of foveation afforded by an intraocular camera or eye‐tracked e xtr aocular
camera. A well‐known visual sea rch t a sk i s pe r f or me d for t h e firs t t ime w i th
prosthetic v ision simu lation a nd c ompared to h istorical visual search data from
normally sighted subjects. A novel functional reach and grasp exper i ment t ha t
makes use of m onocular d epth c ues is t hen p r esented, f o llowed b y a navigation
and mobilit y task.
The second half of the thesis focuses on improvements to implantabl e
devices in g eneral, as w ell as o n a ser ies o f improvements a nd analyses o f the
latest intraocular camera design. In Chapters 4 and 5, a posit ive mobile i on
contamina t ion sensor i s presente d tha t c an v alidat e the her m eti city o f
biomedical d evice pa ckages e nc apsulated by h ermetic coatings, and may be
int e gra t ed w ith future i mplantable dev ices t o provide ea rly warni ng o f le aks or
contamina t ion that m a y j eopardiz e the impla n t. A n integ r ated CMOS clock circuit
capable of driving the image sensor array on a commercial image sensor is also
22

presen ted, a long w ith s e ver a l key experiments proving the utili ty o f such devices.
The final t w o chapters p resen t some advan c ed i nt raoc ular c amera desig ns,
summarize t he w ork presen ted in t his th esis, a n d s u ggest f utu r e r es ea rch
directions.

23

CHAPTER 1 REFERENCES
[1] R. D . Jager, W . F. M ieler, and J. W . Miller, “ Age‐ Related Macular
Degeneration,” New England Journal of Medicine, vol. 358, no. 24, pp.
2606–2617, 2008.
[2] N. M . Bressler, S. B . Bressl er, and S. L . Fine, “Age‐relate d macular
degeneration,” Survey of Ophthalmology , vol. 32, n o. 6 , p p . 375–413, 1988.
[3] D. T . Hartong, E . L. B erson, a nd T . P. D ryj a , “ R etinitis p i gmentosa,” The
Lancet , vol. 368, no. 9549, pp. 1795–1809, 2 006.
[4] G. J . Farrar, P . F. K enna, an d P. H umphries, “On th e g e n e ti cs of re tin i ti s
pigmentosa a nd o n mutation‐ind ep end e nt a pproaches t o therapeuti c
int e rv ent i o n ,” The EMBO Journal , vol. 21, no. 5, pp. 857–864, 2002.
[5] M. S . Huma yun, E . de J uan, J r., G. D a g nelie, R . J. G reenberg, R. H. Propst, and
D. H . Phillips, “Visual Percepti on E licited by E lectrical Stimu lation o f
Re ti na i n Bli n d Humans,” Archives of Ophthalmology, vol. 114, no. 1 , p. 40,
1996.
[6] M. S. Humayun, E. de Juan, J. D. Weiland, G. Dagnelie, S. K at ona,
R. Greenberg, and S. Suzuki, “Pattern electrical stimulation of the human
retina,” Vision Research , vol. 39, no. 15, pp. 2569–2576, 1 999.
[7] M. S. Humayun, J. D. Weiland, G. Y. Fujii, R. J. Greenberg, R . Williams o n,
J. L ittle, B . Mech, V. C immarusti, G . van Boemel, G. D agnelie, an d E. de Juan ,
Jr., “Visual perceptio n i n a blind s u bject with a c hronic m icroelectr onic
retinal pros thesis,” Vision Research , vol. 43, no. 24, pp. 2573–2581, 2 003.
[8] J. D . Weiland, W . Liu, a nd M . S. H umayun, “Retinal p rosthes is,” Annual
Review of Biomedical Engineering, vol. 7, pp. 361–401, 2005.
[9] J. D . Weiland, A . K. C ho, and M. S . Humayun, “ Retinal prost heses: c urrent
clinical r es ults a nd f u t ure n eeds, ” Ophthalmology , vol. 118, n o. 1 1, p p.
2227–2237, 2011.
[10] N . R. B . Stiles, B. P . McIntosh, P. J . Nasiatka, M. C . Hau er, J. D . Weiland,
M. S. Humayun, and A. R. Tanguay, Jr., “An Intraocular Camera f or R etinal
Prosthes es: Restorin g Sight to t he B lind”, i n Optical Processes in
Microparticles and Nanostructures: A Festschrift Dedicated to Richard
Kounai Chang on His Retirement from Yale University , World Scientific
Publishing Company, Singapor e, ch. 20, pp. 385–430, 2 010.
24

[11] D . S. F riedman, B . J. O 'Colmain, B. M uñoz, S. C . Tomany, C . McCa rt y,
P. T. de J on g, B . Ne m e su re, P. Mitc h el l , and J . Kem pen, “P reval ence o f Age‐
Related M a cular Degeneration i n the United S tates ,” Archives of
Ophthalmology , vol. 122, no. 4, p. 564, 2004.
[12] D . H. H ubel, Eye, Brain, and Vision , in S cien tific American L ibrary S eries,
vol. 22, W. H. Freeman and Company, New Yo rk, New York, 1988.
[13] R . E. M acLaren, R . A. P earson, A. M acNeil, R. H . Douglas, T. E . Salt,
M. A kimoto , A. S waroo p , J. C . Sowden, and R. R . Ali, “ Retinal r epa i r by
transplantation o f p hotoreceptor p recursors,” Nature , vol. 444, n o. 7 116,
pp. 203–207, 2006.
[14] K . Canola, B. A ngén ieux, M . Teka y a, A . Quiambao, M. I . Naa sh,
F. L . Munier, D. F . Schorderet, and Y. A rsenijevic, “Retinal S t em C ells
Transplanted into Models of Late Stages of Retinitis Pigmentosa
Preferentially Adopt a Glial or a R e tinal Ganglion C ell Fate,” Investigative
Ophthalmology and Visual Science , vol. 48, no. 1, pp. 446–454, 2007.
[15] C . Veraart, M .‐C. W anet‐Defalque, B . Gérard, A. V anlierde, and J. Delbeke,
“Pattern R ecognitio n w ith the Op tic Nerve Visual P rosthesis,” Artificial
Organs , vol. 27, no. 11, pp. 996–1004, 2003.
[16] E . M. S chmidt, M. J . Bak, F . T. H ambrecht, C. V . Kufta, D . K . O'Rourke, and P.
Vallabhanat h, “ Feasibility of a v isu a l prosthesis f or t he b lind based on
intracortical micro stimulat ion of t he v isual cortex,” Brain , vol. 119, n o. 2 ,
pp. 507–522, 1996.
[17] W. H. Dobelle, M. G. Mladejovsky, and J. P. Girvin, “Artif icial Vision f or t he
Blind: Elect rical Stimulation of Vis ual Cortex Offers Hope f or a Funct i onal
Prosthes is,” Science , vol. 183, no. 4123, pp. 440–444, 1974.
[18] G . S. B rindley and W. S . Lewin, “ The sens ations p roduced b y elect r ical
stimulation of t he v isu a l cortex,” The Journal of Physiology , vol. 196, n o. 2 ,
pp. 479–493, 1968.
[19] R . R. L ak hanpal, D. Y anai, J. D . Weiland, G . Y. F ujii, S . Caf f ey,
R. J . Greenberg, E. d e Juan, Jr., and M. S . Humayun, “ Advanc es in t he
development of v isual prostheses, ” Current Opinion in Ophthalmology , vol.
14, no. 3, p p . 122–127, 2003.

25

[20] E. Margalit, M. Maia, J. D. Weiland, R. J. Greenberg, G. Y . Fujii, G . Torres,
D. V . Piyathaisere, T. M . O'Hear n, W . Liu, G . Lazzi, G. D agneli e, D . A.
Scribner, E. d e Juan, Jr., and M. S . Humayun, “ Retinal Pro s thes is f or t he
Blind,” Survey of Ophthalmology , vol. 47, no. 4, pp. 335–356, 2002.
[21] L . Renier, O. C olligno n , C. P oirier, D. T randuy, A. V anlierd e, A . Bol,
C. Veraart, and A. G. De Volder, “Cross‐modal activation of vis ual c o rtex
during d epth p erception usin g auditor y s ubstitution of v is ion,”
Neuroimage , vol. 26, no. 2, pp. 573–580, 2005.
[22] M . Ptito, S . M. M oesgaard, A. G jedde, a nd R . Ku pers, “Cros s‐modal plas ticity
revealed b y electrotac t i le stimulat i on o f the t o ngue i n th e con genitally
blind,” Brain, vol. 128, no. 3, pp. 606–614, 2 005.
[23] M . S. H umayun, J. D . Dorn, L. d a Cruz, G. D agnelie, J .‐A. Sahel, P. E. Stanga,
A. V. Cideciyan, J. L. Duncan, D. Eliott, E. Filley, A. C. Ho, A. S antos, A . B.
Safran, A. Arditi, L. V. Del Priore, and R. J. Greenberg, “Inte rim results from
the in ter n at ional tr ial o f S econ d Sight's visual p rosthesis,” Ophthalmology ,
vol. 119, no. 4, pp. 779–788, 2012.
[24] D . Zhou a n d E . Greenbaum, Implantable Neural Prostheses 2: Techniques
and Engineering Approaches, Springer S cience+Business Media LLC, New
York, New York, 2010.
[25] M. P. Barry and G. Dagnelie, “Use of the Argus II retinal prosthesis t o
improve v i sual g uidance of f ine hand m ovements ,” Investigative
Ophthalmology and Visual Science , vol. 53, no. 9, pp. 5095–5101, 2012.
[26] M . S. H umayun, “Intr a ocular r et inal p rosth e sis,” Transactions of the
American Ophthalmological Society , vol. 99, pp. 271–300, 2001.
[27] P. J. Nasiatka, M. C. Hauer, N. R. B. Stiles, A. R. Tangua y, J r., and
M. S. Humayun, "Intraocular Camera for Retinal Prostheses", U.S . Patent
8,197,539 B 2 , issued June 12, 2012.
[28] M . C. H au er, P. J . N a siatka, N. R . B. S til e s, J .‐C. L ue, R . Agrawal,
J. D . Weiland, M . S. H umayun, and A. R . Tanguay, J r., “Intraocu lar Camera
for Retinal Prostheses : Optical Desig n ,” Annual Meeting of the Optical
Society of America, San Jose, California, 2007; FiO/LS/OMD Conference
Program , Paper FThT1, p. 142, 2007.
[29] M. C. Hauer, “Intraocular Camera for Retinal Prostheses: Refractive a nd
Diffractive Lens Systems,” Dissert ation, University o f Southern Califo r nia,
Los Angeles , Califo r nia, 2009.
26

[30] P . J. N asiatka, A . Ahuja, N . R. B . Stiles, M. C . Hauer, R . N. Agrawal, R. Freda,
D. Güven, M. S. Humayun, J. D. Weil and, and A. R. Tanguay, Jr., “Intraocular
Camera for Retinal Prostheses”, Investigative Ophthalmology and Visual
Science , vol. 46, no. 5, ARVO Abstract 5277, P aper B480, 2005.
[31] P . J. N asiatka, A . Ahuja, N . R. B . Stiles, M. C . Hauer, R . N. Agrawal, R. Freda,
D. Güven, M. S. Humayun, J. D. Weil and, and A. R. Tanguay, Jr., “Intraocular
Camer a D esign for Retinal Prost h eses”, Annual Meeting of the Optical
Society of America, Tuscon, AZ, 2005; FiO/LS Conference Program , Paper
FThI4, 124, 2005.
[32] P. J. Nasiatka, M. C. Hauer, N. R. B. Stiles, L. Lue, S. T akahash i , R. N . Agrawal,
R. Freda, M. S. Humayun, J. D. Weiland, and A. R. Tanguay, Jr., “Intraocular
Camera for Retinal Prostheses”, Investigative Ophthalmology and Visual
Science , vol. 47, no. 5, ARVO Abstract 3180, P aper B554, 2006.
[33] P. J. Nasiatka, M. C. Hauer, N. R. B. Stiles, L. Lue, S. T akahash i , R. N . Agrawal,
J. D . Weiland, M . S. H umayun, a n d A. R . T a nguay, J r., “An Intra ocula r
Camera f or R etinal P ro stheses”, Proceedings of BioMed , Irvine, Califo rnia,
7‐8 June, 2006.
[34] P . J. N asiatka, B . P. M cIntosh, N . R. B . Stiles, M. C . Hau er, J. D . Weiland,
M. S . Humayun, a nd A . R. T anguay, Jr., “An Intraocular Camera f or
Provis ion o f N atural F oveat i on i n Retin a l Pro stheses”, Proceedings of the
2010 Neural Interfaces Conference , Long B each, California, 21‐23 J une,
2010.
[35] N . R. B . Stiles, B. P . McI n tosh, P. J . Nasiatka, J. D . Wei land, M. S . Humayun,
and A. R . Tanguay, J r., “Intraoc ular C amer a for Retinal Prosth e ses:
Psychophysical A n a lysis of I mag e S ampling an d Filter ing”, Proceedings of
the 2010 Neural Interfaces Conference , Long Beach, Califo rnia, 21‐23 June,
2010.
[36] C . Y. Chen, T. Y. Wong, and W. J. Heriot, “Intravitreal Be vac i zumab
(Avastin) fo r Neovascular Ag e‐re lated Macular Degeneration: A Short‐
term Study”, American Journal of Ophthalmology, vol. 143, no. 3 , pp.
510–512, 2007.
[37] N . Sabbah, C. N. Authie, N. Sanda, S. Mohand‐Said, J.‐A. Sahel, A. B. Safran,
“Importance of Eye Position on Sp atial Localization in BlindSub jects
Wearing an Argus II Retinal Prosth esis”, Investigative Ophthalmology &
Visual Science , vol. 55, no. 12, pp. 8259–8266, 2014.

27

CHAPTER 2
VISUAL PROSTHESIS SIMULATOR
2.1 Introduction
A visual p rosthesis simulator (V PS) has been d eveloped t o allow s ighted
volunteers to wear an apparatus that simulates the experience o f bein g
implanted with a retinal prosthe sis in b oth head‐pointed a nd e y e‐pointed camer a
modes, as well as with both RP an d AMD conditio ns, as described earlier. Such a
system allows for the study and quantification of the effects of head and eye
pointed cameras, as well as the effects of pixellation in confi guratio n s similar to
those exper ienc e d by subjects with i ntraocular r et in al p rosthes es. A dditionall y ,
a key feature of this simulator is the ability to dynamically a djust the simulate d
electrode a rray par a meters ( including blur, electrode density, and field of v iew)
all in a portable and wearable system to allow a wide range of psychophysical
exper i ment s to be performed.
The system i ncludes a wide a ngl e c amera affixed to a h ead‐moun ted
display (HMD), and incorporates eye‐tr ackin g to det e rm ine the direction o f gaze
and to extract the appropriate region of the visual field for display in the HMD. In
addition t o quantifying the bene fit of f oveation o r comparing v arious s ystem
configura t ions, the VPS can also b e used t o restor e fov e at ion for blind patients
with t he c urrent i mplementa t ion of a n intr aocular retinal prost hesis.
28

Furthermore, the VPS serves as a platform that can eventually be miniaturiz ed
and developed into a n eye‐tracke d extraocular camera f or i ntrao c u lar retinal
prostheses, which would serve a s a n additional t herap e utic a lte rn ativ e for R P
and AMD should a fully implantable intraocular camera prove not to be a viable
option for c ertain subjects.
2.2 Prior Art
Several res e arch g rou p s have d em onstrated various levels o f pro sthetic
vision s imulation, w ith the most c ommon tasks being visual a cui ty t ests a nd
read ing t e xt [1–13]. T ypically, 6 to 12 subjects w ere r e cruit e d for each
exper i ment , and a stro ng l ear n in g effect w as o bserved in n early all studies. A
summary o f key simulation v ariabl es, experimental t asks, s i mula tion p arameters,
and sample size is shown in Table 1. Common to all known visua l prosthesis
simulations is t he s tudy o f RP, whereas both R P and AMD have b e en s t u died only
in t his thesis. I n ad ditio n , ther e have b een sev eral k ey v isua l prosthesis
simulations in which the percept is stabilized in such a way that the prosthetic
simulation is aligned with the direction of gaze. Although the v isual prosthesis
simulations presented herein a ls o stabilize the imag e by a lignm ent with t he
direction of gaze, they additionally allow for both head‐pointe d and eye‐point e d
camera m odes, which will serv e to a llow the complete s imulatio n of both
extraocular and intrao cular cameras. I n summary, the key disti nction b etween
prior art and this t hesis work i s the ability to simulate both RP and AMD modes
29

while also a llowing f o r h ead‐poin ted and ey e‐pointed c a mera m odes for each
conditio n.
Ta ble 1. S umma ry o f key prost h e t ic v ision simul a tion ref erences i ncludin g con ditions simula ted
a n d pa ra me t e rs releva n t to t his t hesis resea rch.
Ref.
Primary
Variables
AMD/
RP
Percept
Stabilized
Camera
Control
Amount of
Blur
N Tasks
[8]
Number o f
pixels an d
spacing
RP N o He ad N one 6 Vi sual a cui t y
[9]
Number o f
pixels an d
spacing,
mag n if ica t i o n
RP N o Head N one 7
Navi ga ti on
throu g h maze
[5]
Number o f
pixels an d
spacing , size,
drop ou t, levels
RP N o Mouse None 4
Fa cial
recogni tion
[2]
4 4, 6 10,
a n d 16 16
array s , con t rast
RP N o Head L igh t 8
Recog n it ion,
reading, an d
co ordin a tion
[1, 6, 7]
Number o f
pixels an d siz e,
array
eccentrici ty
RP Y es Eye L ig ht 5 R ea ding
[13]
Rectan gular vs.
hexagon a l gri d
RP N o Joyst i ck M edium 13 V isua l a c ui t y
[11]
Number o f
pixels an d
spacing , size,
drop ou t, levels
RP Y es H ea d L i g h t 4 Rea d ing
[12]
Pixel size,
fi lte ri n g
methods
RP N o Hea d M edium 12 V isua l a c ui t y
[4]
FOV mapping
methods
RP Y es H ea d Medium 1 2
Acuity,
recogni tion,
and
co ordin a tion
30

Due to the complex nature of these psychophysical experiments a nd t he
associated burden placed on subjects both mentally and physical ly, the duration
of the trials is usually kept relatively short and the number o f subjects r ecruited
is typically on the order of 6 to 12. In addition, a majority of e xper iments w ith
visual p rosthesis simulation t o date h ave used v isual stimuli f rom a computer
monitor fo r tasks such a s visual acuity, reading, and facial re cog n it ion. N o
exper i ment s have b een f ound i n th e literature o n visual search using a wide f ield
of view display, a functional reach and grasp task that involve s both h ead and eye
movements, or a full‐sc ale naviga tion and mobility experiment. Whil e there was
a navig a tio n e xper iment w i th s im ula t e d pr ost he t i c vi si on i n one r eference, the
navigat i on w as p er for m ed i n a virtual g a me e nv iro n ment w it h no actual
ambulation [9]. In Chapter 3 of this thesis, results on each o f thes e experiments
are pr esented to fill this gap in the liter a ture.
2.3 Hardware Implementation
Simulation o f an a ccurate p rosthe tic visio n e xp erien c e in r eal tim e
requires a substantial amount of computing power to process the video frames
captured b y an e xt er n a l scene ca mera. F urt h ermore, th e simulat io n of r estor e d
foveation, a s would be p rovided b y a n intraoc u lar camera, requi res a wide angl e
scene c a mera u sed in c onjunction w ith on e or m ore ey e tr ackin g methods in
order to p r o vide t he c apability fo r close to f ull peripheral v ision in the case of
31

AMD, and to provide the capability for a full range of foveation in the case of an
eye‐ p oint e d ca m e ra. In stead of al w a ys e xtr acti ng th e cen tr al reg ion o f the scen e
camera a s with t he h ead‐pointed camera, the eye tr acker provide s the subject’s
real‐time g a ze dir ectio n so that t he a ppropria t e reg i on o f the scene image can be
extracted and displayed in t he d ir ection o f g a ze, thereby simul ating the percept
that would be acquired by an intraocular camera.
F i g u re 11. F irst g enerat ion visua l p rost hesis simul a tor ha rdw a r e overview s howing t wo PCs,
with one runn i ng cus tom VP S s o ft ware a n d the o ther ru nning eye tr ac ki ng s o f tw are .
In t he f irst g enerat ion VPS shown in F igur e 1 1 , two P C s were u s ed: one
with an Intel Core 2 Duo processor and PCI video capture card ( EPIX I nc., Buffalo
Grove, I L) t hat is u sed for the scene camera p rocessing, and an other PC w ith an
Intel Co re 2 D uo p rocessor and c o mmercial eye tracking s oftware ( Arrington
32

Research, S c ottsdale, AZ) that i s used f or g aze ext r act i on. A n E t h ernet link
between th e PCs allows f or f as t tr ansfer o f g a ze c oordinates to b e used i n the
simulated image construction. A third monitor is used to split the video from the
HMD and show the experimenter a live view of exactly what the s ubject i s seeing.
In the current second generation VPS, the system was reduced to a single
PC b y instal ling m ultiple video cap t ure cards in a single machi ne, optimizing t he
software to fully take advantage of the dual core architecture, and utilizing an
upgraded G eForce G TX 480 graphics c ard (Nvidia Corporation, S an ta C lara, CA).
C o mpute Unif ied Device A rchitecture (CUDA) p rogramming l ibrarie s that
execute d i rectly t hro u gh t he g raphics processing u nit (GPU) wer e als o
implement e d to r eg ain real‐time blur p roce ssing w ith the reduce d computing
resources. The reason that a push was made to consolidate the hardware i nto
one machin e is t he i d e a that e v e ntu a l miniaturization of t h e s y ste m could le a d to
both a fully mobile visual prost hesis simulator as w ell as a n e ye‐track ed
extr aocular camera d evice for treatment of R P and AMD. T hes e a dd it iona l
application s are e xplored in d eta i l in Sect i on 2 .7 of this chap ter.
In a ddition t o the computing hard ware, a head‐mounted display ( HMD) i s
worn b y the subject and is u sed t o display t he simulated p rosth etic i mages th a t
are v iew ed b y the subject. U sing a n HMD as t he displa y devic e, rather than a
computer m onitor, allows f or m obile operation and can be u sed as a platform f or
mounting o ther s ensors, such a s cameras, h ead‐trackers, and acc el er ometers. A
wide‐an g le s cene c amera is m ounted t o the HMD for image acquisi tion o f the
33

environment, a nd a second ca mera p oint ed t oward the eye allows gaze
coordinate tracking. The first generation VPS system used the Z800 H MD
(eMagin Corporation, Bellevue, WA), and is shown in Figure 12. This HMD is
primarily marketed as a consumer p rodu ct for gam ing an d ente rtain ment, but is
relatively inexpensive and served as a good development platform. A 50‐foot
tether cable carrying power and video signals connected the HMD to the PCs in
order to all ow for navigatio n and mobility ex p erimen ts.
F i g u re 12. The f irst g enera t i o n visua l p rost hesis simula tor Z80 0 he ad‐ moun t e d d i s pl ay ( HM D )
hardw a re sh o wing an eye‐tracking c amer a, a s cene cam era, an d c a b l i n g m o unte d t o the H MD .

34

While the Z800 prov ed t o be i nval uable during t he d evelopment a nd
test ing of th e V PS, as w ell as de v el opme nt of v a riou s psyc hophy sical experiment
protocols, i t ultimately proved to be ill‐suited for calibrated v isual psychophysics
exper i ments due to t he a mount of u nwanted ambient light leakage , poor h ead
mounting system, bulky cabling, overall weight, and limited fie ld‐of‐view of the
display.
The current s econd generation V PS s ystem replaces t h e Z 800 w i th the
zSight HMD ( Sensics, I nc., Columbia, Mary land) shown in F igur e 13, which is
marketed toward military and research applications. One key im pr ovement of
this HMD is the more sophisticated mounting of the eye‐tracking camera, which
provides a nearly on‐axis view of the pupil, als o shown in Figu re 13.

Figure 13. Sensics zSi ght head mounted display showing the scen e camera m ount ed t o t he fron t
( L eft ) a nd t he eye t rac k ing ca m e ra with ne arly on‐ axis vi e w of the pu pil (Righ t ).

35

Other impr ovements a fford ed b y the zSig ht i nclude a m uch higher
resolution, much w ider f ield o f view, an d gr eatly improv ed p hys ical stability o n
the head. T his new HMD also i nc ludes a wireless video option f or c omplete
freedom from c abling. In t his configurat ion, o nly a small‐di ame ter and
lightweight 75‐foot cable is required for the eye‐tracking and scen e ca mera v id eo
feeds back to the PC. A comparison of the first generation (Z800) and second
generation (zSight) HM D key spec ificatio ns is s u mmarized in Tab le 2.
Ta ble 2. Technica l compa rison of Z800 a nd zSig ht H MD performa n c e cha r act e rist ics.
eMagin Z800 Sensics zSight
Field of View 40° 70°
Display Technology OLED OLED
Contrast 20 0:1 10 00: 1
Brightness 5 f L 11 5 f L
Weight 23 0 g 45 0 g
Resolution 80 0 × 600 16 80 × 10 50
Wireless Video Option No Yes (100 m)
Geometric Distortion 1% 5%
Battery Option None 5 V, 2 H ours
Diopter Adjustment None –4 to +4 Dio pters
2.4 Software Implementation
The eye t r a c king softw are used i n the visual p rosthesis simulat or ( VPS)
was purchased from A r r ing t on R es earch, I nc., including a dynami c‐lin k ed l ibrar y
(DLL) written in C ++ t hat was provided t o inter f ace to o ther s oftware. A
commercial C++ library (EPIX, I nc ., Buffalo G rove, IL) was purc hased to h andle
any need ed i mage p r o cessing f unctions i nc luding q uantiz atio n (p ixellatio n ),
image r e siz i ng o r ro tat i on, and image mask in g. A n add i t i onal c ust om cor e C ++
36

library was developed to c apture t he scene c amera image, p erfor m the scene
camera s ub‐regioning b ased o n the gaz e d irectio n , perform any s ubsequent
image proc essing o peratio n s such a s post‐pixellatio n b lurring, an d provi d e
output t o the display devic e . A f lowchart r epresenting the ove rall image
processing algorithm implemented by the VPS is shown in Figure 14.
To g en era t e the simulated perc ept, f irst a b ackground image is formed
based on t he s imulated b lindness condit ion (AMD o r RP), a nd t he n a percept is
superimposed o ver th e background i mage d epend i ng o n the des i r e d camera
mode ( eye‐pointed or h ead‐pointed). I n the case o f simulated RP, the background
image is s imply black, b ut i n the case o f simulated AMD, a f ull ‐color
repres ent a t i on i s pres erved from t he scen e c a m era to s im ulate i ntact peripheral
vision surrounding the central sco toma (region of visual impair ment p rior t o the
addit i on of prosthetic v ision).
Care is taken to provide natural scaling of the background imag e
compared w ith what w ould b e seen b y a normal o bse r ver without t he H M D
apparatus. To accomplish proper scaling, the field of view of t h e HMD is
compared against the field of view of the scene camera and the captured image is
cropped or scaled to match the field of view of the HMD as view ed b y the
observer. T he g oal is t o replace the normal v iew o f the na t u ra l env i ro nment w i th
a camera‐deriv ed a nd H MD‐presented v iew t h at i s as w ell matched to the normal
view a s possible.
37

Figure 14. F low char t of t he v isual pr os thesis s imula t or c o d e s h o wing t he a l g ori t hm u sed to f or m
images for t he AMD an d R P mode s as well as the he a d‐p o inted an d ey e ‐pointed c a m era m odes.
38

Addit i onall y a simple dewarping algorithm can optional ly b e emp loyed
for scene cameras with very wide angle fields of view [14, 15]. The dewarping
algorithm used follows from the observation that the image surface of a fisheye
lens is a nearly spheric al surface that is projected onto a fla t image p l ane. W hen
solving a s i mple set o f geometr i c r e lation s bet w een t he distort ed a nd u ndistort ed
dista n ces measured f rom the center o f the image pla n e using t h i s spheric a l
mapping assumption, the result is a function that converts a di storted fisheye lens
image back t o a rectilinear ima g e of t he o riginal object ( scene ), a s given by
Equation 2.
tan




d
u
r
rf
f
(2)
In Equation 2, r
u
i s the distance f rom center t hat a pixel should b e placed t o
compensat e f or t he dis torted p ix el dista nce, r
d
. The focal length, f, or the distance
between the second principal plane of the lens and the image plane, may not be
known expl icitly, but may be t reat ed a s an a djustable para meter t hat, i n this c ase,
is s et e xperiment a lly s uch that t he b ackground scene camera i m a ge a ppears
undistorted. A simple look‐up ta ble based on t his equatio n c an be generated as
the parameter, f, is adjusted in order to quickly map an image with fisheye
(barrel) distortion to a rectilinear image in a computationally e fficient wa y.
Figure 15 s h ows a comparison o f a typical scene camera i mage b efore and after
the dewarp ing algorithm is u sed. T he e ffect is m ost pronounced t oward the edg e
of the image, where the pixel distances from the image center, r
d
and r
u
, are large.
39

F i g u re 15. Imag e s capt ured f rom t he scene ca mera b ef ore (L ef t) an d after (Ri g ht) dewarpi n g,
with the dewarping parameter, f, set to 1.39. In this case, the effective focal length is exp ressed
a s a mult i pl e o f t he senso r’s ha lf ‐d iag o nal d i st a nc e .
After the background image has been formed, the simulated perce pt i s
then g en er ated, d e pen d ent o n t he s elect e d ca mera m ode (eye‐poin ted or h ead‐
pointed). I n the head ‐pointed m ode, a sub‐regio n in the center o f the scene
camera image frame is taken to represent the simulated prosthet ic r egio n. A
similar pro c edure is f ollowed in t he e y e ‐pointed m ode, b ut i n t his case a sub‐
regio n centered on the current gaze coordinate is used. This sub‐region image is
then processed to make it appear close to what a subject with a n advance d r etinal
prosthesis m ight see, including:
1. A pix e llatio n e ffect wh ereby each e lectrode o n the ar ray giv e s th e
impression of a large d i screte p ixel, depending on t he e lectrod e siz e
and spacing. T his is a ccomplished using a spatial quantization fil ter.
2. A separ a tion o r grid ding e ffect whereby empty spa c e appears
between el ectrodes d epend i ng o n the fill factor o f th e electrod e
40

array. T his is a ccomplished b y multiplying an i mag e o f the gr id
pattern by t h e sub‐region pixe llat e d image, po i nt by poin t .
3. A blurring o f the edg e s of t he p ix els due to e lectric field and current
spreading caused b y the el ectr ic f ield dis tribution a b ove each
electrode a n d by t he p hysical sep a rat i on o f t h e electrod e from the
retinal tissue. This is accompli shed b y a real‐time Gaussian
convolution blur f unction employed o n the GP U using a CUDA k er n el.
4. A quantiza t i on o f perc ept in ten sit y, l imited t o as l ow a s 8 discrete
brightness levels, as w ith the current v isual processing u nit u sed
with the Argus II device [18]. This is accomplished by quantiz ing the
output pixel greyscale v alues to discrete l e vel s .
A demonstr ation o f t he p ixellatio n , gridding, a n d blurring step s for three
levels o f pixellation a nd t hree levels of Gaussian blur are shown in Figure 16
(6 10) a nd F igure 17 (12 20 and 24 40) below. I n each c ase, the effect of
blurring a cts counter‐intuitively to partially restore the imag e to a r ecogniz a ble
form. The most benefit from blurring is observed with the lowe st r esolution
array s.
41

6 10 Resolution
No Blur

30% G aussia n Blur

60% G aussia n Blur

Original Image

F i g u re 16. A d emonst rat i on o f t h e visua l prost hesis simula tor i ma g e proce s s i n g o u tpu t
showing a resolut i on o f 6 10 (simila r t o t he Arg u s II d evice) wit h va rio us sta g es o f
Gaus sian b lur of n one ( F irst R ow), 30% G a u ssian blur ( S e cond Ro w ) , a n d 6 0 % G aus s i an
blur (Third R o w). T he original imag e is shown for compa rison (Bot t om).

42

12 2 0 Res olution 24 4 0 Res olution
No Blur 30% G aussia n Blur 60% G aussia n Blur Original Image

Fi gure 1 7 . A d e mon str a ti on o f th e visu al p ro sthesi s s i mulator i ma g e p roce s s i n g o u tput s howi n g
a resolut i on o f 12 2 0 ( L e f t Column) a nd a resolut i o n of 24 40 (Right Column). Various stages
of Gaussian blur are shown of none (First Row), 30% Gaussian bl ur ( Secon d R o w), an d 60%
Gaussian blur (Third Row ). The orig ina l imag e is show n f o r com p a ri son (Bo t tom ) .

43

The simulation softwa r e has been u pgraded on t he second gener a t io n VPS
to include a real‐time Gaussian blur algorithm running on an Nv idia C UDA ‐
enab l e d GP U with 13 4 4 cor es, ra t h er than im plemented on the CP U. Th i s allows
for a highly p arallelized implementation of Gaussian blurring t o acc o mplish t he
composite simulation i n real‐tim e. This G P U b lur implement a tio n consists o f th e
following st e ps: (1) the image is Fourier tra nsformed to freque ncy space, (2) the
resulting image is m ultiplied b y the Fourier t rans form o f a Gau ssian k ernel, a nd
(3) the resulting imag e is i nv erse F ourier t ra ns formed t o provi de a t ru e Gaussian
blur w ith a processing t ime that i s independ ent of k ern e l size. This algorithm
allows the simulator to model the field and current spreading f rom eac h
electrode t h at i s caused b y the nonuniform e l e ctric field distr ibution above each
electrode modeled as a n eq uipotent ial surface, a nd b y placing the
microstimulator arr a y some dista nce from th e ret in al ga n glion c ell layer.
Finally, before being displayed to the subject wearing the HMD, the
simulated p e rcept imag e is t ra nslat e d to t he c urren t e ye‐tr a cking g az e coordinat e
and added to the background image. This is true for both cases of head‐pointed
and eye‐pointed cameras, r egardless of where the source of the sub‐region w as
taken. This s tep effec t ively st abil izes t he p ercept o n the sub ject’s v isual field,
simulating t he e ffect o f h aving a microstimulator array affixed t o the r e tinal
surface.

44

A custom g raphical u ser interfac e (GUI) was developed in C # usi ng t he
Microsoft .NET Framework Version 3.5 to interface with the core image
processing C ++ library described above. F igure 19 shows a scre en capture of the
GUI implementat i on t hat includes a series o f paramet e r settings c ontrols and
buttons to quickly load preset values. A set of real‐time benchmarking outputs is
also displayed for system health and performance monitoring. T he G UI a llows
for a n e xperiment e r using the V P S to q uickly c hange s i mulation pa ra me t e rs i n
real t ime a n d inst antl y monitor the video that i s being display ed t o the subject
wear ing th e HMD (on two separa t e monitors, a s shown in Figure 1 8).
F i g u re 18. The c urrent g enera t io n vis u al p ro sthesi s simulator w i t h experimenter monitors for
the gr aphi cal user interf a ce s, head‐m o un t e d dis p lay vie w , and s cene ca mera view.
45

Fi gure 1 9 . I ma ge c ap ture of the vi su al p ros t he si s si mulat o r gr a phica l user int e rf ace.
46

Sets of configuration parameters can be saved into groups and l oaded at a
later time. This allows for we ll‐defined pro t ocols and removes human error
when changing simulation parameters. The GUI also allows for i nt eract i o n
between th e VPS a n d t h e Arr i ngto n ey e‐track i ng softwar e for suc h purposes a s
data m ark i ng a nd e ye‐tracking c a libratio n without the need t o s witch to a
different software GUI. Lastly, a full set of benchmarks can be monitor e d in r eal
time f or e ar ly i nd ications t hat the simulation i s taxing s ys tem resources beyond
a reasonable degree. The benchmarking is also valuable for fur ther r efinemen t
of th e VPS p roce ssin g e f f ici e ncy, an d for fu t u r e ef forts to m in ia tu rize th e sys tem .
Further dis c ussion a nd a nalys i s of t he l ates t benchmarking r esu lts is p resented
in the following sectio n .
2.5 Benchmarking and Latency
Trad itio nally, benchmarking i s performed in c omputing a pplicati ons to
gauge the operatio nal performan c e of a sin gle system, to a ssess t he r elat ive
performanc e between t w o systems , or to analyze portions of the software t hat
need i mprovements in e fficienc y [16, 17]. B enchmarking has bee n used
ext e nsively throughout t he c ourse of t his inv e stiga t ion to a nal yze and optimize
the key soft ware c omponents. H erein, a n op erat ional p e rforma nc e benchmark
analys is o f the lates t h ardwar e and software i mplement ation obt ained to d at e
will be described in detail for the VPS. The value of benchmar king t o the
development of this visual prost hesis simulator application is threefold:
47

1. The benchmarking shows how close the simulation of prosthetic
vision currently is to real‐time operation, in the sense that s ubjects
can functio n r easonably well in t he e nv ironm e nt w hen u sing the
full image processin g capabil ity of the VPS co m puter.
2. Estab l ish i n g a b asel ine of th e VPS perform anc e to date wi l l h e l p in
comparing the per f ormance to f uture systems and in
understa nd ing how miniatur iza t io n effor t s m a y be a ble to r each
real‐time o p erat ion.
3. Monitoring system health o ver ti me c an h elp determin e if t he
hardware is aging o r failing.
The bench m arking p r e sen t ed i n this sect i on i s based on t he m ost r ecen t
hardwar e t o host t he V PS t o d a te, w i th h ardwar e specific atio ns that a r e
summarized in Table 3 below. It is important to note that thro ughout t h e
exper i ence o f upgradin g and r e fur b ishing t he V PS, the c o mbinati on o f various
computer h ardwar e co mponents t hat ar e com b ined i nto a syst em c a n sometimes
int e ract i n unpredicta ble ways d ependin g o n firmwa re v ersion s, dr iver v ers i on s,
and ev en f luctuations in t he sy stem temperatur e. T herefore, re al‐world
benchmarks are arguably more important than simple (and often o verly
optimistic) system specifications.
48

Ta ble 3. S yst e m speci f ica t i o ns o f t h e Visua l P rost hesis Simul a t or P C use d i n benchmar king
experiment s.
Parameter Value
Opera t ing Syst em Window s 7 Ul t i mat e 32‐bit , Service Pack 1
Motherboar d ASUS P5 N ‐D
Chipset NVIDIA nF o rce 750i SL I
Proce ss or Int e l Core 2 D uo, E8400 a t 3 .00 GHz (Dua l Core)
Memor y 4096 MB
DirectX Versi o n 11
Graphi c s C ar d NVIDIA Ge F or ce GTX 670
GPU Clock 980 MHz
Video Mem o r y 3273 MB, 192 .2 GB/sec
CUDA Cores 1344
Video D r iver V ersion 8.17.1 3 .1 42 (5/15/2012)
Long‐term benchmarks w ere run on t his syst em i n a variety of si mulation
configurations to compute an average execution time for all of the key
com pon ents of th e VPS code. Th e code segm e n ts tes ted, a l ong wi th descript i ons
and ex ecution times, a re g iv en i n Table 4. These benc hmarks w ere run once
ever y seco nd, and sta t istics w er e collected o ver a per i od o f 10 minutes on two
differ ent d a ys. T he f irs t c ode seg m ent consists o f a set of p r e‐processin g s t e p s
that include the capture of the video frame, conversion of colo r and image
formats, d ewarping a nd o ther l ens corr ections, a n d f ormatt ing o f the
experimenter’s display. Next are a set of three core image pro cessing step s
including q u antization ( pixellation), the GPU CUDA blurring alg orith m , and other
modelling of the elec tro d es such as fill factor and electrode s hape.
49

Ta ble 4. Compa rison of minimum, avera g e, a nd m a x imum p roce ssing t imes i n milliseconds f or
various c ode s egments o f t h e visual pr os t h esis simula t o r .
Code Segment Description Min Average Max
Pre‐Proces s ing Cap t ure o f im a ge f ro m vide o input s
including any color c onversi o n or
dewarpin g; drawing of visual
overla ys for e xperiment e r view
14 ms 18 ms 27 ms
Quanti z a ti on Pi xe llati on o f sub ‐ re gi on o f i n te re st
a n d g ridding bet ween pixels to
simulate a given fill‐factor
0.22 ms 0.30 ms 2.8 ms
GPU Blurring Blurring of th e re gi on o f i n t e re st
using a parallelized CUDA GPU
blurring algor ithm by me an s o f
Fourie r tran sf orms
7.4 ms 9.9 ms 15 ms
Proce ssing To tal Tota l ima g e processing t i me,
i n cludi n g qu a n ti za ti on a nd b lurri ng,
but also inclu ding s ub‐regioning and
other minor i m a g e pr oces s ing ta sk s
22 ms 27 ms 33 ms
Post ‐Proce ssing 1 Addi tion o f regi on o f simula ted
prost h et ic visi o n in t he l ocation of
current g a z e c oor dina tes
13 ms 16 ms 22 ms
Post ‐Proce ssing 2 Dis pla y of f inal i ma g e o n t he H MD 7.2 ms 8.7 ms 26 ms
Grand Total Total of all ste p s, includin g p re‐
proce ssing, p r oces s ing to t a l, and
b o th p o s t‐ pr o c e s si ng s te ps
56 ms 70 ms 110 ms
F r ames Per
Secon d
Effe c t i v e fr am e rate c al cula t e d a s t he
inverse o f t he g ra nd t ota l pro c e ssing
ti me i n se c o n d s
18 14 9.1
L a st are a s et of two p o st‐ p roce ssi ng segm e nt s th at cre at e th e final image
b a sed on th e cam era m ode ( h e ad‐ poi nted or eye‐ point e d ) and act u al l y write th e
final image to the HMD. The benchmarking results indicate that with a full
processing l oad (all possible si mulation a lgorithms running), t he a v e rage f rame
rate a tta in ed w as a pproximately 1 4 f rames per second, which is above the 10 to
12 fr ames p er second need ed f or f licker‐fusion of t he v ideo str eam. T herefore,
the system i s deemed a s qualifying for “real‐time” for this applicatio n ( that o f
visual prosthesis simulation).
50

It i s also u seful and necessary t o consider s ystem latency when gauging
performanc e, t hat is, the total delay between the time t hat the scene or gaze
changes to the time that the change is reflected in the simulat ion. This l at enc y
becomes difficult t o m e asure without highly specialized equipme nt, so t h e s e
values w ere computed b y analys is. F or t his s y s t em, the scene c amera v i deo frame
can change due to the motion of an object in the scene, or due to t he m ovement
of the camera itself (i.e., a change in head position). These types of changes are
captured b y the scen e camera, which operates a t 30 frames p er second. The time
between when a g iven c hange occurs i n a scene or b y camera m ove ment, and
when this change is first displayed on the HMD will be referred to as “scene‐to‐
display latency”.
If a digital camera with a global shutter is used for the scene c amera and
eye camer a ( as i s the case w ith this v isual prosthesis s im ulato r design) then t he
image of t h e e nv ironm e nt i s captu r ed w ithin a small fraction o f the frame time,
following b y a period o f data t ransfer to t he c omputer, a nd t he n followed by a
period o f w a it t ime unt i l the nex t f rame c aptu re b eg ins. I n this case, it can be
assumed that t he l owes t scen e‐to‐display l aten cy c an b e ac hieved when the scene
or head position changes just before the scene camera captures a new image.
Conversely, the highes t scen e‐to‐display l at ency w ill be m easur ed when the
scene or h ead positio n c hanges j ust aft e r the s c ene camer a c aptures a new image,
as this change will not be detected until at most 33 milliseconds later when a new
video ima g e is captured by the s cene camera in the following fr ame.
51

The second t ype of l a t ency i n this system is e ncounter ed w hen t he e ye
position changes and the gaze coordinates should therefore be u pdated b y the
eye track e r . Eye p osition chan g e s ar e capt ured b y the ey e cam e ra, which
operate s at 60 fram e s per secon d ( t wice th e ra t e of th e sce ne c amera). The time
between w h en t he e y e p osition changes, a nd w hen this c hange is first displayed
on the HMD will be referred to as “gaze‐to‐display latency”. T he l ow est gaze‐to‐
display lat e ncy can be a chieved w hen the eye position ( g a ze d ir ect i o n ) changes
just before the eye c a mera captur e s a new image.
F i g ure 20. Swim la ne d iag ram sho wing t he sequenc e o f event s t ha t produ ces the l o west p os sible
syst em l at ency. The e xt erna l event s h appen just b efore (1) the c ame ras c ap ture i ma ge s, and (2 )
t h e correct g aze coordina t e s a re ca lcul a t ed a nd m a d e a v a i la ble just b ef ore t h ey a re n eeded t o
e x tra c t t h e re gi on o f i n te re s t .

52

Similar to the analysis of the scene camera latencies but now r ef erring t o
the eye c a mera, th e h i ghest gaz e ‐to‐display latency will b e mea sured when t h e
eye posit i o n c hanges j ust after the eye camera c aptures a new i mage, as t his
change w ill not be d etected unti l at m ost 17 milliseconds later , when a n ew v ideo
image is captured by the scene camera. An example of both best ‐c ase latency
scenar ios happening s imultaneousl y is s hown s chematical ly i n Figure 20. In this
case, the ex tern al e ven t s happen just b efore the corresponding vi de o f ra m e s a re
captured by the camera s.
A comprehensiv e Ex cel spread sheet w as c rea t ed t o track la tency
calculations for the best and worst cases of each. Figure 21 s hows a n excerpt
from t his sp readsheet showing external e vents and the su bsystem s that m ake use
of these events in the processing chain. Each lane in the diag r a m repres ents
either an external event that can happen, a processing thread r unning o n a
computer, or t he d ispla y d ev ice its e lf i n the ca se o f the las t la n e . Le tte rs in box e s
stemming f rom each segment i nd icate th e outputs (below t he seg m ents) and
inputs ( above the segments) that can be given to, or accepted f rom other
processing l anes. F or e xample, t h e calculated g az e coordina tes (box B) are
need ed dur ing the pr e‐processing stag e i n the V P S soft ware, and so the best
overall latency can be a chieved w hen the correct g az e coordina tes are available
just before the pre‐processing s eg ment.

53

Some p roce ssing l anes d epend synchronously on t he p revious lane and
these dependenc ies a re s hown w ith a solid line c on necting the lanes. For
example, t he V iew Point ey e t r a c king soft w are w a it s until the ey e camer a
produces a v alid i mage, so t he V iew P oint t hrea d is e ffect i v e ly frozen in time until
the eye c a mera process is complete.
F i g u re 21. Excerpt o f t h e Excel file u se d t o c alcul a te w ors t c ase and best case latency scenarios.
Each pair of letters represents an output from a previous step that can be used as an input in a
subsequent processing st ep.
The computatio nal lat e ncy descr i bed at t he b egin ning o f this se ction that
was summa rized in Tab le 4 is repr esented by the “VPS Software” process lane i n
Figure 21. The time delay of this line is entirely limited by the speed o f the CPU
and GPU in the VPS. On this timing chart, each square represen ts o ne m illisecond
in time. Considering only the ordering in time for the various i nputs and outputs
on e ach processing l ane can help t o determine the overal l laten cy o f the system.
54

For example, t he s cen e ‐to‐display latency can be g iv en b y the t otal time between
point C of the “Scene Change” external event and point E of the display sub‐
system, at which point the subject can see the result of the en tire p rocessing
stream.
Monte Carl o simulations were r un t o place each a synchronous output in
various del a ys w ith respect to t he c orresponding i nputs in s ubsequent lanes. In
some scenarios, each output can be provided just before the inp ut i s needed t o
result in the lowest value for each type of latency. In other s c enar ios, e a c h output
can be p ro vided just a fter a n input is n eeded, i n which case o ne t i me p eriod must
be advanced before the output va lue can be used. The results of this analysis fo r
each type o f system latency are summarized in Table 5.
Ta ble 5. Compa rison of minimum, avera g e, a nd m a x imum p roce ssing t imes i n milliseconds f or
t h ree diff erent syst em lat encies.
Latency Min Average Max
Scene‐t o ‐displa y 73 ms 96 ms 118 ms
Gaze‐t o‐displ a y 56 ms 88 ms 120 ms
Compu t ation a l 56 ms 70 ms 110 ms
The effects of i ncr e as ed c omputa tional l oad (effectively l oweri ng t he
frame ra te) and increa sed scene‐t o ‐display a nd g aze‐to‐display latencies c an b e
investig at ed i n future studies. I n fact, the VPS was designed to a llow for
addit i onal l atencies t o be a rtificially a dded by t he e xper iment er. H igh
computational latenc y creat e s a very c hoppy v ideo s tr eam , w hile h igh scene‐to‐
di spla y la te n cy ca use s a n ef fe ct whe re the subje ct f e e ls like h e or she i s wa tchin g
55

a video o n delay. S ubjects using the system i n its most e ffici ent state did not
report any difficulty with using the system, and qualitative data shows that the
70 to 100 millisecond s of scene‐to‐display latency is wel l tolerated.
2.6 Visual Prosthesis Simulator as an Eye ‐Tracked Extraocular Camera
A secondary goal o f the visual p rosthesis sim u lator is t o possi bly enhance
the experience o f pat i ents c urr e ntly u sing a n ex traoc u lar camer a by a dding
foveation capability either before the intraocular camera becom es a vailable o r
for cases in which the crystalline sac may be compromised and i mplantat ion of
an intraocular camera may be infeasible. This can be accomplis hed by
redesigning the visual p rosthesi s simulator (VPS) into a s mall and portabl e
battery o perated system, and addi ng a n ultra m inia ture e ye‐track ing camera t o
the ext r aoc u lar camera m ounting (typically a p air of s unglasses). Instead of
simulating a nd display ing the r e sults on a h e a d‐mounted display , the VPS gaze‐
compensation c ode w o uld then b e lever a ged to i ns tea d d isplay t h e unalter e d
video on the actual implanted retinal prosthesis through the ex is tin g ( perhaps
slightly modified) vid eo processing unit (VPU).
In fact, we have already demonstrated the basic principles of t his
technique using the cur r ently im plement e d VP S, w ith a subject who has an Argus
I device implanted. Because traditional eye tracking calibrati o n t echniques rely
on the subject looking toward visual indicators on a screen, this t ype of
calibration is i mpossible to p erform w ith a blind subject. I ns tead , the eye
56

tracking c al ibratio n w a s a ccomplished w ith a tactile boar d with evenly spaced
felt squares. The tactile board replaced the traditional displ ay c alibration; th e
subject use d a c hin rest f or h ead stabilization and was instruc ted to l ook ( using
foveation only) toward the location of the hand when placed on one of t he f elt
squares. After calibration, a proof of concept demonstration was arranged in
which the subject was asked to l ocate a s m all white tar g et a g a i nst a blac k
background and then point to it. The subject stated that he wa s in f act able t o
move h is e yes to scan the scene, a f ea t nev e r before p erformed using a head‐
pointed visual p rosthesis. The su bject was not only s ignifican tly fas t er a t tar g et
location i n t h e eye‐poin ted mode, b u t also c ommented t ha t this mode w as m uch
more natural and easier to use. A similar study has been prese nte d u si ng t he
Argus II device [21].
In a dd itio n to difficulties i nherent with e y e t rack ing calibrat io n, a
wear able e ye‐track ed v isual prost h esis w ould n eed addit i onal i mpro v ements t o
the eye tr a c king syst e m employed i n order to b e viable u nder v a ried l ight in g
conditio ns ( particularly o utdoor s), based on e ye t racking obser vations from
numerous e xperiments w ith the V P S. F irs t , t h e ey e‐trac king c al ibration r outine
should b e very s imple and n e arly a utomatic t o avoid user f rustr at ion and ey e
tracking errors. Further research into eye tracking calibratio n methods is n eeded
for such a d evice to b e r e aliz ed a nd w id ely used. T he a dded ey e camera a lso
poses one a dditio n al c hallenge: the camera s hould be r elativel y unobtrusive on
the user’s h ead, w hich f acilitat es t he n eed f or a v er y sma ll camera. One idea to
57

mitigate t his problem is t o simply use a modified version of th e intraocular
camera a s a n u ltraminiature eye t r acking c am era. F in ally, use of the eye‐tracking
system b ased o n the extract i on o f the pupil centro id f ro m a vid eo s tream will
require a w i de d ynamic r ang e c a m era for op erat ion in b oth indoo r and outdoor
lighting conditions, as well as careful design to minimize spur ious r eflections t h a t
could frustrate th e cent roid ex t ract ion algorithm.
At first glance, the latency analysis in the previous section may c ause
concern because real‐t ime operat ion is j ust achieved, even w ith a powe rf u l du a l‐
core C P U a nd m odern CUDA‐en a bled G PU. H owever, the image proce ssing
requiremen ts f or a n eye‐track e d ext r aoc u lar camera ( EOC) w ould be
considerably relaxed compared with the full visual prosthesis s imulation. F or
example, there would be no need for the scene camera or at leas t i m age
processing softwar e to h andle col o r images, which reduces the c o m plexity of
sever a l ima g e process i ng st e ps. I n addit i on, there is n o need for an e x p erimen ter
display, and no need for GPU blurring, which is present in the VPS only t o
simulate e lectric field and curren t i nhomogen eit ies a nd spreadi ng b etween t he
implanted electrodes a nd the r etin al tissue.
Table 6 summarizes s o me o f the key differen ces in r equirement s bet ween
the VPS and an eye‐tracked EOC. The primary savings on computing power is
expected to be the lack of needs for both an experimenter display and the GPU
post‐simula t ion blurring s tep. I n additio n , the output r esolution of the processed
image is s ig nifican t ly s maller, a nd o nly 8‐bit monochrome p roce ssing i s needed.
58

Due to t hese differ e n c es, and th e fact t ha t minia t urized F PGAs and DSPs a r e
becoming competitive even with desktop CPUs, the argument can be made t hat
real‐time p r ocessin g of an ey e ‐trac k ed EOC is e nt irely fea sible .
Ta ble 6. Compa rison of s imulat io n requirement s f or t hree diff e r e n t ap pli c a t i o ns o f the vi s u al
prost h esis s imula t or, a n d t h e rela xe d requirement s f or a n eye‐t ra ck ed e xt ra oc ula r c amera
system.
Simulation Requirements
VPS
AMD Mode
VPS
RP Mode
Eye ‐Tracked
EOC
Scene Camera Reg io n of Interest 800 600 256 256 256 256
Color / M o nochrome Color Monochrome Monochrome
Required Bit ‐Dept h 24 bit s 8 bit s 8 bit s
Experimenter D ispla y Yes Yes No
GPU Blur Yes Yes No
Eye Tracker C a libration Visual o n HM D Visual o n HM D Ta ct ile or Ot h er
Video Ou tpu t R esoluti o n SXGA SXGA Comp osi t e
To f urther e xplore t he d igit al s ig n a l processing r equir e ments o f a n e y e ‐
tracked EOC, an effort was funded by Texas Instruments (Dallas, TX) to determine
the optimal pixel bit‐depth for pr ocessing o f images i n both i n door a nd o utdoor
environments. In this study, the goal was the determine the op tim a l floating‐
point repr esent a tio n o f a wide‐dy n amic r ang e p ixel v alue g iven that any real‐
world scene may have u p to 7 o rders of m agnitude difference in il lumination
levels. Although optical intensity in a physics context is typ ically c onver t ed t o
decibels using 10 times the logarithm, it is common practice in the digital image
sensor i ndu stry to define th e dyn a mic rang e o f t he s ensor in d e cibels a s 20 times
the logarithm of the optical intensity because the voltage outp ut o f the image
sensor i s ty pically l inear with r es pect t o the incid e nt i nt ensi ty, a n d voltage is a
59

field quant i ty. T her e fo re, the definition o f decibel as p resen ted in t his thesis w ill
be assumed to be 20 times the logarithm, and so 7 orders of mag nitude i n
intensity w o uld requir e an i mage sensor with 140 dB o f dynamic range . T he key
assumption i n this s tu dy w as t hat the focus should b e entir e ly on d igit al sign a l
processing o f the imag e, a nd t hat a camer a w ill be a va ilable i n t he f uture to
capture the full 140 dB of dynamic rang e.
One add i t i onal k ey a ss umption is t hat the pr ocessor should b e a s small
and as l ow p ower a s possible for use in a p p lications such a s th e intr aocular
camera. Given these assumptions as well as a survey of the sma llest p rocessors
currently a vailable o n the market, the IEEE 754‐2008 stand a rd 1 6‐bit (half‐
precision) f l o atin g poin t architectu re w as c hosen a s t he f r a mew ork for this s tudy.
This f orma t consists o f three fi elds: a s ingle sign b it, ten si gn if i ca n d bi t s
repres ent i n g a b ase‐10 n umber, a n d f iv e exponen t b its that r epres en t a bas e ‐10
exponent number. Key features o f this standard are the use of a sign bit in order
to r epres e n t p ositiv e a n d neg a tive n umbers, a fix e d offs et t hat i s subtracted f rom
the exponent i n order to r epres e n t p ositiv e a n d nega tiv e e xponents, and special
representations such as infini ty and NaN (not a number).
While this s tanda r d sp ecifies t he n umber of e xponent bits t o be f ive, t his
study relax e s the number o f expo nen t b its to b e used w hile f oll owing all other
specific atio ns o f the s t andar d. T h e r eason fo r the r e laxa tion is t o p r opose the
optimal number o f ex ponent b its to b e used f or d igit al i mage p r oc essing, but
while also c onforming to t he I EE E 754‐2008 standard i n ord e r to l evera g e
60

exis tin g h ardwar e arc h itectures that h ave already been wid ely t est e d and
implement e d.
In a dd it ion to a v ar iable number o f exponen t b its, t he c as es o f fractional
and int e ger sign ificand s w er e explored. F or t he c ase o f f racti onal s ignificands, a
hidden bit ( w hich a lways has the value of 1 ) is a ssumed to p rec ede th e radix point
such that the minimum value of the significand approaches 1 and the maximum
value appr oaches 2 . U nder t his assumption c oupled w ith the exp onent offset
given by t he I EEE 754‐2008 s t and a rd, it c an b e s h own that t he m aximum p ossible
value that c an b e r e pr esen ted as a f unct ion of t he n umber of e x po nen t b its is
given by Equation 3, i n which b i s the nu mber o f ex ponent b its used i n th e
repres ent a t i on.
 
22

max
22
 

b
offset
Value
(3)
The subtraction of 2 in the exponent is due to the fact that th e la rgest
possible exponent value is given by 2 raised to the number of e xponent bits m inu s
1, and that by IEEE convention, t he largest value is reserved f or special meaning,
so a nother 1 m ust be s ubtracted. F urthermore, by I EEE 754‐2008 convention,
the offset value is given by Equation 4, which is half of the l arg e st e x p onent value
in order to allow for the larges t sw ing of n eg at ive a n d positiv e expon e nt value.
1
21

 
b
offset
(4)
61

The maxim u m possible 16‐bit flo a tin g p oint v alue c an b e expr ess ed b y
combining E quation 3 with E quation 4 to p roduce E quation 5 bel ow. One
interesting note is that this equation holds for any number of bits a nd i s only
governed by the number of exponent bits. The reason for this is in the choice of
the hidden one so that all of the bits in the significand allow for additional
precision (which is also a desired property), but do not affect the largest possible
value.
 
1
22

ma x
22



b
Value
(5)
For example, i f the nu mber o f ex ponent b it s is c hosen to b e fiv e, t hen the
largest exp o nent v alue i s 11110 (11111 is res erved to r epresent i n f inity), which
is a base 10 equivalent value of 30. The offset in this exampl e would be 15, s o the
maximum exponent value is also 15. Finally, 2 2
15
yi e lds the maximum value
of 65,536. In order to determine the ratio between the m aximum and minimum
values (i.e. dynamic range), a similar equation is needed for the minimum value
for a 16‐bit floating point number and is given by Equation 6. The difference in
this case is that the minimum value of the significand sh ould approach 1 (due to
the hidden leading 1) and the smallest exponent value is one (z ero is r eserv e d for
a special meaning d i sc ussed next) minus the same offset.
1
22

min
1.0 2

 

b
Value
(6)
For example, if the number of exponent bits is again chosen to be five, then
the smallest e xponent value is 0 0001 (00000 is r eserved for spe cial m eaning),
62

which is a base 10 equivalent value of 1. The offset in this e xample i s again 15, so
the minimum exponent value is –14. Finally, 1 2
–14
yi e lds the mi ni mum value
of 0.000061. F or t his example, t he d ynamic r ange i s approximately 9 decades, or
180 dB. A s mentioned above, t here i s a spec ial meaning given t o th e number
when all of the exponent bits are zero. This representation is known as
“subnormal” and can be used when values closer to zero are needed than
typically afforded by the normal minimum value. In this case, the hidden 1
changes to a hidden 0 before the radix point, and the minimum s ignificand i s no w
govern ed b y the number o f bits i n the sign ific and (b less than 16 in this case).
The minim u m value that can be re presen ted by a 16‐bit floatin g point number is
given by Equation 7.
1
(16 ) 2 1
su b m i n
22

  


b
b
Value
(7)
Because ea ch o f these equations express the mi ni mum and maxi mu m
possible values o nly in t erms o f the number o f expon e nt b its, F igure 22 shows a
plot o f Equations 5, 6, and 7 converted t o dB o n the vertical a xis and plotted as a
function o f number o f exponent b its, b. The subnormal curve is shown as a
dashed line, and while it is possible to use this representation, there are
processing a nd a lgorithmic conce rns with h andling d i fferent r eg imes o f the
hidden bit. W hen looking at t he difference between the maximum and minimum
values in decibels (equivalent to looking at the ratio of the a ctual values), n ote
that f o r a s election o f 4 expon e nt b its, ther e is a pproxim ately 90 dB o f dynamic
63

range, while for a selection of 5 exponent bits, there is about 180 dB o f dynamic
rang e, a s c a lculated i n t h e example used i n this s ectio n . T her efor e, t o repres en t
the expected scene dynamic range of 140 dB, 5 exponent bits is t h e preferr e d
choice. T hankfully t h i s choice c oincides p erfectly w ith the IE EE 754‐2008
stand a rd, which defines 5 exp o nent b its for half‐precisio n f l o a tin g p oint
numbers.
Figure 22. P lots o f the range of s ub‐normal minimu m, n orm a l mi nimum, a n d ma xi mum va lues i n
dB f or f lo ating point nu mbers with f ra cti o n a l signifi c an ds a s a f unc t ion of t he n umber of e x p onen t
bit s .
A similar a n alysis w as completed f or a nother r epresen t a t ion o f a 16 ‐bit
number, but inst ead u s ing a n i n t eger s ign i ficand t hat departs f ro m the IEEE
standard of a fractional significand. Similar results were obt ai ned i n t hat 5
exponent b its are needed t o reach at l east 140 dB d ifferenc e be tween the
mi ni mum and maxi mum possi bl e value. I t should also be noted t h at i n
64

applications where the intensity value of illumination is repre sented w ith
floating p oint n umbers , negativ e v alues are no t nec e ssar i ly n ee ded. T h erefor e, i t
is suggested that the sign bit be used for additional precision o f the significan t if
spec i ali zed hardware i s developed f o r posi ti ve ‐onl y f l oati ng p o int
repres ent a t i ons.
2.7 Summary and Other Applications
The visual p rosthesis simulator will be featured again in the n ext chapter,
and has proven to be an invaluable tool for understanding the e ffects o f fove atio n
with a retinal prosthesis. A unique mix of hardware and software has enabled
this technology to run in real time, and may even enable additi onal t reatment
options for RP and AMD by way of an eye‐tracked EOC before an I OC b ecomes
commercially a vailable, or w here s urgical constraints m ay p recl ude th e use o f a n
IOC. Eve n w ith thes e prim ary applications, num e rous o ther a ppl ications f or a
visual prosthesis simulator have b een i dentified.
Similar to t he e ffort to c reat e a miniatur ized versio n o f t h e V PS t o be u sed
as an eye‐tracked EOC with relaxed requirements on latency and computational
load, it m ay b e possible to c rea t e a minia t urized v ers i on o f th e VPS that i s fully
functional a s a more general‐purpose visio n simulator to b e used in furth er
studies. F o r e xample, longer‐term studies in o phthalmology o r vision sciences
may be enabled if the VPS is fully portable and wearable. The current g en er atio n
VPS has already been demonstrated to be very mobile with a wireless video link ;
65

however, the subject still needs to be within range of the larg e base‐stati on P C
that actually runs the simulation code.
An effort has been made to explore the idea of miniaturization of the entire
p ro c e ssing c h ain o f th e VPS , inc l u d i ng th e b l u r tec h niq u e c u rre ntly e mployed on
th e CUDA‐enabled GPU . For this test, a LogicPD (Minneap olis, M N) d evelopment
kit featuring a Texas Instrument s OMAP p ro cessor that r uns Windows CE was
used to test various blur benchmarks that could be applied to t he visual
prosthesis simulator. The open source Cimg image processing library was used
for these tests, and b e nchmarks w ere tak e n for a Ga ussian con vo lution, an
anisotropic filtering technique, and a simple moving average blur. A comparison
of the image output res ults are shown in Figure 23. The k ernel size, filter siz e, or
correspond i ng a djustm ent par a meters w ere v a ried i n eac h c ase un til the amount
of blur was visually equivalent.
F i g u re 23. An ima g e t h at h a s b een b lurred using the OMAP p roces sor an d Cimg l ibrary b y
means o f G au s s ian conv o luti on ( Lef t ), a nis o tr opic f il tering ( Ce nt er), a nd a moving avera g e b o x
fi lte r ( Ri gh t).

66

As expected, the Gaussian convolution filter processing time depends
strongly on the size of the convolution kernel, with a quadrati cally i ncreasing
processing t ime as a function of k ernel size. A s was employed in the visual
prosthesis s im ulator, a Fourier transform based Gaussian b lur co u ld also be u sed
such that the processing time does not increase significantly w ith kernel s ize;
however, such an algorithm was not available for the OMAP processor. Both
anisot ropic filter ing a n d moving a ver a ges a r e constan t with r es pect t o kernel
size; however, none of these techniques were able to achieve a real‐time
processing rate greater than 16 Hz (processing times of under 6 0 milliseconds).
These ben c hmarking r esults a r e s ummarized in F ig ure 24 (n ote th at t h e
processing or execution time is on a logarithmic scale). Althoug h real t ime
operation was not achieved in this test, miniature digital sign al p rocessors a re
continually improving. Because the moving average filter produ ces a reasonable
output w ith only 90 milliseconds o f processing t im e, i t is h ope ful that f utur e
generations of DSPs will easily b e able to accomplish this in r eal time.
67

F i g ure 24. P lots o f exe c ut ion t i me a s a funct i on of k e rnel o r f ilter size for t hree d i fferent
blurring a lgo r it hms using t h e Ci mg library running on a LogicPD O MA P de ve lop m e nt bo ard.
In a dd it ion to a minia turized versio n of t he V PS , several oth e r a pplications
have b een i d ent i f i ed, i n cluding th e study of e ye d iseases by op hthalmologists.
For example, a n ongo ing invest ig ation using the VPS is e nabling research into
treatment for Leber’s hereditar y optic neuropathy (LHON). Othe r studies o n
saliency cues, a s well as a lterna te f orms of visual prostheses such a s the “vOICe”
sensory su bstitution d evice that c onver t s vis u al s ignals t o aud itory signals, a re
being purs ued. For a ny s tudies i n which a real‐tim e alter e d v i ew of the
environment that can be stabilized on the retina is desired, the visual prosthesis
simulator described herein provides a comprehensive solution.

68

CHAPTER 2 REFERENCES
[1] J. S ommerhalder, E . Ou eghlani, M . Bagnoud, U . Leonards, A. B. Safran, and
M. P eliz zone, “Sim ula t ion o f a rt ific ial v i sio n: I. E ccen t ric read ing of
isolated w o r ds, and pe rceptual l ea rnin g,” Vision Research , vol. 43, n o. 3,
pp. 269–283, 2003.
[2] J. S . Hayes, V. T . Y i n, D . Piyathaisere, J. D . Weiland, M . S. H umayun, and
G. D agnelie, “ Visually guided pe rfo r mance of s imple tasks using simulated
prosthetic v ision,” Artificial Organs , vol. 27, no. 11, pp. 1016–1028, 2003.
[3] D. K . Freeman, J. F . Rizzo III, a nd S . I. F ried, “Encoding visual i nformation
in r e t in al ganglion cell s with p rosthetic st im ulation,” Journal of Neural
Engineering , vol. 8, no. 3, p. 035005, 2011.
[4] J. J . van Rheede, J . Joram, C . Kennard, a nd S . L. H icks, “S im ulating pros thetic
vision: Optimizing t he i nformati on content o f a limited visual display,”
Journal of Vision , vol. 10, no. 14, p. 32, 2010.
[5] R. W . Thompson, G. D . Barnett, M . S. H uma yun, and G. D agnel ie, “Facial
Recognition U sing S im ulated P ro sthetic Pixel i zed Vision,” Investigative
Ophthalmology and Visual Science , vol. 44, no. 11, pp. 5035–5042, 2003.
[6] J. Sommerhalder, B. Rappaz, R. de Haller, A. P. Fornos, A. B. Safran, and
M. P elizzon e , “Sim ulation of a rtific ial visio n: I I . Eccentric reading o f full‐
page t ext and the lear ning o f this t ask,” Vision Research, vol. 44, no. 14, pp.
1693–1706, 2004.
[7] A. P . Fornos, J. S omme rhalder, B . Rappaz, A. B . Safran, and M. Pelizzone,
“Simulation of a rtific ial vis i on, III: d o the spatial or t emp o ral
character i stics of s timulus pixelizatio n r eally m atter ? ” Investigative
Ophthalmology and Visual Science , vol. 46, no. 10, pp. 3906–3912, 2005.
[8] K. C ha, K. H orch, and R. A . Normann, “ S i mulation o f a p h osp hene‐based
visual f ield: visual a c u ity in a p ixeliz ed v ision sys t e m ,” Annals of
Biomedical Engineering , vol. 20, no. 4, pp. 439–449, 1992.
[9] K. Cha, K. W. Horch, and R. A. Normann, “Mobility performan ce w ith a
pixeliz e d vision s ystem,” Vision Research , vol. 32, n o. 7 , pp. 1367–1372,
1992.
[10] M . S. H umayun, “Intraocular retinal prosth esis,” Transactions of the
American Ophthalmological Society , vol. 99, pp. 271–300, 2001.
69

[11] G . Dagnelie, D. B arnett, M . S. H umayun, and R. W . Thompson , “Paragraph
text reading using a pixelized prosthetic vision simulator: par a m eter
depend enc e a nd t ask learning i n free‐view i ng conditions,” Investigative
Ophthalmology and Visual Science , vol. 47, no. 3, pp. 1241–1250, 2006.
[12] S . C. C hen, L . E. H allum, G . J. S ua ning, and N. H . Lovell, “ A quantitative
analys is o f head m ovement behaviour during visual acuity a ss ess ment
under prost h etic v is ion sim u lation, ” Journal of Neural Engineering , vol. 4,
no. 1, pp. S108–S123, 2007.
[13] S. C. Chen, L. E. Hallum, N. H. Lovell, and G. J. Suaning, “Visual acuity
measurement of p rosthetic vis i o n: a v i rtual ‐ reality s i mulation study,”
Journal of Neural Engineering , vol. 2, no. 1, pp. S135–S145, 2005.
[14] R . Szeliski, Computer Vision: Algorithms and Applications , Springer‐Verlag
London Limited, Lond on, United Kingdom, 2011.
[15] C . Brauer‐Burchardt a nd K . Voss, “A n ew a lgorithm t o corre ct f ish‐ey e and
strong w ide‐angle‐le ns‐distort io n from s ingle im a g es,” i n 2001
International Conference on Image Processing , vol. 1, pp. 225–228, 2001.
[16] R . W. H ockney, The Science of Computer Benchmarking, vol. 2, the Society
for Industrial a nd A pplied Mathem atics, Philad elphia, Pennsylva nia, 1 996.
[17] J . Dongarra, J. L . Martin, and J. W orlton, “Computer bench marki n g: P aths
and pitfalls: The most p opular w ay o f rating computer performan ce c an
confuse as well as inform; avoid misunderstanding by asking just wha t t he
benchmark is measuring,” IEEE Spectrum, vol. 24, no. 7, p p . 38–43, 1987.
[18] J . D. W eiland, W. L iu, and M. S . Humayun, “ Retinal prosthe sis,” Annual
Review of Biomedical Engineering , vol. 7, pp. 361–401, 2005.
[19] B. P. McIntosh, P. J. Nasia tka, N . R. B . Stiles, J. D . Wei land, M. S . Humayun,
and A. R. Tanguay, Jr., “The Importance of Natural Foveation in Retinal
Prosthes es: Experimen t s with a V isual Prosthesis S imulator,” Proceedings
of the 2010 NIC , Long Beach, Califo rnia, 21‐23 June, 2010.
[20] B . P. M cInto s h, N . R. B . Stiles, M. S . H u mayun, a nd A . R. Tanguay, J r., “Effects
of Foveation on Visual Search Task with Visual Prosthesis Simul ation,”
Journal of Vision , vol. 13, no. 9, p. 685, 2013.
[21] A . Caspi, A . Roy, G. Consed ai, R. Gr eenberg, A. S afran, and J.‐A. Sahel,
“Retin al Pr o sthesis ‐ St eer ing th e Line o f Sight with Eye M ovem ents”, in
36
th
Annual International Conference of the IEEE Engineering in Medicine
and Biology Society , 2014.
70

CHAPTER 3
VISUAL PSYCHOPHYSICS EXPERIMENTS WITH THE
VISUAL PROSTHESIS SIMULATOR
3.1 Introduction
The visual p rosthesis simulator presented in Chapter 2 has numerous
applications i n the res e arch o n and furtheranc e of visual p r ost heses, p a rticularly
intr aocular retinal pros theses. Wit h over fifty individual par ameters that can be
dynamicall y adjusted t o change t he s imulation, a v ery l a rge num ber of s ystem
stat es c an b e achieved a nd s tudied. H owev er, four k ey h ypotheses enabled by
certain com binations of p arameters were chosen to be the focus of initial studies,
given the fact t hat s e veral a r e c o mpletely u nique to t his visua l prosth esis
simulator implementa tion.
1. The comparison of vis ual prostheses u sed by p atien t s with R etin it is
Pigmen tosa a nd A g e ‐Related M acular D egen erat ion is o f particula r
importance, because these are t h e two groups w ith the grea tes t
potential benefit from s uch prostheses, and yet the latter c a se has not
been studied to date.
2. The comparison of e ye‐pointed a nd h ead‐pointed modes for visual
prostheses i s also o f significan t importance d ue t o the fact t h at o nl y
one par a digm ( head‐pointed) is c urrently i n use by commercial
71

devices such a s the Argus II. T he r estor a tio n o f foveation in the eye‐
pointed mode by means of implantable imaging devices such as th e
intraocular camera (IOC) may benefit patients greatly, and the visual
prosthesis simulator can be employed to show to what extent.
3. Electrode count and density are two additional parameters of ut most
importance, as p rostheses today p r ovide up t o 60 electrodes, bu t the
next few generations may increase the number to as many as 1,00 0.
Understanding the possible benefits afforded by increasing the
number o f electrodes m ay i n f luen ce f uture research d irections a s w e ll
as their corresponding techn ological implementations.
4. Lastly, the level of b lur in i ntraocular retinal prostheses s ig nific a ntl y
affects the perceptio n o f pixellated images, b u t is n ot o ften s tudied o r
simulated to t he e xtent that c an b e accomplished by t he visual
prosthesis simulator in real time. The results of studying thi s
paramet e r may influence how the electrodes a re d esig ned, a nd h o w
they rest on the surface of the retina.
In o rder t o demonst r ate the effectiveness (or ineffectiveness) of these key
parameters, a set of experiments was envisioned to capture a wide range of daily
living tasks and fundamental visual system properties that may affect p atients
with implanted devices. The experiments described in this chap ter w e re i n some
72

cases derived in part from succe ssful publishe d experiments in visual prosthesis
simulation, and in part from ongoing evaluations of subjects im planted with
retinal pro s theses. A s ummary o f the key published studies to date will be
presented in the next section, followed by the results of three key new
experiments designed a nd conduc ted specifically to address th e four h ypotheses
above and thereby to t est the be nefits o f fo veat ion usi n g the v isual prosthesis
simulator for both RP and AMD cases.
3.2 Literature Review of Human Subject Testing
Even b efor e the initial exper i ments o n e l i citing t he p erceptio n of
phosphenes by elect r ic al stimula tion of the r e tina were p e rform ed in 1996 [1], a
series of papers was published that described a set of human su bject
psychophys ics experim e nts using an e arly visu a l simulator consisting of a head‐
mounted CRT monitor with a p erforated scr e en p laced between the m onitor a nd
the observer to simulate the effect of percepts generated by an a rr ay o f small
discrete e lectrodes [2–4]. T his si mulator sought t o mi mi c the phosphenes
observed i n experiments with cor tical visual p rostheses that i n m any ways a re
similar to retinal prostheses, in that an image can be formed b y separated and
discrete regions of electrical stimulation. These papers inclu ded performance
measurements of human subjects participating in visual psychoph ysics tasks,
including viewin g a let t e r “E” at va r ious siz es t o det e rm in e v i sual acuity, reading
simple s entences, and nav i gating t hrough a n indoor o bstacle cou rse to t es t
73

mobility. I n each c ase, p erformance was com pared among various resolutions
ranging from a 10 10 array to a 32 32 array, w ith va ryin g electr ode spacin g
and coverage of t he visual field. A f undamental conclusion was that
approximately 625 e lec t rodes (25 25) a re n eeded w ithin the macul a r r e gio n t o
allow for a 20/30 v i sual a cuity and to p erfor m a variety o f tas ks reasonably well.
This r einfo r ces a result t hat was suggested a s early as 1965 th at 600 visual
channels a re n eeded fo r efficient reading [5]. Aspects of t hes e studies, such a s
the concept of a stabilized percept using eye tracking, and per formance m etrics
such as the speed to complete a ta sk and the quality of c omplet ion by measuring
errors o r o b stacle t ouches, has been c arried f orward i nto a number of the new
exper i ment s presen ted in this thes is.
In anticipation of the initial human subject trials of the Argu s I device,
which began in 2002, a series of studies specifically targeted toward t he
simulation o f in traocular r e tina l p r ostheses a nd s ubject p erformance with a set
of d aily l iving tasks w a s published [6]. In this p aper, a set of p sychophysical tests
of tasks regarded as essential to function in daily living were a dm inist e re d,
including facial r ecognition, le tter/symbol r ecognition, visual acuity
measurements using a tumbling “E” test, object recognition, and a set of manual
dexterity tasks. T he p reviously published result o f a 25 25 a rra y wa s ca rri e d
forward as a baseline and variab le electrode size was added to th e simulation, in
addition t o simulated electrode dropouts t hat were b ased o n pro jected e lectrode
fabrication yields of 70 to 90 pe rcent. Th e results of this st udy again established
74

that p erfo r m ance i s maximized in g en er al w hen the electrode arr ay h as a
resolution o f at l east 2 5 25 elements, a n d further con cluded t hat 6 grey l evel s
and a 33% fill factor ( ratio of electrode size to gap size) are sufficient for good
speed an d a ccuracy in t h e execut io n of the dail y living tasks.
In 2003, i nitial r esults e merged f rom the first implanted Argus I subjects,
showing that they were able to r eport the location o f individua l phosphenes
generated by electrode stimulation, could distinguish between t wo a djacent
percepts, and could c orrectly i de nti f y hori zontal and ve r ti cal sets of moving bar s
[7]. During the same year, a set of visual psychophysics papers was published for
new visual p rosthesis simulation s and experimental p rotocols fe atur ing
simulations that w ere more p orta ble and acc u rate t ha n previous simulations
[8–10]. A s before, these visual p sychophysics studies f ocused primar ily on f acial
recognition, reading, and daily living tasks such as object recognition, eye‐hand
coordination, and cutting a ccuracy . A s with p revious experimen ts, variables such
as array resolution and fill factor were tested, but with an em phasis o n the
comparison of the Argus I device (in human subject trials at th e t i me) to t he
planned Argus II that was then under development. Several pape rs f ocused on
contrast o p timization [ 9–10], whil e another paper [8] began exp loring
performanc e differences b etw e en m onocular and binocular viewing during
read ing tas ks with a visual prosthesis. T he c o n clusion in t his s tudy w as t hat with
a binocular system, subjects a ch ieved faster l earning and a hig her overall
performance as compared with a m onocular system. Y et i ntraocul ar r etinal
75

prostheses a re c urrent ly i mplanted i n only a s ingle eye, a nd i t i s no t yet k n ow n
whether or n ot b inocular fusion c an b e achieved w ith binocular re t i nal
prostheses. Aspects of both monocular and binocular perception of depth are
explained in Section 3.4, and it will be shown herein that the addit i on o f foveat i o n
can improve depth perception a s part o f a m a nual d ex terity t ask , even w ith a
monocular retinal prosthesis.
Results of f urther r es earch publi s hed during t he m id‐2000’s [11 –13]
coincided w ith the announcement t hat the Argus II device w ould begin human
subject trials s hortly t hereafter. In these studies, the focus shifted somewhat to
the study of head movements and observations about how subjects used the
Argus I device for navigation and mobility. It quickly became clear t h at b ecause
the percept provided b y the intrao cular retina l prosthesis i s f ixed w it h respect to
the ret i nal surface, a n d b ecause t he c amera that p rovid e s input to the visual
processing u nit is m ounted t o a set of s unglasses on t h e h ead, the s u bjects a re
required to make sets of unnaturally large as well as unnatural ly small and
controlled head m ovements i n o r der to s can t h e scen e. F urthermore, the subjects
were trained to maintain alignment between the eyes and head in order to
prevent mis a lignmen t b etween t he p erceiv ed l ocation o f o b jects in the scene and
their actual e ye g az e dir e ction. T his real izat ion provides s up port f or t he p otential
value of t he i ntraocular camera t o restore natural foveation, a s well as t h e
impetus fo r the dev e l o pment of t he v isual prosthesis s imulator presented in
Chapter 2 in order to simulate the effects of restored foveatio n with a n
76

intr aocular retinal pr osthesis. T he visual prosthesis s im ulato r also a llowed
researchers to e xp erience the h e ad‐pointed c amera m o de u sed by Argus I
subjects for the first time, as well as to share the experience o f the eye‐pointed
camera mode with a subject implanted with the Argus I retinal p rosthesis, also
for the firs t time.
The most recent studies of the results of the Argus II huma n su bject trials
show little to no seriou s advers e effects from chronic implanta tion, and also that
in a vast majority of cases, subjects performed daily living ta sks better w ith th e
prosthesis on as opposed to off [1 4]. Similar prosthesis‐ on ve rsus prosthesis‐off
performanc e results w e re a lso shown by a s et o f human subject e xperiments
performed with an implanted subretinal implant [15]. The Argus II device w as
gran ted the CE m ark for trea tment of R etin itis P igmen t osa (RP) in 2011, a nd w as
approved b y the FDA in 2013. B ecause t his dev i ce i s now commer cially a vailable,
a focus should b e placed o n simulating, understand ing, a nd i mpr oving this a nd
future prostheses, which is one goal of this thesis. One such focus on
improvement pres ented herein is on the effec ts of restored fov eation, which can
be a ccomplished with a n intraocular camera o r an e ye‐tracked e x traocular
camera; both o f which can be a dded to a n existing i mplanted s ub ject t o “upgrade”
the subject’s visual e xp erience to i nclude f oveation. T he f oll owing set of t hree
visual p sychophysics e xperiments f ollows to a ce rtai n e x te nt th e st yle and types
of experiments historically performed for evaluation of such de vices b y
simulation, but seeks to highlight the performance improvements that are
77

afforded by an added foveation ability, and further seeks to describe t he
electrode p a ramet e rs t hat maximize this impr ovement.
3.3 Visual Search
3.3.1 Introduction
Visual s ear c h experim e nts are a well‐established method t o quan tify the
efficiency w ith which a subject ca n locate a t arget object, such as the letter “F”,
among a field of d istractor obje cts, such a s the letter “ E” [ 16 –20]. M any studi es
have shown that the average over many trials of such a task ten ds t o exhibit a
strong l in ea r rel a tio n ship b etw een t he r espons e t i me t o locate the tar g et o bject
and the nu mber of distractor objects present. Typic ally, both target‐present and
targ et‐absent tr ials a r e completed i n order to verify th at t he subject i s i ndeed
locating each target well above chance. A similar visual searc h experiment i s
presen ted here f or s u b jects usin g the visual p rosthesis simulat or, in a n a ttem p t
to q uantify the differ en ce b etween a h ead‐poin ted and a n e ye‐po int e d paradigm
as applied to visual prostheses. To date, no literature has be en p ublished o n the
performance of a visual search u nd er limited f ield of v i ew a nd limited resolution
vision, and under the c o nditions o f RP a nd A MD w hen using a hea d‐poi nted as
opposed to an eye‐pointed camera.
3.3.2 Methods
For this v is ual search e xper iment , b lack l ett e rs w ere p r ojected onto a
white screen t hat spanned a 90 ° h o rizont al f ield o f view f rom the point of v iew o f
78

the subject wear ing the visual p rosthesis simulator. T he h oriz ontal field of v iew
of the scene camera is only 45° and the simulated prosthetic vision area is 20°
h ori zon t a lly, so bot h he a d a n d e ye move m e n t s a re re q ui re d t o cov e r the entire
display area. The subject sits in a darkened room in front of the sc reen w ith a
white background such that the subject is fairly certain when the edge of the
testing area is encountered to prevent head motions that are too large. The
subject was instructed t o look a round the screen w ith the goal of f inding t he
targ et l ett e r “F” among a field of dis tractors c on sisting of du plications o f the letter
“E”. A subject performing the visual search in front of the af orementioned s creen
is shown in Figure 25.

F i g u re 25. S ubject perf o rming t h e vi sua l s e a rch experiment , showing projected letters in the
background and the visual prosthesis simulation and MATLAB soft ware running a nd d isplayed
on a set o f co mputer m onit ors in t he f ore g roun d.
79

Generation o f the random t arget and distractor l ocations, as w e ll as
collection of e xp erimental data, was accomplished with t he u se of a custom
MATLAB script, which was designed to perform a double‐blind exp erimenta l
protocol. In this p rotocol, neither the experimenter n or t he subject knows how
many d istr actors w ill be o n th e s c reen, whether the ta r get w ill b e presen t or
absent, nor where th e t a rget w ill be l ocated. A graphical user interface (GUI) was
designed to facilitate ease of setup, and is shown in Figu re 26. After eye tracking
calibration, the subject pressed a button to start a trial, and then pressed either a
green button (if the tar g et w as f ound), o r a red button (if the t arg e t was absent)
to s top the trial. Both t he t ime to f ind the tar get a nd t he a c curacy o f response
were recorded.
F i g u re 26. Screen c a p t u re o f t he a u tomat e d visua l s ea rch dat a c ollection MAT L AB a p p lica ti on
showing t h e pla c ement s o f lett ers on t he s creen ( L e f t ) as w ell as t he d at a co llec t ed f o r a n
experiment ( Right). T he respon s e times are in units o f secon ds .

80

Subjects w ere recruit e d using posted f lyers ap proved b y the Ins titutional
Review B oard ( IRB) a nd u nder IRB s tudy n umber HS‐12‐00182 at t h e Univers ity
of S outhern Califor n ia H ealth Sciences c ampus. E ntrance criteria for the study
included the ability to speak and understand English so that in structions could
be w ell und e rstood, tol e rance to m otion sickness and claustroph obi a so that t he
HMD could b e well tolerated, a nd a visual acuity ( including cor recti o n) o f 20/40
or better. Before participating in the study, subjects were asked to si gn an
informed consent form approved by the IRB, and their visual acu ity and type o f
vision correction were documented.
Furthermore, a s eries of p re‐exp e r im ent v i su al a cuities were m easured
for each subject while wearing the head‐mounted display in each of the
experimental modes described below, in order to ensure that the letters could be
easily identified on the projection screen. In general, the si ze o f the letters for a
given experimental trial was set to be approximately 1.5 times the visual a cuity
measured for that particular pixellated resolution. For exampl e, i f a s u bject was
able t o successfully r ead the 20/ 600 line o f a randomly generat ed S nellen eye
chart for a display mode o f 24 40 pixellated simulated visio n , then t h e
correspond i ng e xperim ental trials f or t hat resolution u sed a le tter s iz e of 20/ 900.
The primar y advantag e of t h i s technique is t hat all letters ar e guar a nt eed t o be
readable by the subject. One disadvantage of this technique wa s th at l ower
resolutions below 24 40 wer e n ot f eas i ble, a s the lett ers became s o large that a
sufficient number o f distractor o bjects c ould n ot b e displayed in the 90 °
81

horizontal field of view of the sc reen. A vast majority of sub jects reported visu a l
acuities c lose t o 20/300 for a full resolution d isplay, 20/600 for a 24 40 d i splay,
and 20/1200 for a 12 20 display.
The experimental protocol for the visual search task is as foll ows:
1. Measure th e subject’s visual acu ity and verify t hat he o r she m eets
all of the study entranc e criteria.
2. Explain how the experiment works and seat the subject at a pre‐
determined l ocation t o provide a 90° horizontal field of view of the
screen.
3. Allow the s u bject to g o through a dry run o f t he e xperim ent with
and without the HMD until they f eel comfortable with the task.
4. Per f orm the visual s ea rch experiment w ith no H MD a s a baseline
of performance.
5. If the subject reports too many incorrect responses more than
10% incorrect, retrain and restart the experiment.
6. If the subject at any time admits to incorrectly answering or
presses the response b utton too quickly by m istake, m a rk t hat
particular trial as bad data and discard this data during analy sis.
7. Cont inue w ith the pla nned exper i ments, a nd a llow a 5 minute
break if des ired bet ween each one.
8. Ask the subject to c omplete a qu estionnaire to p rovide f eed back on
the experimental met hods and offe r improvements to th e process.
82

Data was first collected using the method described above from 7 naïve
subjects using the full resolution of the display, but with a l im ited h orizontal field
of view of 2 0°, corresp onding a pproximately to t he ( diag onal) field of view of the
Argus II retinal prosthesis. D ata sets f o r b oth head‐pointed a nd e ye ‐poi nte d
modes were collected, and 20 t ri al s were completed f or e ach mod e of o per a tio n .
Data sets were then collected for 9 additional naïve subjects using a simulated
24 40 pix e llated a n d blurred representa tion o f a retinal prosthes is w ith a 20°
horizontal field of view and a 12° vertic a l field of view.
A Gaussian blur o f 60 % that a llowed fo r significant o verlap w i th the
neighboring pixellated regio n s was used i n ord e r to o v e rco m e th e gr id ding e ffect
of the 50% duty cycle, as described in Section 1.5, and the percept was presented
in grayscale. In both sets of experiments, both AMD and RP con ditions w e re
simulated, a s well as h ead‐pointe d and eye‐pointed camera m odes. Lastly,
several sets o f baseline d ata we r e c ollected, inc l uding the s u b ject’s n ormal visual
search performance without the H MD, and the su bject’s visual se arch
performance when using the HMD, but with the full resolution an d the full fiel d
of view of the HMD (60 ° horizontal . For each subject group, the experiment was
repeated for both AMD and RP simulated conditions, for a total of 80 trials p er
subject. One such trial in the AMD condition is shown in Figur e 27. A s a set of
control t rials, both groups (i.e. , full r esolution and 24 40 resolution) were a ls o
asked to perform the same visual search task while wearing the head‐mounted
display using the full resolution o f 1280 720 p ixels and the unrestricted
83

horizontal field of view of 60°, and with normal unrestricted vision. The data sets
from these full resolution and normal vision trials were combin ed i nto a common
data s et to establish a measure of b aseline p e rformance.

Figure 27. E xample o f si mulate d AMD conditi o n with a n intr aocul ar r e t i n al p ro sthe si s duri n g a
visual search experiment. (Top) Image captured by the scene ca mer a s howing t he e xtent of t he
HMD field of view (Yellow Box) and where the subject is currently looking (Red Box). (Bottom)
View displayed on the HMD that the user sees, in this case simu lating A MD w ith a central sc ot om a
sc aled t o an a ppro x ima t ely 28° h orizon ta l field of v iew in o rder to extend beyond the extent o f
the ce nt ral vi s u al p r o s t he si s si mula ti on. N o t e th a t t he i ma ge portion within the red box and the
displa ye d percept a re of fset b y a slig ht a mount , l ik ely due t o the dri f t i n c ali b ra ti on o v e r the le ngth
o f t he e x periment a s t hes e ima g e s were e x t ra ct ed af t er ex perime nt completion.
84

3.3.3 Results
Data was collected from 7 naïve subjects for the full‐resolution
exper i ment , and 9 different na ïve s u bjects f or t he p ixellat e d e xpe r im ent, w ith 4
trials a t each o f 5 levels o f distraction for a total of 20 tri als per experimental s et.
Plots of time to find the target letter as a function of the number o f distractors for
the cases o f r es tric ted field o f view but full‐resolution p rost hetic vision, and
restricted field of view and pixellated prosthetic vision, both for head‐pointed
and eye‐pointed modes with t he R P condition, are shown in F igur e 2 8. Th e tim e
to find the target is the typica l m e t r ic used in visual se arc h experiments, and has
been shown to exhibit a linear tr e n d in n orm a lly sighted s ubjec ts [16–18].
The same e xper iment was perfo r med again with s imulated p rost het ic
vision, but this time with the AMD condition. Once again the e xper iment wa s
performed with cases of both restricted field of view with full resolution, as well
as r estric ted field of v iew and pixellated and blurred modes for both the head‐
pointed and eye‐pointed camera modes. In this AMD case, the su bject retain s
peripheral vision, w hich i s presented as a full‐color b ackgroun d behind t he
prosthesis s imulation region in the HMD. Plots of response tim e as a function of
number of distr a ctors for these A M D experiments ar e s h own in Fi g u re 29.
85

Fi gure 2 8 . Plot s of ti me t o fi n d the t ar g e t i n a vi s ual s e arch wi t h s i mula t e d R e t init is P igmentosa ,
comparing experiments using a limited field of view but full resolution display (Top) and a
24 40 pixella t e d displa y (Bot tom). Curves shown are f o r head ‐point ed ( Red), eye‐point e d
(Blue), full field of view HMD (M a g ent a ), a nd n orm a l vi sion (G reen).

86

Figure 29. Plots of time to find the target in a visual search wi th s i mulate d A ge ‐ R e l ate d M acul ar
D e ge ne rati on, c o mp ari n g e x pe ri me nts u s i n g a li mi te d fi e l d of v iew but full resolution display
(Top ) an d a 24 40 pixellated dis p lay (B ot to m). Cu r ves s h own a re for he a d ‐pointed (Red), eye‐
pointed (Blue ) , full field of vie w HMD (Magenta), a nd no r mal vi sion (Green).
87

Each response time plot in Figure 28 and Figure 29 shows data for head‐
pointed and eye‐poin ted camera m odes a s wel l a s for the t w o bas eli n es t aken f or
all subjects. The baseline d a ta agrees w ith literature values of visual search data
sh o wing th at 90% o f al l p e o p l e will h ave a sl o p e o f l e ss than 37 milliseconds p er
item [16]. Error bars on these plots indicate the standard error f or t hat particular
data point, and each plot was tested for significance by using a pair ed t‐test
between th e head‐poin t ed a nd e y e ‐pointed d a t a points f o r t he s a me n umber of
distractors, e ach with a significance threshol d of a p ‐value o f 0. 05. Significant
values and the consequences of this experiment will be discusse d in S ection 3 .3.4,
but the data is also presented in tabular form with associated p ‐values in T able 7.
Ta ble 7. Summa ry o f da ta collect ed d uring t he visua l s e a rch exp eriment compa ring t he hea d ‐
point e d a n d eye‐point e d ca mera m ode s for e a ch e xperiment . The p ‐values displaye d in
parenthese s c o mp are c o rre spon ding p oin t s for the he ad ‐pointed and e ye‐pointed d at a se ts, wi th
green values l e ss th a n or eq u al to a thresh old of 0.05 indicati n g si g ni fi can t d i ffe re nc e s .
Mode Res Cam
Average Time to Find Target in Seconds, with (p ‐value)
5 Objects 15 Objects 25 Objects 35 Objects 45 Objects
RP
F u ll
Head 13.75
(0.05)
17.82
(0.01)
17.14
(0.03)
15.16
(0.29)
18.94
(0.02)
Eye 9.64 10.28 9.80 12.40 11.01
24 × 40
Head 12.32
(0.20)
15.51
(0.02)
20.91
(0.02)
24.23
(0.04)
22.31
(0.22)
Eye 9.48 9.06 13.94 15.07 17.80
AMD
F u ll
Head 4.68
(0.23)
4.89
(0.13)
6.30
(0.02)
7.09
(0.07)
7.49
(0.00)
Eye 3.83 3.46 4.00 4.91 3.99
24 × 40
Head 4.32
(0.04)
5.42
(0.02)
6.02
(0.02)
6.18
(0.02)
6.41
(0.03)
Eye 3.24 3.55 3.78 4.02 4.84
HM D Full FO V B a s eline 2.29 2.53 2.56 3.04 3.13
No H M D B a seline 1.05 1.25 1.61 1.95 2.31
In order to further quantify the difference in subject performa nce
between th e cas e s, lin e ar r eg ress ion lin e s w e re f it t o eac h c urve, and slope and
int e rcept v a lues were extract e d a n d summarized in Figure 30.
88

Figure 30. Summary of slopes (Top) and intercepts (Bottom) of r egression lines fit to e ach pl o t
o f respo nse t i mes as a f unctio n of t he number o f d ist r ac to rs.
In t his case, signific ance w as d et er mined usin g MATLAB’s b uilt‐ in a nalysis
of covariance (ANOCOVA) tool, and these results will be d iscuss ed i n the
following s ection. A n additional e xperimenta l data s et i s also i ncluded here t hat
89

was omitted from the time to fin d targ et p lots: simulated AMD, but with n o
prosthesis a t all. This condition w ould b e similar to a s ubjec t with A MD w ho
performs th e task by using peripheral visio n o nly.
Lastly, each s et o f trial s p resented i n the plots shown in F igu re 28 and
Figure 29 also included a number of target absent trials in order to verify that the
subjects w ere performing w ell above chance i n their rep o rts of targ et a bsent o r
target present. For the data sets presented here, all subjects r eported correct
answers gr eat e r tha n 90% o f the time. I n a v i sual s earch exper im en t, i t has bee n
shown that a s ubject w ill typicall y termina t e the s e arch a s soon as the target is
found, which requires a scan of about half of the screen on ave r a ge. H owever,
when the target is absent, the subject will typically scan the screen t wo o r three
times to be sure that the target was not missed on a previous s can [16–18].
Typically this target absent data is not analyzed for subject p erf o rmance, but i t
will be shown that this data may be particularly useful in the case o f visual
prosthesis s im ulation d a ta.
Because tar g et a bsent trials t ook s ignific a ntl y l onger to p erform due to
multiple searches to feel confident of a correct “absent” respo nse, f ewer o v e ral l
trials o f the targ et a bsent cas e w er e per f ormed. H owev er , the experimental t rials
were c arefully p lanned and randomized such that each level of d istraction w as
shown at least twice without a target present for each subject. S ignificanc e can
be observed in the equivalent time to terminate the search vers us n umber of
distractor plots, albeit the times are much larger than those observed f o r the
90

targ et p r e s e nt s et. S uch target‐a bsent plots for the A M D mode are shown in
Figure 31.

Figure 31. Plots of time to find the target in a visual search wi th s i mulate d A ge ‐ R e l ate d M acul ar
Deg e nera t i o n, in t he t a r g e t ‐ a b s e nt c a s e, and c ompa ring ex perime nts u s i n g a li mi te d fi e l d o f vi e w
but full resolution d isplay ( To p) a nd a 24 40 pixella t e d disp lay (Bottom). C ur ves shown are for
hea d ‐point ed ( Red), eye‐point e d (Blue), f u ll f i eld of v iew HMD (Magenta), a n d n orm a l vision
(Green).
A discussio n o n the significanc e of this set of collected data and t h e
meaning of these results now follows.
91

3.3.4 Discussion and Conclusions
In the case of simulated prosthetic vision with Retinitis Pigme ntosa, an
approximately linear r elationshi p between response t ime and num ber of
distractors was observed f or e ac h case o f head‐pointed, eye‐poi nted, normal
vision with HMD, and normal vision without HMD, for both full r es olution and
24 40 electrode ar r a y resolution. T his linear r elat ionship was e xpected fo r
norm al vision cases, as has been shown previ ousl y in the visual s ear c h literatur e
[16–18]. However, the fact that a linear relationship is also observed f or l ow‐
field‐of‐view vision validates the visual search experiment as a quantita tiv e
comparison of t h e d egradation o f performance as comp a red to n or mal vision.
This linear relationship is also observed in the cases of simul ated A ge‐Relat ed
Macular Degeneration and lower electrode densities of 12 20 as w ell.
After extr acting t he k ey c haracter is tics o f the l i near t r e nds , namely s lope
and y ‐axis interc epts, differ ent s i mu lation conditions can be compar ed t o each
other and compared to t he corresp onding c ontrol c ases o f normal v ision. The
slopes o f the linea r t rends are of i nter est bec a use they i ndicate the additional
time n eeded to f ind a targ et f or e ach additio n al d istr actor tha t is a dded to t he
scene. I n t h e cases pr esen ted her e , ther e wa s not a sign ifica n t change i n slope
between the head‐pointed a nd e ye‐pointed c amera cas e s when comp aring th e
slopes using an analysis of covariance for RP, although the tim es w ere
significantly larger for the head‐pointed case. However, there was a significant
change i n slope between the head ‐pointed a n d e ye‐point ed c amera cases when
92

comparing t he s lopes us ing an a nalysis of covariance for AM D in the case of a full‐
resolution p ercept, and once a ga in t he t imes w ere s i gn ifican tly l arger for the
head‐pointed case. The results of the statistical analysis of covariance, comparing
the head‐pointed and e ye‐pointed c amera modes ar e summarized in Tab le 8.
Ta ble 8. R esult s o f t h e MATL AB A NOCOVA t ool, w hich c ompares t h e slopes and intercepts of
t h e linea r reg ressions o f t h e head‐pointed and eye‐pointed visual search time to find the target
data. V alues i n green indic ate st at ist i ca lly sig nif i ca nt diff e rences wit h a t h reshol d of 0.05.
Mode Resolution
Comparison
Data Types
Slope p ‐Value Intercept p ‐Value
RP
Full
Head‐P ointe d vs.
Eye‐Pointed
0.692 3 0.038 3
Head‐P ointe d vs.
Baseline with HMD
0.405 8 0.000 4
Eye‐Pointed vs.
Baseline with HMD
0.410 7 0.000 2
24 × 40
Head‐P ointe d vs.
Eye‐Pointed
0.499 4 0.113 0
Head‐P ointe d vs.
Baseline with HMD
0.011 6 0.003 8
Eye‐Pointed vs.
Baseline with HMD
0.002 2 0.003 9
AMD
Full
Head‐P ointe d vs.
Eye‐Pointed
0.020 9 0.368 5
Head‐P ointe d vs.
Baseline with HMD
0.000 7 0.002 0
Eye‐Pointed vs.
Baseline with HMD
0.814 7 0.025 7
24 × 40
Head‐P ointe d vs.
Eye‐Pointed
0.359 7 0.007 4
Head‐P ointe d vs.
Baseline with HMD
0.055 3 0.000 5
Eye‐Pointed vs.
Baseline with HMD
0.087 2 0.008 9
For the case of full resolution with AMD, there is a statistica lly
insig n ifica n t d iffer e nce in t he s lopes betwe e n the eye‐point e d camera c ase and
93

the case of normal vision but we arin g an H MD, but this i s not t he c ase when
comparing the head‐pointed camera case and the case of normal v ision but
wearing an HMD. This indicates that using an eye‐pointed camera is much closer
to a natural interpretation of the scene in this case, when com pared with t he
head‐pointed c amer a mode. T hese s ame c o nclusions can also b e d raw n b y
comparing each pair of time to find the target using the paired‐t test, but the
ANOCOVA analysis tends to provide a more h olistic comparison of eac h plot.
The collected d ata shows that t he y ‐interc e pt s are s i gn ificantly different
between the head‐pointed a nd e ye ‐pointed c amera cas e s for both RP (in full
resolution o nly and AMD (for the 24 40 case only . T he i n t er cepts ar e o f
interes t b ecause t his in dicates the overall time t hat it t akes a person t o locate j ust
a single target in the scene. A larger intercept indicates an overall inhibition o f
the task. It is worthy of note that at every distractor level in F ig ure 28 (RP),
Figure 29 (AMD), a nd F igure 31 (AMD, Target A bsent), the averag e time t o fin d
the tar get w as larger for the head ‐pointed ca s es and f or t he ey e‐pointed cas e s.
Lastly, analysis of head and eye motion plots may offer some evide n ce o f
improved p erforma n ce i n the visual search task when looking at the time i t ta kes
for the subject to f ind the target o bject [21]. P lots o f hea d and ey e motion a s a
function o f time d uring a targ et‐absent trial from a s ingle sub ject w ith RP
simulation are shown in Figure 32. On the left side from top t o bottom are plots
of t he h orizontal gaze coordina te, vertical g aze coordinate, ho riz o ntal h ead
motion, and vertical head motion for the case of RP with the he ad‐pointed c amer a
94

mode a nd 24 4 0 electrode r e so lution f or a t arg e t‐absent t r i al. A t right i s the
same set o f plots, but for the case o f the ey e‐point e d camer a m ode.
F i g u re 32. P lots c ompa ri ng h ea d‐pointed (L eft ) a nd e ye‐point ed (Righ t ) m o des for RP w he n
performing a target‐absent trial. Plots from top to bottom are h oriz on tal g a ze c o o rdi n ate, v erti cal
ga ze c oor dina te, horiz o nt al h ead mot i o n (ya w ) , a nd vertic a l head m o tion ( pitch ) .
In a dd ition to t he o bserved shor ter total times to t ermin a te t h e s e ar ch i n
the eye‐pointed case, more d elib erat e horiz o ntal a nd v ertic a l h ead motions can
be observed, as well as more succinct and deliberate eye motions. In addition,
when the subject intends to fixate the eye and prevent vertical m oti o n, thi s can
more easily be achieved in the eye‐pointed mode. Future work o n the
aggregat ion of h ead a n d ey e motion d a t a, a nd o n the extr actio n of s pecific
parameters , are likely w orthwhil e efforts to further q uantify t h i s improvement
in prosthetic vision performance.
95

3.4 Functional Reach and Grasp
3.4.1 Introduction
A second experiment to test the c apabilities and improv ements affo rded
by eye‐pointed rather than head‐pointed modes, amount of retina l blurring, and
electrode d e nsity was tested w ith the visual p rosthesis simulator and a functional
reach and grasp task. R eaching and grasping i s one of t he f und amenta l functions
need ed f or a variety o f daily living t asks for patients w ith a retinal prosthesis
[22–24]. T he r each a nd grasp t ask presented herein r epresents a tes t o f
ambulation and fine motor skills of the limbs in conjunction with vision‐base d
recognition a nd m onocular depth perception o f objects, as w ell as a v isual search
with both head and eye motio n to locate a target object.
3.4.2 Methods
Because any depth perception i s inherently m onocular with i ntraocular
retinal pros theses ( the FDA at p resent i mposes a r estr ic tion t o i mplant o nly in
one eye should a b ett e r t r ea tment become available fo r the o t he r eye), o n ly
monocular depth cues such as con verging lines, famil i ar o bject size, occlusion,
and object shadowing and shading can be used in such a task [25 –28]. I n
addit i on, b e cause the subject is f ree to u se h ead an d eye movem ents, motion
parallax can also be used to estimate the relative depth of objects. To test the
capability f or m onocular depth pe rception, a functional reach a nd grasp t ask was
designed as follows. A subject is seated at a table with stati c familiar o bjects
96

(cereal boxes) p laced along two differ ent depth planes ( as s how n in t h e
photograph in Figure 33 and in the diagram in Figure 34). The subject (wearin g
the visual p rosthesis simulator HMD) i s then t asked with f ind i n g a randomly
placed hand‐held familiar object (in this case a deodorant bottle) in one of four
locations in a third depth plane that is deeper than that of th e cereal b oxes,
occluded from the front in each l ocation by one of the cereal b oxes.
Fi gure 3 3 . S ub j e ct p e r fo rmi n g the re a c h and gr as p ta sk u si ng t h e visual p r o s t hesis simulator,
wi th t he t ar g e t ob j e ct v i s i b le i n the b a c k gr ound b e h i n d the ce nter occluder object. The VPS
soft w a re ca n be seen in t h e f o re g round.
As w ith th e v i sual s earch exp e riment, th e reach and grasp task was
performed in b oth head‐pointed and eye‐pointed modes, and in bo th R P and AMD
conditions. However, unlike the visual search experiment, this reach and grasp
task was performed with b oth 24 × 40 and 12 × 20 resolutions, b oth blurred and
completely un‐blurred so t hat each d iscr ete electrode ca n be c learly s een. F igur e
97

35 shows a screen capture of the VPS during one reach and grasp trial as seen by
the experimenter (top) and the subject (bottom). In this case, the subject is in the
eye‐pointed , A MD, 24 × 40 resolution, and blur red configu ratio n, reaching for the
targ et object.
F i g u re 34. Schema t ic d iag ram o f t he re a c h a n d g r a s p experimenta l set u p. The subject is s eat e d
a s c lose t o t h e t a ble a s possi ble, a nd t he v i s ua l prost h esi s s i mul a tor scene ca mera e xt ends s lig h t l y
into t he reach a nd g r a s p area. T he t arge t ob j e ct i s pla c e d r an d o mly in one o f the fou r loc ati o ns.

98

F i g u re 35. S creen c a p t u re o f t h e vi sua l p rost hesi s si mula tor, showing one reach and grasp trial
in t he e ye‐poi nted, AMD , 24 40 res o luti on, and blurred con fig ura t i o n. The e xperiment e r view
is shown (T o p ) as well as th e subject’s vie w in the H MD ( Bo ttom ).
Between tr ials, the subject is a s k ed t o clo s e their ey es ( ver i f ie d by
observation of t he e ye‐tracking ca mera o utput), and the target object i s moved to
one of t h e f our locations. A fter a s tarting tone i s played f ro m a comput e r, t he t otal
99

time t o both l ocate t he t arget ob ject b y looking around t he occ luding o bjects, and
then t o r e ac h and gra s p the ta rget o bject was recorded b y t he e xper imenter and
stored digitally for each trial. A computer was also used to gen er a t e r a ndom
target placements so as to provide an equal number of total tri als at e ach location.
Lastly, the number of occluder o bject touches was recorded by t he e x p erimen ter
for each trial. A description of the results obtained is prese nted i n the following
section.
3.4.3 Results
Data was accumulated by running six trials at each of the four possible
target locations. Time to grasp the object varied quite a bit from t rial t o trial,
because each l ocation p osed a sli ghtly different difficulty d ep ending o n if t h e
subject was right or l eft ha nded, as w ell as o n subtleties i n how the subject
executed t he s earch for the tar g et. O ver the course of m any t r ials, thes e
variatio ns were avera ged, and th e standard error became s uffici ently small. F or
exper i ment s where the percept is b lurred, a G aussian b lur of 6 0 %, p roviding
significant overlap with the neighboring pixellated regions was used in order to
overcome the gridding e ffect of t he 50% d uty cycle, a s describe d in S ec tion 1.5,
and the per c ept was pr esen ted in g rayscale.
Subjects w ere recruit e d using posted f lyers ap proved b y the Ins titutional
Review B oard ( IRB) a nd u nder IRB s tudy n umber HS‐12‐00182 at t h e Univers ity
of S outhern Califor n ia H ealth Sciences c ampus. E ntrance criteria for the study
100

included the ability to speak and understand English so that in structions could
be w ell und e rstood, tol e rance to m otion sickness and claustroph obi a so that t he
HMD could b e well tolerated, a nd a visual acuity ( including cor recti o n) o f 20/40
or better. Before participating in the study, subjects were asked to si gn an
informed consent form approved by the IRB and their visual acuity and type of
vision correction were d ocumented. Data sets were collected fr o m 10 naïv e
subjects for every combinatio n of the followin g param et e r s:
1. AMD and RP modes
2. 24 40 and 12 20 electrodes in a 20 ° horizontal field of view
3. Head‐pointed and e ye‐ p ointed camera modes
4. Blurred and un‐blurred electrodes
The combination of all variations among these parameters produc ed a
total of 1 6 data s ets for each o f th e 10 naïve s u bjects. P lots o f comparison p a irs
between pixellation w ithout b lur and pixellation w ith blur f or the 2 4 40 case
are shown in F igure 3 6 w ith the total time t o grasp av erag ed o v er all object
positions and trials on the vertical axis. A similar plot for the 12 20 case i s
shown in Figure 37. Error bars represent standard error, and t he d ashed line
represents the mean time required to grasp the object while wearing the head
mounted display, b ut w ith an u nr estr icted field of view (60 ° horizontal) and full
resolution.
101

Figure 36. Bar ch art comparing the time to grasp a target objec t f o r the pi xe llate d mo de (re d) t o
t he pixellat e d a n d blurred mode ( blue) f o r ea ch o f t he 24 40 resolut i on e xperimenta l pa ra met e r
combina tion s.

Figure 37. Bar ch art comparing the time to grasp a target objec t f o r the pi xe llate d mo de (re d) t o
t he pixellat e d a n d blurred mode ( blue) f o r ea ch o f t he 12 20 resolut i on e xperimenta l pa ra met e r
combina tion s.
102

The dashed l ine repr es ents t he b a s eline for a verag e subject per form ance
with t he b urden o f j ust wear ing the HMD itsel f , and repr esents the best t ha t a
su b j ec t is e xp ec ted to d o . Th e sa m e d a t a p re sent ed in ta b u l a r fo rm c an b e fo u n d
in Table 9.
Tab l e 9 . Ave rage t i me to g ra sp t he t a rge t ob j e ct i n t h e func ti o na l dept h perception e xperiment s .
The standar d error of t he m ean is sh o wn i n paren t heses below ea ch a verage t ime.
12 20 24 40
RP AMD RP AMD
Pixellated Only
Head‐Pointed
23.1 7
(1.29)
11.8 6
(0.56)
24.3 3
(1.19)
15.5 6
(0.92)
Eye‐P o inted
21.9 3
(2.74)
9.50
(0.30)
20.5 8
(1.58)
10.7 3
(0.59)
Pix e llated and
Blurred
Head‐Pointed
20.8 0
(1.82)
12.0 7
(0.47)
29.2 6
(3.29)
14.3 4
(1.06)
Eye‐P o inted
14.8 4
(1.18)
8.56
(0.44)
21.5 5
(1.39)
10.3 5
(0.68)
A more d etailed statistical analysis will be presented on this data i n the
following section, but in general, if the error bars shown in these figures do not
overlap, it is likely that they are significan tly differ ent wit h a hypothesis t est p ‐
value of 0.0 5. In most c ases, th e pixellated and blurred mo de compared with the
pixellated only m ode did not show statistically significant tim e differ ences ;
however, it will be shown later that the blurring is essential to m axim iz e the
improvement gained w hen using an e ye‐pointed i ns tead o f a head‐ pointed
camera. Using exactly the same data sets that were shown previously, but plotted
in a d ifferent order produces th e plots shown in Figure 38 and Figure 39.
103

Figure 38. Bar chart comparing the time to grasp for the eye‐po i n ted m ode ( bl ue) to t he h ea d‐
pointed m ode (red) f or ea c h of t he 24 4 0 resolut ion experimen t a l pa ramet e r combina t ions.

Figure 39. Bar chart comparing the time to grasp for the eye‐po i n ted m ode ( bl ue) to t he h ea d‐
pointed m ode (red) f or ea c h of t he 12 2 0 resolut ion experimen t a l pa ramet e r combina t ions.
In t hese p l o ts, the ey e‐pointed a n d head‐point ed c amera modes a re
directly c o m pared for the case o f 24 40 resolution a nd 12 20 resolution ,
104

respectively. For each pair of data points comparing pixellate d and blurred to
pixellated only, and comparing h ea d‐pointed to e ye‐pointed c ame ra m odes, the
percentage i mprovement a fter sub tract i ng t he b aselin e v a lue (da shed l ine) w as
calculated and summarized in Table 10. A paired t ‐tes t w a s perfo r med for each
comparison, a nd o nly sign ifican t va lues w ith a p ‐value o f less than 0.0 5 are shown
highlighted with yellow.
Ta ble 10. P ercenta g e i mprovement af te r b a se li ne s ub tr ac ti on f or various functional depth
percept i on e xperiment a l condit io ns. Any d ifferences b elow a t hreshol d p‐va l ue o f 0.05 ha ve b een
highlighted with yellow, an d rep lace d with values of z ero other wise.
12 20 24 40
RP AMD RP AMD
% Im prove d
with Blur
Head‐P ointe d 0 0 0 0
Eye‐Pointed 47.5 0 0 0
% Im prove d
with Eye‐Poin t ed
No B lur 0 48.6 21.6 56.4
With Blur 43.2 69.2 34.7 54.4
Although a complete discussion o f the impact o f these res u lts w ill follow,
it i s wort h mentioning t hat in n early ev ery case, t h ere is a s ignific ant
improvement in p erfo rmance w h e n comparing t he e ye‐ p ointed t o th e head‐
pointed configurations, ranging fro m 21.6 to 6 9.2 percent. W he n comparing t he
blurred and un‐blurred configura tions for this p ar ticular task, only the lowest
resolution in RP mode showed a significant im p rovement.
Lastly, an a nalysis was performe d of t he n umber of o bstacle tou ches i n th e
head‐point ed m ode compared w it h the eye‐pointed mode, which can b e used a s
another metric by which improvement can be measured. A summary of th e mean
105

number of obstacle touch es for the case of 24 40 resolution is shown in Figure
40. No baseline number is shown on this plot, because with the f ull‐resolution
HMD, each trial was completed without any obstacle touches.
Figure 40. B ar c h a rt c om paring the number of obstacle hits for t h e hea d ‐point e d mode (red) t o
t h e eye‐pointed mode ( blue) for e a ch o f t h e 24 40 resolut i on e x pe ri me nt al p ar ame t e r
combina tion s.
A discussion of the significance of this set of collected data and the
meaning of these results now follows.
3.4.4 Discussion and Conclusions
A summary o f all comparison res u lts between head‐po i nted a nd e y e‐
pointed modes, as w ell as c ompa risons o f blurred and un‐blurred electrodes
(pixellation) is given in Table 10 in the previous section. In terestingly, w h e n
comparing the eye‐pointed and head‐pointed cases, there is a si gn ific ant
im provem e n t (p ‐values less than 0.05) f or a ll cases in a 24 40 resolution
simulation, and in 3 out of 4 cases in a 12 20 resolution simulation. The two
106

cases that s howed the least improvement w h en u sing a n eye‐point ed c amer a
were simulations of RP with no electrode blurring. This sugges ts t ha t the larg est
improvement afforded b y an e ye‐p ointed c amera is r ealiz e d when this m ode is
used in conjunction with blurring of the electro d e shapes, whic h will t o a certain
extent occ ur n aturall y, depend ing in m ag nitude o n the proxim ity of the
electrodes t o the ret i na l tissue.
In a dd it ion, n o signific ant im prov em ent with a n ey e‐pointed cam er a was
shown in the case of RP with no blur and a 12 20 resolution. Review of patient
commentar y and v i deos r ev eals t hat the in terpr e ta tio n o f the s c ene is s o
confusing w ith so f ew e lectrodes and without blur, t h at t he e ye ‐pointed c amer a
mode did not add any additional information that was useful in the task.
However, when blur was added in this same simulation mode, a 43 .2 p ercent
im provem e n t in t im e t o reach and grasp was observed.
Analyz ing the same d ata using differ ent c o mparisons , n o s i gnifi cant
improvement was observed w hen comparing b lurred with u n‐blurred percepts
for any case but the 12 20 resolution simulation in the RP mode with an eye‐
pointed camera. This suggests that a lower resolution electrode array would
benefit most greatly from additional electrode blurring for thi s reach and g r asp
task. A bove a r esolution of 12 20, the effect of blurring has no significant
impact on this task; however, it is suspected that below a reso lution o f 12 20,
such a s the 6 10 resolution currently employed by the Argus II device, blurr ing
would not only p rovide i mprovement, but may be a bs olutely essential for the
107

successful completion of the task. Additional experiments are planned to t es t
exactly th is h ypothesis.
3.5 Navigation and Mobility
3.5.1 Introduction
Yet a n other exp e riment i s pres ented that s eeks to f urther q uantify the
effects of f oveation a nd e lectrode parameters on a subject’s fu nctional vision
ability. This e xpe r im e n t con tinue s t o build upon t he c o m plexity of the visual
search and reach and grasp tasks to now include full‐body ambul atory motion,
whereby visual r ecognition i s n eeded t o avoid obstacles and nav iga t e toward a
targ et b utto n at t he e nd o f a cou rse using only v isual prosthes is s im ulation. O ther
navigat i on e xper iment s i n the literatur e w ere used a s a basis t o design this
navigation t ask, incorp o rating q uantifiable metrics, course des ig n, a nd h uman
subject safety [29–34].
3.5.2 Methods
Due to t he n ature of s uch a complex e xper imen t in t erms o f data collection,
logistics, and safety, three people are required to run the nav ig atio n a n d m o bility
experiment, and the time to run each trial can be upwards of 15 minutes. Because
of t his com plexity, a n init ial pi lot study was envisioned a nd p er formed t o prov e
the usefulness of the experiment with a small set of participan ts a nd
experimental configurations. Because the primary thrust of thi s research i s to
quantify the effect of adding foveation ability to intraocular ret i nal p r ostheses,
108

the primary comparison in t his ex p e riment i s between the head‐pointed and eye‐
pointed camera modes. In additi on, the differ ence i n performan ce b etween
24 40 resolution, and limited fiel d of v iew but full resolution a re compared.
Although a limited number of sub jects and experimental configur ations a r e
presented, t he m ethods a nd e xperimental protocols can be a pplied to a wide
rang e of configu rations for a larger scale stud y in the future.
The basic idea f or t his navigati on a nd m obility experiment i s t o create an
obstacle c ourse whereby a subjec t enabled by p rosthetic vision starts i n one
location, and navigates around v arious o bstacles t o reach a tar get location. F or
the desig n o f the course itsel f , as w ell as d ata collection dur ing t h e trials, a
MATLAB application was created, and the user interface is shown in Figure 41.
At t he t op, a diagr a m depicting a long h allway c an b e seen w ith various randomly
placed o bstacles r epresent ed b y the colored, unfilled rec t angles. Each number
represents a different obstacle, and the colors represent diffe rent ob j ect heights.
The black rectangles r epres e nt s tationar y obstacles that a re b u ilt in to t he
hallway.
In o rder t o present the subjects w ith randomiz ed courses o f
approxim at ely equal difficulty, a sc oring m e tric w as d ev is ed t o compare d ifferent
course configurations with r esp e ct t o their corresponding l evels of difficulty. The
estimated course path to be taken by the subject is shown as th e solid black line
in F igure 41 , and is gen erat ed b y finding the midpoint b et ween each o bstacle and
t h e ne arby wall se ct i o n wi t h t he largest lateral gap. The sele cted point along the
109

path is then slightly shifted later a lly toward th e direc t ion of the next point in t he
path to prevent dramatic angular changes in the path. After predicting the path,
the absolut e values of t he a ngles of e ach direction change w ith respect to the
previous d irectio n a ngl e s ar e sum m ed, which produces a s ingl e s core. I t can be
assumed that a larger score repr esents a m ore difficult course, either because
more direction changes are required, or because the set of dire ction changes is
larger i n m a gnitude on the averag e .
Figure 41. M ATLAB GUI f o r cour s e gen e ration a nd d a t a collecti o n for the navigation and mobility
experiment .
Using a M ATLAB script, 2,000 random courses w ere generated, a nd a
histogram was created showing th e difficulty s cores i n bins of 4 (Figur e 42). A ft er
analyzing the histogram of p o ssible course difficulties, and after building s ome
sample cou rses w ith cardboard boxes as o bstacles, it w as d eterm ined t ha t a
difficulty s core of 120 to 140 wo uld provide a rang e that i s bo th c hallenging
110

enough for a subject, and also easily repeatable with other ran domized obstacle
placements. F our such courses w ere s e lect ed t o be u sed in t he navigation a nd
mobility e x p erimen t, i n addition t o one course to b e used f or t raining each
subject.
F i g u re 42. H istogra m plot of t he dif f i cult y scores for 2,000 ra ndomly g enera t e d courses.
The purpose of t he t raining cours e is t o allow the subject to b ec ome
familiar with the visual prosthe sis simulator apparatus, so t ha t the new
exper i ence o f wear in g the dev i ce i s not the limiting f actor on performance.
However, f or e ach trial in w hich d ata is collected, the course should b e new to t he
su bject to force the use of th e vision system and to den y any r elianc e on m emory
to aid in navigation. The training course and two of the selected courses a re
shown in Figure 43.
The final course maps were placed on a fine grid using fiducial m arkings
within t he h allway i n order to a llow for rap i d and accu rate a ss em bly of t he
40 60 80 100 120 140 160 180 200
0
50
100
150
Total Degrees of Heading Change
Number of Courses
Histogram of Difficulty Rating for 2000 Random Courses
111

courses. B ecause m ultiple subjects w ill perform the task on t h e same course, i t
is important to ensure that the courses are assembled as identi cally a s possible.
Also s hown o n the map are the individually labeled and color‐co ded placements
of each obstacle.
F i g u re 43. Two e xperiment a l cour ses (“C o urse A ” and “Cour s e B”) and one training course
(“Tr aining”) d esigned for the navigation and mobility experiment. Fidu cial mar ks a n d col or
co ded obst a c les aid in the e ase of cour s e assembly.
The selection of the obstacles themselves proved to be a diffic ult decision,
because eac h o bstacle would ideally b e an o bject that could commonly be f ound
in d ay‐to‐d a y navigation t asks aro u nd t he h ome or o ffice, would n ot p ose a safety
threat t o su bjects t hat may collide w ith th e o b ject, and would be light enough to
allow for rapid reconfiguration o f the object p lacements in t he room. The
decision was made to simply construct obstacles using cardboard b ox es, and to
print facades onto th e sides of the b oxes. Because of this, th e obstacles should b e
convex solid objects, and should h ave dimensions t hat are roughly multiples of
two feet (th e size o f a standard large cardboard box).
112

Some openly shared public domain 3D models o f common objects al ong
with P hotoshop t ouchups were u se d to complete a set of 6 u nique objects: a
bench, a bookshelf, a pillar, a filing cabinet, and two variati ons o f sets of generic
wooden drawers. Figure 44 shows front‐side images of the books helf, the filing
cabinet, and the wooden drawers in the 4‐drawer configuration ( a 2‐draw er
configuration w as a lso created). A ll obstacles were a ssembled and used i n all of
the subsequent na vig a t ion a n d mobility courses descr ibed herein.
F i g u re 44. Three exa mpl es o f f r on t fac ad es o f o b sta c les used i n t he n avi g ati o n an d mob i li ty
experiment. From left t o right : a bo ok shelf , a f iling ca binet , an d a mul t i ‐ le ve l set of drawers.
Figure 45 below shows a photogra ph o f one subject performin g t h e
navigation a nd m obility exper i ment o n “Co u rse A”. T he f inished b ookshelf
obstacle can be seen on the left side of the photograph. Note that a guideline was
followed that r equir e d the pillar o b stacle ( also s een i n the ba ckground) to f ully
113

occlude the target a t th e end of t he course, w hich i s a red b u t ton to b e pressed by
the subject to end the t rial.
Fi gure 4 5 . P ho to gra p h o f a s ub j e ct p e rfor mi n g the navi g a ti on a nd mobility experiment. The
bookshelf obst acle c a n b e se en i n t h e f o reground t o t h e l e f t , a nd the pillar obstacle can be seen in
the ba ck grou nd o cclu d ing t h e tar g et loc a t ion.
The red button was placed on a small table at the end of the co urse a nd
was large enough t o be recognized by the subjects under all sim ulated v isio n
condi ti ons. F urthermore, a mi ni mu m clearance around o bstacles of a t least 3 feet
was provided in order to allow a reasonable chance for subjects to navigate
without any obstacle c ollisions. Lastly, to prevent unintentional viewing of t he
course by l ooking aro u nd t he e dges o f the HMD screen, a black s hroud was
attached to the HMD such that it draped over the subject’s face , as c an a lso be
seen in Figure 45. The experiment itself was performed using the f ollowing
protocol:
114

1. Before the experiment, the order of courses for each of the
experimental configurations was r a ndomized such that e ach trial
was performed on a course that was new to the subject.
Additionall y , the rando m ization w a s perform e d such t hat for eac h
exper i ment al configura tion, all fou r courses were used.
2. Before the first trial, the trai ning course was assembled and the
subject was run through all of t he e xperimental configurations tha t
he or she would see, and was given plenty of time to explore th e
course in each of the simulation modes.
3. During t he e xper iment , t he e xper imenter s configur e d th e course to
match the corresponding course map for each t rial . E ach
exper i ment er s ign e d o ff that t he course mat c hed the m a p to a
reasonable expectation.
4. One exper i menter w as s tatio n ed a t the visual s imulator computer
to m onitor e ye t racking and simu lation q uality, while th e other two
experimenters were responsible for subject safety, timing the t rial,
indicating any direction changes on the floor with dropped
markers, and watching for any obstacle hits.
5. After the trial, the time to reach the target was recorded, and the
exact distance between each marker dropped on the floor was
measured t o record th e actual p ath length. O bstacle hits w ere also
recorded.
115

Th e nu mber of obstacle hits, th e time to navi gate the course, a nd the total
path l ength were a ll rec o rded. In additio n , a video recording of the entire session
was saved in a locked drawer for later review. A subject exit in ter v iew w a s also
performed to capture any qualitative experiences and comments f rom the
subject.
3.5.3 Results
Typically, p e rformance of n avigat ion tasks i s evaluated in t erm s of t he
percent a ge o f prefer red walking s p eed, tha t i s, how close the s ubject p erforms
the task with respect to walking at a comfortable speed, which varies f rom person
to p erson [32–34]. In t h is c ase, h owever, because the subject is required to wear
the HMD with a visual shroud, the baseline walking speed was taken to b e the
speed a t w h ich the subject could n avigate th e training c ourse w ith the HMD on,
but with f ull resolution a nd a n unrestricted f ield o f view ( 60 ° horizontal). The
preferred walking sp eeds fo r the case o f normal w alking d ow n a straigh t
unobstructed h allway w ith no H MD, the walking speed s while navi gating t he
training course (again with no HMD), and the walking speeds whi le n avig ating
the same t r a in ing cour s e but with t he H MD a t full resolution a r e summarized in
Figure 46. The ind i cated erro r bars represent the standard err or.
116

Figure 46. B ar c h a rt o f three cases of p referred w a l k ing speed mea s urement s i n t h e ca se o f a
norm al s ub j e ct w alki n g i n a s t r a i g h t l i n e alon g the ob st acle c ourse with no obstacles (Left), a
norm al sub j e c t n avi g ati n g t h e tr ai ni ng co u rse (Ce n te r), an d a s ubject wearing the V PS HMD w ith
full resoluti on (Right).
Note that there is not a significant difference between the pre ferred
walking speeds of all of the sub jects combined when w alking d ow n a s t raig ht
hallway, and their walk ing sp eeds when n avig ating the tr aining cours e. H owev er ,
there is a significant slowdown when performing the same traini ng n av iga t ion
task while wear ing the HMD even w ith full res o lution a nd a n unr estr icted field o f
view. T h i s latt er c as e will be r eferred to a s the “ e ffect i ve p referred w alk i ng
speed” hereinafter, a nd p erformance will be presented in terms of the percentage
of the effec t ive pr eferred walking s p eed.
After performing the experim en t in the RP mode with both full r esolution
b u t restric t ed fi e l d o f v i ew ( 20 ° h o rizo nt al in e ac h c ase) , and 24 40 resolution,
each with head‐pointed and eye‐pointed camera modes (a total of 4 variations),
117

comparisons w ere made b etween t he h ead‐pointed and eye‐pointed modes wi th
respect to b oth the percentag e o f t h e effec t iv e preferr e d walki ng s peed, and the
number of obstacle hits. A Gaussian blur of 60% was used in or d e r to o vercome
the gridding effect of the 50% duty cycle, as described in Sect ion 1.5, a nd t h e
percept was presented in grayscal e . T he i mprovements afford ed by t he e ye‐
pointed camera mode are shown in Figure 47 for a total of four subjects, each
well trained on t he t raining course, i n which the error b a rs o n ce a g a in i ndic ate
stand a rd error.
F i g u re 47. Ba r c ha rt s howing t he percent ag e improvement in e f f e ct ive pref erre d wa l k ing spe e d
when in the e y e‐point e d mode as compared wit h t he hea d ‐point ed mo d e , a n d compa ring t he f ul l
re soluti on (b u t li mi te d fi e l d of view) m o d e to the 24 4 0 reso lution m ode.
A discussion of both the significance of this set of collected data and the
meaning of these results now follows.

118

3.5.4 Discussion and Conclusions
Although the experimental data presented herein is preliminary in n atur e
with a relatively small sample set, both the goals of the exper iment and a proof o f
its ability t o s how significan t res u lts have b een a chiev e d. N a mely, a significant
im provem e n t (p‐value < 0.05) was observed in both the number of obstacle
collisions and in the percentage of the effective preferred walking speed when
using the eye‐pointed camera m ode inst ead of t he h ead ‐ pointed c ame ra mode .
Of t he t wo, the improvement in o bstacle avoidance is a m uch lar ger percentag e ,
and it c an b e argued t hat being able t o avoid obstacles while walking a t a 1 0 %
faster speed is a very impactful result.
Another interes t ing observation is t hat both w alking s p eed a nd obstacle
avoida nce a r e improved t o approximately the same e xt en t in b oth f ull resolution
(but r estric ted field of view) a nd 24 40 resolutions. T his indicates that t he
improvement is largely driven by the added benefit of foveation, although t he
effects o f b lurring a nd e ven lower electrode resolutions have n ot y et b een t est e d .
Indeed, a larger sample size to reduce the magnitudes of the er ror bars, as w el l
as additional trials to test for the effects of lower resolutio ns a nd b lur are planned
for the futu re.

119

3.6 Summary
In t his chapter, t hree k e y visual p sychophysics e xperiments w it h human
subjects w ere pr esen t e d to q uantify the effects of e y e ‐pointed as o pposed to
head‐point ed c amera modes, the effects of t he d egr ee o f e lectro de b l u rring, th e
effects o f e l e ctrode d en sity a nd cou nt, a n d how these effect s i nfluence a subject’s
capability t o perform tasks of d aily l iving with b oth AMD and RP. Each of the
three experiment s wa s tailored t o demonstr ate incr easing comple xity o f the
tasks, but to b uild u pon fundam ental visual p sychophysics a s a basis. K ey
summaries of each exp eriment a r e as follows.
Visual Search – There was no significant change in slope between the
head‐point ed a nd e ye‐pointed c a m era modes in genera l , but the y ‐intercepts
were significantly different between the head‐pointed and eye‐p ointed c amera
cases for both RP and AMD. For the case of a simulated 24 40 elec trode ar ray
with AMD, there is a statistically insignificant difference in s l ope between t he e ye‐
pointed camera c as e and th e case o f nor m al vision but wear ing a n HMD,
indica ting t hat using an e y e ‐poi nted c amer a comes very close t o a natural
int e rpretat i on o f the scene, e sp ecially w hen compared with t he head‐pointed
camera.
Reach and Grasp – When comparing eye‐pointed and head‐pointed cases
while per f o r ming a r each a nd g r a sp, the ey e‐pointed camera m od e shows a
signific ant improvement over t he h ead‐point e d camera m ode for a ll cases in a
24 40 resolution simulation, and in 3 out of 4 cases in a 12 20 resolution
120

simulation. N o improv ement was observed w hen compa ring b lurred with un‐
blurred percepts in any case but the 12 20 resolution simulation in the RP mode
with an eye‐pointed camera. This suggests that a lower resolut ion electrod e
array woul d benefit most greatly f rom optimal electrode blurring, but that adding
foveation al most a lways affords an i mprovement. H owever, it i s a lso shown that
blurring was essential to maximize the improvement gained by re storing
fovea t ion w i th an eye‐p o inte d ins t e a d of a head ‐ pointed ca m e ra.
Navigation and Mobility – A significant improvement was observed in
both obstacle avoidance and walking speed when using the eye‐po in ted cam e ra
mode instead of the head‐pointed c amera during a n avigation task. The
improvement in walking speed and obstacle avoidance is approxim ately the
same in both full and 24 40 array resolutions. T his ind i cates that t h e
improvement is largely driven by the added benefit of foveation a t these
resolutions. The effects of blurring and even lower electrode resolu tions have
not yet been tes ted.

121

CHAPTER 3 REFERENCES
[1] M. S. Humayun, E. de Juan, Jr., G. Dagnelie, R. J. Greenberg, R. H. Propst, and
D. H . Phillips, “Visual Percepti on E licited by E lectrical Stimu lation o f
Re t i na i n Bli nd Humans,” Archives of Ophthalmology , vol. 114, no. 1 , p. 40,
1996.
[2] K. Cha, K. Horch, and R. A. Normann, “Simulation of a phosp hene‐based
visual f ield: Visual a c u ity in a p ixeliz ed v ision sys t e m ,” Annals of
Biomedical Engineering , vol. 20, no. 4, pp. 439–449, 1992.
[3] K. Cha, K. W. Horch, and R. A. Normann, “Mobility performan ce w ith a
pixeliz e d vision s ystem,” Vision Research , vol. 32, n o. 7 , pp. 1367–1372,
1992.
[4] K. Cha, K. W. Horc h, R. A. Normann , and D. K. Boman, “Readi ng s peed w ith
a pixelized vision s ystem,” Journal of the Optical Society of America A:
Optics, Image Science, and Vision , vol. 9, no. 5, pp. 673–677, 1992.
[5] G. S. Brindley, “Number of information channels needed for e ffic i en t
read ing,” The Journal of Physiology , vol. 177, no. 2, p. P44, 1965.
[6] M. S . Humayun, “ Intraocular retinal prosth esis,” Transactions of the
American Ophthalmological Society , vol. 99, pp. 271–300, 2001.
[7] M. S. Humayun, J. D. Weiland, G. Y. Fujii, R. J. Greenberg, R . Williams o n,
J. L ittle, B . Mech, V. C immarusti, G . van Boemel, G. D ag nelie, and E. d e Juan,
Jr., “Visual perceptio n i n a blind s u bject with a c hronic m icro electr onic
retinal pros thesis,” Vision Research , vol. 43, no. 24, pp. 2573–2581, 2003.
[8] R. W . Thompson, G. D . Barnett, M . S. H uma yun, and G. D agnel ie, “Facial
Recognition U sing S im ulated P ro sthetic Pixel i zed Vision,” Investigative
Ophthalmology and Visual Science , vol. 44, no. 11, pp. 5035–5042, 2003.
[9] J. S . Hayes, V. T . Y i n, D . Piyathaisere, J. D . Weiland, M . S. H umayun, and
G. D agnelie, “ Visually guided pe rfo r mance of s imple tasks using simulated
prosthetic v ision,” Artificial Organs , vol. 27, no. 11, pp. 1016–1028, 2003.
[10] J . Sommerhalder, E. O u e ghlani, M. Bagnoud, U. L eonards, A. B. Safran, and
M. P eliz zone, “Sim ula t ion o f a rt ific ial v i sio n: I. E ccen t ric read ing of
isolated w o r ds, and pe rceptual l ea rnin g,” Vision Research , vol. 43, n o. 3,
pp. 269–283, 2003.
122

[11] G . Dagnelie, D. B arnett, M . S. H umayun, and R. W . Thompson , “Paragraph
text reading using a pixelized prosthetic vision simulator: par a m eter
depend enc e a nd t ask learning i n free‐view i ng conditions,” Investigative
Ophthalmology and Visual Science , vol. 47, no. 3, pp. 1241–1250, 2006.
[12] S . C. C hen, L . E. H allum, G . J. S ua ning, and N. H . Lovell, “ A quantitative
analys is o f head m ovement behaviour during visual acuity a ss ess ment
under prost h etic v is ion sim u lation, ” Journal of Neural Engineering , vol. 4,
no. 1, pp. S108–S123, 2007.
[13] D . Yanai, J . D. W eiland, M. M ahadevappa, R. J . Greenberg, I. F ine, a nd
M. S. Humayun, “Visual performance using a retinal prosthesis i n three
subjects w ith retinitis pigmentosa,” American Journal of Ophthalmology ,
vol. 143, no. 5, pp. 820–827, 2007.
[14] M . S. H umayun, J. D . Dorn, L. d a Cruz, G. D agnelie, J .‐A. Sahel, P. E. Stanga,
A. V. Cideciyan, J. L. Duncan, D. Eliott, E. Filley, A. C. Ho, A. S antos,
A. B. Safran, A. Arditi, L. V. Del Priore, and R. J. G reenberg, “Interim resu lts
from the international trial of Second Sight's visual prosthesi s,”
Ophthalmology , vol. 119, no. 4, pp. 779–788, 2012.
[15] E. Zrenner, K. U. Bartz‐Schmidt, H. Benav, D. Besch, A. Br uck m ann,
V.‐P. Gabel, F . Gekeler, U . Grep p m aier, A. H arscher, S . Kibbel, J . Koch,
A. K usnyerik, T. P eters, K . Stin gl, H. S achs, A. S tett, P . Szur man, B . Wi lhelm,
and R. W ilke, "Subretinal electro n ic c hips a llow blind patients to read
letters and combine them to words," Proceedings of the Royal Society B:
Biological Sciences , vol. 278, no. 1711, pp. 1489‐1497, 2011.
[16] J . M. W olfe, “What Can 1 Million T rials Tell U s About Visu al S ear c h?”
Psychological Science , vol. 9, no. 1, pp. 33–39, 1998.
[17] J. M. Wolfe and S. L. Franzel, “Binocularity and visual se arch,” Perception
and Psychophysics , vol. 44, no. 1, p p . 81–93, 1988.
[18] J. M. Wolfe, “Guided search 2.0: A revised model of visua l search,”
Psychonomic Bulletin and Review , vol. 1, no. 2, pp. 202–238, 1994.
[19] M. P. Eckstein, “Visual search: A retrospective,” Journal of Vision , vol. 11,
no. 5, p. 14, 2011.
[20] H. Mazyar, R. van den Berg, R. L. Seilheimer, and W. J. Ma , “Independ e nce
is elusive: Set size e f fe cts on encoding precis ion in visual s earch,” Journal
of Vision , vol. 13, no. 5, p. 8, 2013.
123

[21] A. Doshi and M. M. Trivedi, “Head and eye gaze dynamics du ring visual
attention s h ifts in com plex enviro n ments,” Journal of Vision , vol. 12, no. 2,
p. 9, 2012.
[22] S. A. Winges, D. J. Weber, and M. Santello, “The role of v ision on h and
preshaping during reach to grasp,” Experimental Brain Research , vol. 152,
no. 4, pp. 489–498, 2003.
[23] P . Servos, M. A . Goodale, a nd L . S. J akobson, “ The role o f b inocular vision
in prehension: A kinematic analysis,” Vision Research, vol. 32, no. 8, pp.
1513–1521, 1992.
[24] K . M. B . Bennett and U. C astiello, Insights into the Reach to Grasp Movement,
Advanc es i n Psychology S eries, vol. 105, E lsevier, A msterdam, T he
Netherland s, 1994.
[25] B . Rogers a nd M . Gra h am , “Sim ilarit ies be t w een m ot ion para llax and
stereopsis i n human depth perception,” Vision Research, vol. 22, no. 2, pp.
261–270, 1982.
[26] A . R. T anguay, Jr., N. R . B. S tiles, J. C risp, and B. P . M cIntosh, “ Persistence of
Monocular Depth Perception i n t h e Low Resolution L im it,” Journal of
Vision , vol. 12, no. 9, p. 1195, 2012.
[27] G . M. R edding, R. B . M e fferd, J r., and B. A . Wieland, “ E f f ect of Observer
Movement o n Monocular Depth Perception,” Perceptual and Motor Skills ,
vol. 24, no. 3, pp. 725–726, 1967.
[28] B . Rogers a nd M . Graham, “Motion parallax as a n ind e penden t cue for
depth perception,” Perception , vol. 8, no. 2, pp. 125–134, 1979.
[29] V. Prad eep, G. M edioni, and J. D . Weiland, “ R o bot vision for the visually
im paired,” i n 2010 IEEE Computer Society Conference on Computer Vision
and Pattern Recognition Workshops , pp. 15–22, 2010.
[30] D .‐R. C hebat, F . C. S chneider, R. K u p ers, and M . P tito, “N av igation w i th a
sensory s u bstitution d evice in cong e n i tally b lin d i ndivid uals,”
Neuroreport , vol. 22, no. 7, pp. 342–347, 2011.
[31] K. Cha, K. W. Horch, and R. A. Normann, “Mobility performa nce with a
pixeliz e d vision s ystem,” Vision Research , vol. 32, n o. 7 , pp. 1367–1372,
1992.
124

[32] G . P. S oong, J . E. L ovie‐Kitchin, and B. B rown, “Preferred walking speed for
assessment of mobility performance: Sighted guide versus non‐s ighted
guide techn i ques,” Clinical and Experimental Optometry, vol. 83, no. 5, pp.
279–282, 2000.
[33] D . D. C lark‐Carter, A . D. H eyes, and C. I. Ho warth, “ The e fficie ncy and
walking sp eed of v isually i mpaired people,” Ergonomics , vo l . 29, no . 6, p p .
779–789, 1986.
[34] W . D. A . Be ggs, “Psychological correlates of w alking s peed i n the visually
im paired,” Ergonomics , vol. 34, no. 1, pp. 91–102, 1991.

125

CHAPTER 4
HERMETICITY AND CLOCKING OF IMPLANTABLE
BIOMEDICAL CIRCUITS AND DEVICES
4.1 Introduction
The first half of this thesis focused on quantifying the benefi t of
implementing a n ey e‐pointed camera ( intr aocular camera ) as o pposed to a head‐
pointed camera ( extraocular came ra) for intraocular retinal prostheses. T he n ext
two chapters n ow e xpl o re a n umber of i mpro vements to e nsure tha t a camera
that is implanted in the eye (or any implantable device for tha t matter) i s
protected fr om a nd c an d etect the presence o f p o sitiv e m obile i ons and moisture.
Ion contamination i s of p articula r importance t o implantable de vices because this
form o f con tamina tion c an b e d e tr imental to m icroelectro n ic c ir cuits that r ely on
well insulated and neutrally cha rged dielectric surroundings.
Furthermore, a simple low power CMOS clock circuit is presented that can
be u sed in i mplantable b iomedica l devices such a s the intraocul ar c amera to
provide a necessary c lock s ignal for key el ectronic c ircuits. This clock generation
circuit seeks to replace the large and bulky quartz crystal osc illators t hat ar e
currently used in implantable devices. One key focus in this chapter i s on
developing such a low power CMOS clock circuit that is capable of d riv i ng a
commercial image sensor a rray, su ch a s those currently u nder c o nsid erat ion for
use in the i ntraocular camera.
126

This chapter and the next introduce eight novel chip designs created
throughout t he course of t his research o n th e topics o f hermeticity and CMOS
clocks. The lab el “Herm etic Coati ng Tes t Ch ip” (HCTC) was chos en to label these
chips because one goal is to measure the quality of post‐fabric ation deposited
hermetic c o a tings against contam i nation. N evertheless, the chips presented here
actually h av e thr ee d i fferent fund a m ental pur p oses. T he f irst is t o create a s ensor
that c an e asily be i ncluded as a s tand ard d e sig n i nteg rated w i t h microchips
created wi th the wi de sp ec trum of CM OS f ab ri cati on processes th at are sensiti ve
to positive mobile ions. These sen sors can be th ough t of as “c anar y” c ircuits that
would detect the potential for an imminent failure of an implan ted device b y
“chirp ing”, sending an inter ru pt sig nal to a mic roprocessor. The second purpose
is to create a chip that can be used b oth for optimization o f t he h erm e t i c qualities
of implantable VLSI chips by using spare CMOS process dielectri c and metal
layers a s pr otection, a n d a lso for the evaluatio n o f hermetic c oatings or p ackages
th at are de s igne d to p ro tec t aga ins t io nic c o nt am i nat i o n. L ast ly, thre e ite r at ions
of a CMOS clock circuit based on a ring oscillator have been de veloped and
incorporated in several generations of HCTC chips. These CMOS clock c ircuits
have b een s hown t o operat e with very low power consumption a s r equired for
the intraocular camera, and are suitable for driving the clocki ng c ircuits of
commercial CMOS image sensor arrays that were originally design ed t o operate
with external quartz crystal oscillators.
127

4.2 Prior Art
4.2.1 Positive Mobile Ion Contamination Detection
Because of t he d amaging effects of i on contamination o n microel ectronic
circuits, especially f or i mplant able d evices, several patents h ave been a w a rded
and published demonstrating meth ods to d etect positiv e m obile ion oxide layer
contamination t hat is i ntroduced during s tandard CMOS p rocessin g [1–6]. For
example, p ositiv e mobile i ons that d erive pr imarily fro m s odium c ontamin a tion
have the unwanted effect of neutralizing negative charge stored o n the f l oati ng
gates of memory cells [3, 7]. In the absence of contamination or u ltraviolet l ight
sources, EPROM structures such as the one shown in Figure 48 can hold a
negative charge on the floating gate for decades [8]. However, m ost attempts a t
creating p rocesses to q uantify the amount of residual positive mobi le i on
contam i na t i on a fte r c hip fabricat ion have r e s ulted in e ither ve ry c omplicated
characterization o f devices invo l ving high v oltages, or m ultipl e complex p ost ‐
processing steps.
Mitros f irst s uggested a s early as 1 993 that a n egativ ely charg ed E PR OM
device could show a measurable decrease in drain current as pos itive mobile i ons
begin to neutralize the charge on the floating gate [9, 2]. This p rovided th e
starting point for research on simple “canary” circuits that ca n be e asily
integrated o n a wide variety o f implantable devices and provide in situ
monitoring o f potential l y detrim en tal positiv e m obile ion conta mi nati on.
128

Figure 48. Example layout of an EPROM device with a floating gate. The red floating gate is
init ia lly c ha rged t o a negat i ve p ot ent i a l b y t u nnel inject ion o f elect r ons, a nd i s t h en g radua l l y
neutralized in the presen c e of posit ive mobile ions.
Furthermore, Mitros described a process by which a matching P MOS field
effect transistor (FET) can be paired with a given PMOS EPROM t o allow for
calibration of the unknown gate v o ltage as a function of m easur ed d rain current.
Then, if t he c apacitance o f the floating gate is k nown, the total number of positive
charges accumulated following i ni tial n eg ativ e charging o f the floating g ate can
be found by applying th e fo llowing simple relationship:
 
ion ox gate
QC V (8)
This concept forms the basis for all of the sensors presented i n this t hesis,
hereinafter referred t o as p osit ive mobile i on contamination ( PMIC) senso rs .
Hereinafter, the dimensions of the EPROM channel width and chan nel length w ill
follow the convention defined in Figure 48, in which the length is defined as the
distance from the drain regi on to the source region.
129

4.2.2 CMOS Clocks
Commercial CMOS image sensor arrays, as well as other key compo nents
that m ay b e found in t he e lectronic s p ayload o f an i mplantable device, typicall y
need c lock g en er ation circuit s t o drive digital components such as
microprocessors a nd a pplication‐sp ecific i nteg rated circuits ( ASICs). These clock
circuits are most often designed to be driven by a large extern al q uartz crys tal
oscillator d u e to t he h igh quality and stability of t he o scilla tion t hat ca n be
achieved. H owever, within t h e s trict surg ical constra ints o f t h e i ntraocular
camera and other such ultraminiature implantable devices, even the smallest
crystal oscil l ator p ackages to d ate (Figure 49, L eft) h ave proven to be too large
and power consumptive for these a pplications. Another alter n ati ve t hat has
gained popularity due to their very small size and very low pow er d issipatio n a re
MEMS r esonator c ircuits, such a s the one shown in F igure 49 (Ri ght) [ 10, 11].
Although these oscillators also have very good stability, they have p roven to b e
very difficult to fabricate, and also have resonances that best produce clocks in
the lower kilohertz range.
A third solution i s an i n t egr a ted on‐chip, l ow‐power, small foo tprint V LSI
clock c ircui t d esigned entirely w ithin the CM OS p rocess, such a s a ring o scillator
consisting of an odd number of digital inverter stages. Most C MOS ring o scillators
are v e ry fas t (in the g i g a hertz r a ng e), and so m ust be s lowed d own considerably
to o perate i n the megahertz range required b y many C M O S image s ensor arrays .
Prev ious r esearch has been p ublished o n slow, yet relia b le, rin g‐oscillator‐based
130

desig n s, which demonstrate the po tential fo r a n u ltra‐low‐power ring oscillator
de sig n ed sp e c i f i c al l y fo r an im p l an tab l e d e vic e th at ne ed s a c l ocking circuit [12–
15]. Most notably, several key references demonstrate a fully‐ integra t ed c amera‐
on‐a‐chip with a low‐power r ing oscillator clock [15–17].

Figure 49. The s mallest k nown q ua r t z cry s t a l os cillat o r pa ck a g e ( L e ft ) is t he N ihon N X1612AA,
and is s hown com pare d to a m illi meter ru ler. The c rystal o scil lat o r si ze o f 1.6 1.2 0.3 mm i s
st ill t o o la rg e for ma ny micro‐impla nt a b le d evice applica t ions. A ME MS o scillato r (Righ t ) can be
made small, but is difficult to fabricate and usually operates best i n t h e k i lohert z rang e; f rom [11] .
One potential issue with r ing oscillators i s their inher e nt i ns tability w ith
respect to a mbient t em perature, and the difficulty o f fabricating oscillators with
consistent and precise frequency values due to fabrication proc ess variatio ns.
However, when implanted in the human body, a much more consiste nt
temperature can be assumed due to natural regulation of the bod y temperature.
In a dd itio n, r ing oscilla tor circ uits a re s o smal l that s ev er al d esig ns w ith slightl y
differ ent center f r e quencies m a y b e desig n ed, and the closest t o the desired
frequency after fabrication can be selected by a multiplexer. Several CMOS r ing‐
oscillator‐based clock d esigns t hat run reliab ly a t only 6 M Hz (suita ble to d rive
one part icular CMOS image senso r arr ay) are presen ted in Sectio n 4. 4.2.
131

4.3 First Generation Hermetic Coating Test Chips
4.3.1 Hermetic Coating Test Chip 1 (HCTC ‐1)
The herm et ic coatin g t est chip H C T C‐1 chip d esig n was fabricat e d by the
Metal Oxide Semiconduct or Implementati o n Servi c e (MOSIS, Mari na d el R ey, CA)
using the ON S emiconductor 0.5 µm C5 process on August 6th, 200 9 (run n umber
T95S). T he C 5 process uses a n epitaxial sil i con s ubstrate containing 3 metal
layers and 2 poly layers whereby the two poly layers can form c apacitors with a
capacitance per unit a rea of 950 aF /µm
2
. A high‐resistance layer (1050 ohms per
square) is a lso available to c reate resis t ors in a s mall area using the Poly2 layer.
The mi ni mum f e ature si ze i s 0.5 µm with a recommended voltage o f 5 for supply
inputs. A full set of measured process parameters a fter f abric at ion is p rovided in
Appendix A. An overview of the HCTC‐1 chip layout and a photom icrograph of
the as‐fabricated chip are shown in Figure 50. A summary of al l key devices that
were tested follows in Table 11.
132

Fi gure 5 0 . T he C ade n ce l ayou t (T op ) a nd an o p t i c al m i c ro gr aph ( Bot t o m ) of H e rmeti c C oa ti ng
Test C hip 1 ( HCTC‐1).
133

Ta ble 11. Summa ry o f k e y devices tha t w ere t e sted o n herme t ic c oat i ng t est chip HCTC‐ 1
including size, loc a ti on, an d pr esence of a pa ssivat ion oxide c ut f or ea c h device.
Device
No.
Device Type Description / Dimensions
Oxide
Cut?
Location
(x, y in µm
from bottom ‐
left corner)
1 Cap a ci tor
Int e rdig i t at ed Met a l 1 elect rodes wi t h 40 fing ers of
size 9 µm 157 µm wit h 4. 5 µm g aps a n d a desi g n
cap a cit a nce of 1 pF
No 850, 13 3 8
2 Cap a ci tor
Ident i ca l to D evice 1 but with a Metal 2 o verlayer
a c t i ng a s a hermet ic s hield
No 1220, 1 338
3 Cap a ci tor
Polysilicon ov e r substr a t e c a p a cit o r of siz e
20 µm 300 µm a nd a desi g n ca pacit a nce of 1 pF
No 1006, 1 228
4 Cap a ci tor
Ident i ca l to D evice 3 but with a Metal 2 o verlayer
a c t i ng a s a hermet ic s hield
No 1376, 1 228
5 Cap a ci tor
Ident i ca l to D evice 3 but with a Metal 2 o verlayer
a n d Met a l 3 overla yer a c t ing a s hermet ic shield s
No 1513, 1 228
6 PMOS P MIC
P‐ Cha nnel EPR OM t o ac t as a po s i t i ve mobile io n
con ta mi na ti o n se n sor wi th a P oly fl oa ti ng g ate are a
of 30 µm
2
a nd a c hannel of s ize 9 µm 3.6 µm
Yes 606, 12 6 8
7 PMOS F ET
Ident i ca l to D evice 6 but pi nned ou t as an FET for
calibra t ion of t he PMIC sens or1`
Yes 606, 12 4 2
8 Surfa c e Prof ile
Prof ilomet er test st ruct ure to me a sure t hick ness of
various met a l and oxi d e laye rs
Yes 1005, 9 09
9 CMOS Cl o ck
F i ve stag e 116 µm 32 µm ring oscillat o r wit h
ad ded ca pa cit a nce to slow t h e tar g et ou t p u t
f requency to 25 MHz
No 566, 69 0
10 CMOS Cl o ck
Simila r to Device 9 but m odi fied f or a 10% slower
outpu t f re que n cy wi t h o ut c hangi n g the si ze
No 705, 50 4
11 CMOS Cl o ck
Simila r to Device 9 but m odi fied f or a 10% fa s ter
outpu t f re que n cy wi t h o ut c hangi n g the si ze
No 896, 69 0
12 CMOS Cl o ck
Fi ve st a ge 8 8 µm 3 2 µ m r i n g o s ci lla tor de si gne d
with an ou tpu t frequency o f 50 MHz but w i th a
D ‐ Fli p ‐ F lop o u tput s ta ge f o r a 2 5 M H z fi nal ou tpu t
No 1000, 5 04
13 CMOS Cl o ck
Simila r to Device 12 but mo dified f or a 10 % sl o wer
outpu t f re que n cy wi t h o ut c hangi n g the si ze
No 1134, 6 90
14 CMOS Cl o ck
Simila r to Device 12 but mo dified f or a 10 % f a s ter
outpu t f re que n cy wi t h o ut c hangi n g the si ze
No 1267, 5 04
134

The purpose of t his fir s t tes t c hip was to p ro vide a n arr a y of stru ctures
that a r e p otent i ally s en sitiv e t o the presence o f eith er m oistu re o r positiv e (e.g.,
sodium) ions, and that therefore may be used to detect imminent failures of a
bioimplantable device. Such hermetic coating test chips may also be able to
quantify the quality of the hermetic seal for a given thin film coating o r microcase
packaging approach w hen exposed to s alin e solution. T he c hip a lso contains
prototype designs of t he c lock g en erato r s bas e d on r in g o s cilla tors, as d escribed
in the introduction and to be discussed in the following sectio n. In a dd it ion,
several unique structures were i ncluded for positive m obile ion d et ection. T he
most s ensitive s tructures are a pair consisting of an EPROM and a PMOS
calibration FET to be used as described at the beginning of the Prior Art section
and the PMOS structure is illustra ted in Figure 51 below.

Fi gure 5 1 . An i llustra ti o n o f the ge ne rali ze d de si gn o f the PMO S EPROM t o b e used a s a posit i ve
mobile ion c o n taminati on s ensor.
135

The chip a lso contains small poly silicon‐over‐grounded‐substrat e
capacitors, and larger c apacitor s that consist o f two inter d ig i ta ted Metal 1
electrodes o ver the gr ounded s ubstrat e . T h e i nt erdigit a ted el e ctrodes may be
shorted to produce a traditional MOS capacitor with a high degr ee o f penetration
of ions to the intervening dielectric layer, or used to measure a n impedance
variatio n betwe e n the electrodes f or h um idity and m o isture d e t ection. The
capacitors m ay a lso be a ble to d etect ion contamination b y way of a lter ed I‐ V
curves due to changes in the charge distribution within the die lectric layer
resulting from d rift o r diffusio n. O ther t est structures f abr icated on H CTC‐ 1
include a b a nk o f resistors to s imulate thermal hot spots as w e ll as s lightly
increased o p erat ing t e mperatures , such as those found in the bo dy at 37 ° C.
Additional metal and dielectric layers were also added to test the
enhancem ent of h ermeticit y t hat may accrue t o incorporating s up ervening s par e
metal and dielectric l ayers that a re a lready a vailable w ithin t he f abrication
process. Other structures such as bare exposed bonding pads (electrodes) a nd
m e tal trace s w ith varying spatial fr equencies p r ovide basic tes ts o f metallic layer
corrosion in the presence of saline. Initial experiments were com pleted t hat
showed n o visible corrosion o f these st ructures, but longer t er m tests are
planned for future research. Finally, a structure in the center of the chip was
introduced t o allow fo r measurement (to the extent p o ssible given the design
rules) o f the thicknes s of t he i nd iv idual m e tal and oxid e layer s of t his particular
fabrication process.
136

Among these various test structures is the star of the show, an erasable
programmable read‐only memory ( EPROM) d evi c e wi th a f loati n g ga te, wh ich
will be u sed as a p ositive mobile i on contamination ( PMIC) sens or ( Table 11,
Device 6). EPROMs have been shown to exhibit modified I‐V characteristics when
in the p resence of positive ion contamination, which decreases the negative bias
charge. When an EPROM is paired with an identical geometry PMO S FET,
calibration measurement of t he P MOS FET I‐V character i stics can be performed
to determine the ionic contamina tion concentrations measured b y the EPROM
PMIC floating gate sensors. NMOS FETs were also fabricated to determin e the
relative s en sitiv i ty o f their I‐V characteristics to t he p resence of negative mobile
ion contaminatio n .
Fi gure 5 2 . P ho to gra p h o f t he H CT C‐ 1 chi p p ac ka ge d i n a c e ra mi c 40‐pin dua l ‐inlin e pa ck ag e
(DIP) wit h t he cover removed. W ire bon d s (thin gold w ires) c a n be seen connect ing the bonding
pad s o n the p e ri me te r of th e c hip to the packaging pins.
The 2 mm square chip was mounted in a ceramic 40‐pin dual‐inlin e
package (DIP) so t hat accelerate d lifetime t ests could b e later performed at
elevat ed temperatures using a pr e c ision ov en. This pack a ge is designated as the
“DIP40” f ro m MOSIS and is r eported to b e a Kyocera K D‐78163 p a c kage w ith
137

operating temperatures l isted as h igh as 100 °C. A ll contacts are gold t o avoid
any corrosive e ffects o f saline contamination o n the package it self. A p hotograph
of one packaged chip is shown in Figure 52.
4.3.1.1 Per f ormanc e Char acter i stic s of CM O S Clock Circuits
The goal o f the resear ch o n CMOS c lock c ircuits is t o implement an on‐
chip, compact, low power clock gen erator i n implantable biomedi cal devices to
replace the trad itio nal crystal os cillator. T wo variatio n s of clock generator
circuits based on a ring oscillator were designed, simulated, a nd f abri cat e d as
part of the HCTC‐1 design. The key design element of the first variation, shown
in F igure 5 3 , i s a tr adit ional ring o s c illator run n ing at a ppr o ximately 5 0 MHz with
a D‐flip‐flop used a s an o utput stag e to b oth divide t he c lo ck frequency by two (to
approximately 25 MHz) and to prov ide a nearl y perfect 50 % duty cycle.
Figure 53. Electrical schematic of the simulated and fabricated 25 MHz CMOS c lock g enera t or
with a 5‐stage ring oscillator an d a divide‐by‐two output s t a ge cons i s t ing of a D ‐f lip‐f l o p w ith
buffering inve r ters on eithe r side of t h e D ‐ f l ip‐f lop.
138

Each element in the 5‐stage ring oscillator is designed to have a large
parasitic output c apacitance, wh ich has the effe ct o f slowing d own the trans i tions
as much as possible to achieve an output frequency in the megah ertz r ang e .
Another design variation was included on the chip, also consisting of a ring
oscillator but with n o flip‐flop stage, a s shown in F igure 54. In this case, the
c e nte r f req u enc y w as r edu c ed to a p p ro xim ate l y 25 M H z b y addi ng an addi ti onal
delay cell to e ach inverter t hat consists o f a capacitive c ompo nent s hunting the
output of each inverting stage t o slow down th e oscillations ev en furt h er, as w ell
as a dding t wo a dditional i nverter stages. T o account f or p roce ss variations, both
slightly s lower and slightly f aster clock generator c ircuits (d esigned to b e
approximately 10% d iffer e nt i n each direction in frequency) of each o scillato r
type were included on the HCT C ‐1 test chip.
Figure 54. Electrical schematic of the simulated and fabricated 25 MHz CMOS c lock g enera t or
with a 7‐stage ring oscillator an d n o D‐flip‐flop ou tput stage. Buff ering invert ers are included a t
the ou tpu t t o supply su ffi cient curren t t o a c a p a citive lo a d.
The delay cell schematic is s hown in Figure 55, which consists of a
standard i nverter coupled w ith a single NMOS FET with d rain a nd s ource shorted
139

together a n d grounded to f orm a small capacitanc e betw een t he g a t e and the
substrate. This acts to lower the frequency of the ring oscill ator since t here i s
additional l oad capacitance on the output of the inverter, and the time constan t
to change voltage levels increases.

Fi gure 5 5 . S che m ati c d i a gra m o f the d e lay ce ll use d i n the ri n g ‐os c illato r‐based cloc k design o n
the CMO S c l o ck t e s t chi p ( CCTC), sh owi n g an i nve rte r wi th a d d e d l oa d ca pa cit a n c e t o d ecrea s e
t h e f requency of oscilla t i on.
Simulations using PSPI CE w er e per f ormed after ext r act i ng t he n e tlist from
the design in Figure 53 using Cadence Virtuoso. The results we re a nalyzed and
plotted usin g Cadence S p ectre, a nd o ne e x a mple w av eform is s how n in F igure 56.
The simulation s hown h ere assume s a 0.5 pF o utput load, which is consistent
with the input capacitances typical of various on‐chip circuits t hat require a clock
signal. This simulation was run assuming a 5 V power supply, w hich i s the
recommended v alue f or t he ON Se miconductor C 5 pro c ess. T his CM OS c lock
configuration produced an estimated 3 mW of power dissipation, which at f ace
value is not much better than other commercially available micr oelectronic clock
generation c ircuits. H owever, if t he c hip wa s instead fabricated i n a 0.13 µm
140

process, both t he voltage a nd l inewid ths would be s maller, r esu lting in a n
estimated further r eduction b y a factor o f about 16, w hich w oul d be u nder
200 µW o f power dissipation, well w ithin the power budget o f th e intraocular
camera.
Fi gure 5 6 . S i mulate d o u t put w a ve fo rm o f the CMO S ri n g o s ci llat or with a 5 V power supply and
a 0.5 pF loa d.
Figure 57 s h ows a screen p rint o f the layout o f the CMO S r ing o scillator
circuit schematic shown in F igure 53, a s show n in C ad enc e V irtu oso. The 5‐stage
ring oscillator is on the left, and t h e D‐f l i p ‐f lop st andard ce ll is on the r i ght. Note
the elongat e d inv e rt er t rans istor c h annels d es igned to s low the switching time of
each inverter, thereby tuning the clock to the desired frequency. The entire clock
generation circuit is ap proximatel y 88 µm 32 µm i n size.
141

F i g u re 57. Ca d ence l a y out of t he C MO S cloc k genera tion c ircui t sho w ing a 5‐stag e ring o sc illa t o r
(Le f t) c ouple d wi t h a s t an d a rd ce ll D ‐ fli p ‐ f lop ou tpu t s t a ge ( R ig ht).
The CM OS r ing oscillator‐based clock generators were t ested, a n d lo ng‐
term statistical data on both frequency and duty cycle stabilit y were collected.
Output waveform shapes and power dissipation were also measured. Key
measurements are summarized in Table 12 below. Due to process variations, the
slightly lower frequency (label “25– MHz”) design was the close st t o the desired
design frequency of 25 MHz, while this design also had the larg es t V pp voltage
output. A lthough the fl ip‐flop designs were o riginally i ntende d to gen erat e a duty
cycle closer t o 50% t han the designs without the flip‐flop by e liminating a ny d uty
cycle anisotropy in the higher frequency clock signal, the data suggests that the
flip‐flop output stage might not be necessary for achieving nea r 50% d uty cycle
performanc e in f uture itera t ions.
In addition, later experiments performed on commercial CMOS ima ge
sensors sh owed r emarkable duty c ycle i nvarianc e fo r sensor c loc k i ng, and as
such the flip‐flop designs may not provide significant benefits . H owever, due to
the addit i o n al p arasitic r esist a nce introduced b y the delay cel ls i n the 25 MHz
142

design variation, the d i ssipated power was measured t o be s lightly higher for this
case t han for the 50 M H z design w ith a divide‐by‐two flip‐flop stage. The peak‐
to‐peak voltage obtain ed w ith th e delay cell design w as a lso hi gher, which may
contribute t o the highe r power dis s ipatio n.
Ta ble 12. S imula t e d f re quency a s well as mea sure d f requency, d u ty c ycle, peak‐to‐peak v ol tage,
and dissi pa ted p o wer f o r six di ffe rent CM O S cl oc k generati on ci rcuit s on HCTC‐1.
Ring Oscillator
Design
Simulated
Frequency
Measured
Frequency
Mean
Duty Cycle
Peak ‐to ‐Peak
Voltage
Power
Dissipated
25 MHz 25.7 MHz 27.33 MHz 51.65 % 2.55 V 7.7 mW
25– MHz 23.5 MHz 24.80 MHz 51.60 % 2.90 V 6.9 mW
25+ MHz 28.0 MHz 29.76 MHz 52.60 % 2.57 V 7.7 mW
50 MHz wit h DF F 25.7 MHz 27.16 MHz 50.89 % 1.48 V 6.4 mW
50– MHz with DFF 23.1 MHz 24.45 MHz 52.31 % 1.60 V 6.3 mW
50+ MHz wit h DF F 28.8 MHz 30.51 MHz 47.59 % 1.31 V 6.6 mW
All designs showed a h igher than e xpected power diss ip ation, b y a factor
of 2, than the SPICE simulated designs predicted. The reason for this is likely due
to inc r e as e d l o adi ng fr o m th e p o w e r m e asu r e m ent i tsel f . Th e p o wer dissipation
measured f or t he 25 MHz clock variants given in T ab le 12 is b as ed o n an
operatio nal voltage of 5 V, and is in agreement with SPICE simu lations when t he
load c apacitance contr ibuted b y the DIP40 package pin and oscil loscope probe,
along with a ll additional p arasit ic c apacitanc e a nd r esis tance from e xtraneous
traces are included. The power dissipation is expected to be reduced b y
143

approximat ely one to t wo o rders of m agnitude i f the clock c ircu it i s im plem ented
in a process with a smaller feature s i ze, and operated a t a v o l tage of 1 V.
A LabView add‐on t hat works in co n junction w ith a Tektro nix osc illoscope
was acquired in order to accurately measu re key properties of t he ring oscillator
circuit, i ncluding t he m ean oscill ation p e rio d a nd c lock j itter. Clock jitter can be
de fi ned in a nu m b e r o f w ays, b u t i n th i s c as e , th e o sc i l l at o r was char a cter ized b y
the amount o f deterministic perio d j itter, D
J
, using a delta‐delta measurement
method. The delta‐delta measurement seeks to fit the histogram d eriv ed f rom
m a ny p er iod m e asure m ents w ith two Gaussian d is tribut ions t hat r epresent n on‐
determinis tic (noise) jitter a t the locations of t wo d istinc t p oints of t h e h istogram
[18, 19]. The time difference between the two peaks of the Gau ssian fits c an b e
shown to b e the deter m inistic jitter pr esent in t he c lock, ev en in the case of a
noise‐free environment, which characterizes the quality of the clock generator
excluding contributio ns d ue t o noise. T his jitt er d eriv es p r i m arily from
inconsistencies i n the clock d uty cycle over t ime. A n example of the LabView
so ftw ar e m easu ri ng th e 50 M H z with D ‐ f l i p ‐ flo p ring o sc i l l ato r desig n (Table 11,
Device 12) is shown in Figure 58.
144

F i g u re 58. S creen c a p t u re o f t h e L a b V iew GUI a pplica t ion for t h e me a s urement of c lock j itt e r. The
wavefor m i s s h own (T op ), i n ad di tion t o period deviati o n mea s ur ement s o ver t i me w it h respec t
to t he a ver a g e p eriod (B ot tom Lef t ) an d a his t o g r a m of a ll peri od deviati o n measurement s
(Bot to m Rig h t).
Peak t o peak voltages and overa ll output w aveform shapes w ere s en sit i v e
to capacitive loading of the measurement equipment, causing low er t han
expected v oltages and alter e d w a veforms as s hown i n F i gure 5 8. In this case,
there is n o visibly dist inct s ep ar ation between two Gaussian d i stributions due to
the alter e d waveform, so t he c al culated clock jitter may be i na ccurate. F uture
iterations o f these clock generation circuit designs include hi gher c urrent
capacity o u t put buffering, and ne w low‐capacitanc e active p robe t ips have b een
acquired to produce more accurate output waveform measurements, a s will be
145

discussed in Section 4.4.2. Statistical measurements of the cl ock p eriod varia t io n
of the nominally designed 25 MHz clock circuit as measured by the LabView add‐
on showed a mean frequency of 27.13 MHz and a deterministic per iod jitt er o f
511 ps. T h i s correspo n ds t o a deterministic variatio n between 26. 76 MHz and
27.51 MHz. Although the effect of this level of clock jitter i s unknown when u sed
to d rive com m e rcial im age senso r s, Section 4.4.2 prese n ts r esul ts o f a simila r
clock generator driving a commerc ial im age s e nsor, but at a slower frequency.
In order to verify that the distorted wavefo rm is indeed due to a dditional
capacitive l oading, additional s im ulations o f th e orig in al c loc k designs were run,
with results as summarized in Table 13. The original simulatio ns assumed a
capacitive l oad of 0.5 p F, w hich is reasonable for the clocking of other on‐chip
circuits and is consistent with the input capacitances likely e ncountered b y CMOS
trans i stors. H owever, larger c apa c itanc e s can be s een w hen dri v i ng c ircuits on
other chips, including capacitance due to w ire bonds (approximately 1 pF) and
added cap a citance due to t he p ackaging a nd o scilloscope probe ( an a dditio n al 2
to 10 pF). Indeed, sim u lations of t he 10 pF l oading c ase closely matched the
waveform observed du ring measurement.
146

Ta ble 13. Compa rison of s imulat ed w a v ef orm out p ut s f o r f o ur d i f f e rent c apac i t ive lo ad s o n t he
cloc k genera tion cir cuits on H CTC‐1.
1.0 pF ca pa cit a nce due t o b onding p ad s 2.0 pF ca pa cit a nce due t o w ire bonds

5.0 pF equiva l e nt to me a s urement wit h a n
ac ti ve prob e
10.0 pF equiva lent t o mea s urement wit h a
pas s ive probe
To explore the feasibility of using a clock generator circuit such as t his, a
commercial image sensor was chosen to serve as a test of the os cillator. The small
form f acto r OmniVisio n OV6920 i ma ge s enso r array typically r equires a large,
power‐consumptive p assive e xterna l crystal oscillator centered at 6.14 MHz to
function correctly. The oscillator is of the Pierce type, and typically i ncludes an
invert er i n t egra ted on t he V LSI image sensor i tself. T h e r esul ting w avefo r m is
likely sent to a clock generator system with a phase‐locked loo p fo r the signal
shaping necessary f or o ptimal image senso r operations.
147

The crystal oscillator from a n OV6920 dev e lopment board was rem oved
and the effect of externally providing a variety of different w aveform types to t he
CMOS image sensor instead of the signal generated from the crys tal oscillator was
explored. S urprisingly, both an AC square wav eform and a sinusoidal waveform
of 1.3 to 2.5 volts peak‐to‐peak and centered at a frequency of 6.14 MHz were
able t o drive the clock generatio n circuitry and allow for mono chrome sensor
operatio n. Further res u lts of t his experiment a re p res e nted i n Section 4.4.2 as a
continuation of the clock gen erat io n re fin e m e nt e ffort.
4.3.1.2 Per f ormanc e Char acter i stics of P M IC Sensors
The basic EPROM stru cture discussed in t h e i ntroductio n and sho wn in
Figure 48 was fabricated using a standard PMOS FET design with a floating g ate
that is capable of storing a negative bias charge. EPROMs have b een shown to
exhibit mod i fied I‐V c ha racterist i cs i n the pr esence o f posit i v e ion contamination,
which grad ually d ecreases t he n egative bias charge. T herefor e o n e possible
measurement techn i que to d et ect t h e presenc e o f ion cont a minat i on i s to m onitor
the drain current of the EPROM, and then estimate the gate volt ag e over t ime
using a calibration curve. When the EPROM is paired with an id en tic a l‐geometr y
PMOS FET with a non‐floating gate, measurement of the PMOS FET I‐V
character i stics can be p erformed t o determine the pres ence o f p os itiv e m o bile
ions as w e ll as ionic co n ta mination concentrations.
148

The EPROM is initially charged by applying a high differential voltage
(–12 V) b et ween t he s ource and drain, a llowing n eg ativ e charges to be injected
by tunneling into and trapped on the floating gate. The measur ement o f t he d r a in
current u ses a mea s urement s e tup iden tical to t he c hargin g setup, but with only
a –1 V differential voltage applied between source and drain. Figu re 59 show s
the charg i n g a nd m ea surement s etup o n th e left a nd t he P FET c a l ibratio n
measurement on the right. The calibration has been shown to be very consistent
from chip to chip, and is typically only measured once for each experiment. The
c al i b ratio n setu p al l o w s th e g ate v o l tage to b e sw ep t o v er a r a ng e to observe th e
corresponding effect on the drain current. Then, after measuring the drain
current o n the floating gate d e vice, the calibration curve can be used to estima te
the float i ng gate volt ag e.
Fi gure 5 9 . The t wo e le ctr i cal me asure me n t se tups u se d f o r the posit ive mobile ion conta minat i on
ex pe riment s , inc lud i ng t he cha rg ing a n d mea s uring of t he EPR OM (Left) as well as the extraction
of t he P FET ca libration curv e (Righ t ). I n b o th c ases, the drai n i s a t the top an d the so urce i s a t t h e
b o t t om o f the tran si st or. Th e b ulk sub s tra t e an d N‐ we ll are b o th t ied t o g roun d.
149

The PFET calibration setup was used to extract the I‐V curve for a PF ET
device o n HCTC‐1, an d subsequent d evic e measurements w er e s h own to be
within 1 percent of this curve. The gate voltage, V GS , was stepped fro m 0 t o 8 V in
0.5 V increments, and the drain‐source voltage, V DS , was set to – 1 V, a s it i s during
normal data collection of the EPROM PMIC sensor. The voltage a cross a 100 Ω
resistor was measured, and then th e corresponding drain current w as c alculated
to generate the plot in Figure 60, which shows the calibration curve of t h e
measured drain current as a function of the applied gate voltage. The value of the
100 Ω res i s t or w as c ho sen such t hat a majority o f drain‐s o urce volt age appe ars
across the transistor channel, which has a resistance o f ap prox imatel y 3.5 kΩ.
F i g u re 60. P lot of t he m ea sured dr ain current i n milliamps a s a function of the applied gate‐
source voltage for a PFET on HCTC‐1, with V DS = –1 V. This plot is measured from the PFET device
with a n on‐fl o a t ing g a te, a n d can be u s e d to e sti m a t e the flo a t ing gat e v o l ta g e o f a si mila rl y
cons t ru cted E PROM.
A logarith mic fit was applied to t he c alibra tion c urve i n Figur e 6 0 f or
values greater than 2 V; a solution for gate voltage as a funct ion of m easured drain
150

c u rrent c a n th en b e fo u nd, as e xp re ssed b y E q u atio n 9. Th i s p a rtic u l ar fit h as a n
R
2
value (the coefficient of determination) of 0.9997, indicating a n excellent f it.
This calibration curve can be used to find the gate voltage, or m ore specifically
the change in gate voltage for a given change in drain current. When Equation 9
is combined with Equation 8, the corresponding change in the nu mber o f
electrons o n the floating gate c an b e est i mated. Because t he n egat iv e charg e s o n
the gate a r e n eutral ized b y positive i on contaminatio n , an y obs erved d e creas e i n
electric c harge corresp onds t o a change i n ionic contamination accu mulated over
tim e , ther e b y providin g an indica t ion of the h erm e t i city o f the chip.
100 . 7 7
208.21




d
I
gate
Ve
(9)
In w hich Vgate is the voltage on the EPROM floating gate in volts, and Id is the
measured E PROM drain current in µA at a drain‐source voltage of –1 V.
The EPROM‐based PMIC sensor was initially charged by applying a
negative voltage differential between the source and drain, all owing negative
charges to b e injected i nto and trapped on t he f loating gate o f the EPROM via
tunneling. A voltage of 12 to 15 volts is necessary to fully program the EPROM to
a gate‐to‐source differential of about 5 volts. The charged EP ROM device
(packaged, b ut w ith the lid removed to e xpose the H C TC‐1 c hip) was th en
immersion soaked i n a buffered s a line s olution in a d ar kened en vironm ent at
room t emperatur e . T he gat e volt age w a s th en r e‐meas ured ( by m e asuring th e
drain current, as d escribed above) to determine if any unwanted Na+ or K+ ions
151

were a ttr ac ted to t he n egatively ch arged floating gate without any ex ternal b iases
applied to the chip. A photograph obtained during one such sal ine so ak
experiment is shown in Figure 61.
F i g u re 61. P hot o g raph of t he s a l ine soa k e xperiment underw a y w ith the first hermetic coating
test chip (HCT C‐1).
The pres en ce o f saline c aused si g n ifican t re d u ctions i n t h e flo ating gate
voltage, a s expect ed, a n d the gate voltage w a s r etained over s e veral long n on‐
soaking rest periods between soak tests, with intervening rinse s in d eionized
water. Table 14 shows a summary of the calculated gate voltage a fter a pplying
the PFET calibration expressed in Equation 9 to the measured EP ROM drain
current. N ote that o ver the course of t he c umulative soak, the g ate voltag e
dropped significantly. Not shown in the table is a long unsoak ed r est perio d
between the 6 hour a nd 29 hour m easurements wher e no d rop was o bserved.
This confirms the hypothesis that the drop in gate voltage is b ein g c aused by
external contamination at room temperature.
152

Ta ble 14. Cha ng e in t he mea sured dra i n current a nd f loat ing g a t e volta g e a s H CTC‐1 chips a re
soaked in a saline solution for 29 hour s cumulatively.
Cumulative
Soak Time
Measured
Drain Current
Calculated
Gate Voltage
Total Gate
Voltage Drop
0 Hours 123.6 µA 2.938 V ‐‐
2 Hours 122.7 µA 2.925 V 13 mV
6 Hours 122.6 µA 2.924 V 14 mV
29 Hours 119.6 µA 2.882 V 56 mV
Although these results represent only a single soak experiment, they
allowed confid e nce to b e gained i n the EPRO M PMIC s en sor approach. This was
very valuable, as new PMIC sensor and HCTC chip designs were underw ay i n
parallel with the saline soak testing. A new automated data lo gging test s etup
was also e nvis ioned a n d im plem ent e d sev e ral m o nths l ater f or m o re r eliable
long‐term in ‐situ data collection. Although this new data logging approach was
used f or s ubsequent hermet ic coating t est chips, it could b e en visioned f or f uture
exper i ment s if a f ull saline s oak is n eed ed t o complete hermeti city e xperiments
within a reasonable time frame. The results presented here hav e s h own that
using an EPROM device as a PMIC sensor is likely feasible, and that th e measured
drain current can be successfully calibrated to calculate an as soc i ated f loating
gate voltage.
4.3.1.3 Results Derived from Other HCTC‐1 Devices
Key devices remaining to b e disc ussed on H CTC‐1 are the capacit ors and
int e rdigit at ed elect rod e s, as w e ll as the surfac e profile test structure designed t o
153

determine the passivation l ayer t hickness, the passivation c ut etch depth into the
field ox ides, and the po ssible deter m inat ion of the thickn e ss o f the fiel d oxides.
With regard to the capacitors an d in terd igit ated e lect r o des on HCTC‐1,
two in terd igita t ed e lec t rodes fabricated a t the Metal 1 la y e r w ere desi gned such
that e ither the impeda nce betw ee n the two electrodes could b e m easured to
hopefully detect the presence of moisture, or the two electrodes could b e
electrically connected to form a 1 pF capacitor with the substr ate as t he g round
electrode. Three poly c apacitor s were a ls o created, e ach with a designed
capacitance of 1 pF. One pair of interdigitated and poly capac itors was covered
with oxides only (including both field and passivation oxides), while another pair
of i nterdigitated and poly c apacitors was covered w ith Metal 2 in addi t i o n to th e
field and passivation oxides. Lastly, a third poly capacitor w as covered w ith
Metal 2 and Metal 3 layers i n ad dition t o the field and passivation oxides to act as
barriers to ion contaminatio n .
After fabric ation, t he c apacitance values wer e f ound t o be t oo low to b e
measured w ith the LCR meter tha t w as a va ila b le a t the time. B e cause an H CTC
chip redes ign was planned in paral lel with HCTC‐1 testing, the effort t o measur e
these capac i tors w as a bandoned i n fa vor of t esting t he n ext HCT C design. I n
addit i on, th e int e rd igit ated e l e ctro des wer e m easured to h ave a h igh open c ircuit
impedance and no c hange w a s observed i n the pres ence o f ad ded s urface
moisture. T his was h ypothesized t o be d ue t o the fact t hat the e lectrodes were
buried under an oxide layer with no air cavity to allow for dir ec t moisture
154

detection. This issue was addressed in a later design by bring ing t h e electrod es
up to the surface o f the c hip.
Lastly, a surface profile test s t ru c tu re w as i nc l u de d in o rd er to det er mine
the relat i v e t hickness es o f the CM OS l ayers fa bricated i n the O N Semiconductor
C5 process, which has implications for the interpretation of PM IC sensor results.
Namely, knowing the exact th icknesses o f t h e f ield o xid e s, p ass iv atio n oxide, a nd
metal layers c an a llow for hypotheses to be formed regarding diffusion rates and
decay sensitivities of th e PMIC sensors.
Figure 62. C ros s s ec tion o f the ON S e m icon duc t or C 5 CMOS p r o ces s laye r thi c kne s s e s ti m a ti on
st ruct ure wit h rela t ive t h i c knes ses me asu r ed using a Dektak pr o fi lo m e te r.
The profil e structure w a s measured u sing a D ektak pro f il ometer under a
cleanroom e nvironment, and a schematic drawing of t he c ross‐sec ti onal p rofi le
is shown in Figu re 62. As can be seen in th e drawing, when mov in g from a n ar e a
with o nly oxides t o an a rea with a ll oxides p lus all conductive layers (3 metals
and 2 polysilicon l ay ers), a r e lative i ncrease of 0. 95 µm w as m easured, whi ch
Estimated Metal Thickness
0.2 Microns
155

indicates th at e ach conductive (metal a nd p olysilicon) layer is approximatel y 200
nm t hick o n avera ge. F urthermore, the combi natio n o f th e passivation oxide and
the top field oxide w a s measured t o be a pproximately 2 µ m ( 1.95 µm) thick. The
spacing between the to p of the Me tal 3 layer and the top of the Metal 2 layer was
also measured to be approximately 2 µm (2.09 µm) thick. Lastly , because the
Metal 3 layer is designed to be an etch stop for the passivation cut, i t was
u nkno wn h o w deep th e p assi vat i o n c u t wo u l d go in th e a b senc e o f a metal layer.
The measurements o f the test s tr ucture s how that t he p assiva tio n cut extend s
0.77 µm below the top of the Metal 2 layer, but does not reach the Metal 1 layer.
This also has implications on how high the PMIC floating gates can be p laced
without unintentio nally exposin g t h em to the ext e rn al en vironme nt.
4.4 Second Generation Hermetic Coating Test Chips
The second generation h ermetic coa ting t es t chips (HCTC‐2) a dd r ess
several of the shortcomings disco vered from experiments with HC TC‐1, and als o
add additio n al t est stru ctures t o fu rther the understanding of positiv e m obile ion
contamination detection. Issues identified with the clock circ uits f rom HCTC‐1,
including insufficient output b uf f e ring a nd o utput frequencies not desig n ed t o
drive ultra‐low‐power commercial CMOS i mage s ens o r arrays h ave been
address e d in a s ep ara t e new chip d esig n labeled “CMOS Clock T es t Chip” (CCTC) .
Splitting the clock designs onto a separate chip allowed the si ze o f t h e previous
problematically s mall c apacitors to b e increased on H CTC‐2 to a chieve
156

capacitance values that were well within the measurement capabi lity of th e lab
equipment. D ue t o tig h t MOSIS fa brication sc hedules, several e l e ments of t h i s
design had to be submitted for fabrication before all lessons w er e learned from
the HCTC‐1 c hip. T herefore, several issues that were identifie d la te i n the testing
of HCTC‐1 are also present in the design of HCTC‐2. Despite th is, HCTC‐2 p roved
to be an invaluable learning tool to evolve the designs further i n the third and
fourth gen eratio n herm etic coat i ng t est chips.
4.4.1 Hermetic Coating Test Chip 2 (HCTC ‐2)
The second gen e ra tion h ermetic c o atin g test c hip (HCTC‐2) f eatu res four
EPROM‐based PMIC sensors with varying levels of overlayer protection using
spare CMOS process layers. Each PMIC sensor is identical to th e PMIC s enso r
developed in HCTC‐1, but has been placed under varying levels o f pr otection t o
determine t h e relat i ve d ecay r ates, and therefore rela tiv e p rot ectio n a fford ed i n
each case. Also included on HCTC‐2 are redesigned versions of the inter d ig ita t ed
electrode experiment f rom HCTC‐ 1 i n or der t o d etect moisture c o ntami nati o n. I n
this d esig n, w hen the electrodes a re e lect ric a lly connected, th e capacitance is
desig n ed t o be 25 pF, which is w ell above the measurement capab ilities of t he l ab
equipment. The design was fabricated on March 24, 2010 (run nu mber T 9BM )
again using the MOSIS service and the ON S emiconductor C 5 process. A full set
of m easured process parameters a ft er f abrication i s provided i n Appendix A. The
157

layout and photomicrograph of the fabricated chip are shown in Figure 63. A
summary of all key devices that w ere tested f ollows in Tab l e 15.

Fi gure 6 3 . T he C ade n ce l ayou t (T op ) a nd an o p t i c al m i c ro gr aph ( Bot t o m ) of H e rmeti c C oa ti ng
Test C hip 2 ( HCTC‐2).
158

Ta ble 15. Summa ry o f k e y devices tha t w ere t e sted o n hermet ic c o a ti n g t e s t chi p H CT C ‐ 2 ,
including size, loc a ti on, an d pr esence of a pa ssivat ion oxide c ut f or ea c h device.
Device
No.
Device Type Description / Dimensions
Oxide
Cut?
Location
(x, y in µm
from bottom ‐
left corner)
1 Cap a ci tor
Interdigi t a t ed polysilic on electro des wi th 72
f i ng ers of size 9 µm 552 µm wit h 4.5 µm ga p s a n d
a de si gn c a pa c i t an ce o f 2 5 pF
Yes 628, 82 2
2 Cap a ci tor
Ident i ca l to D evice 1 but with a Metal 2 o verlayer
and Me tal 3 o v e rlaye r ac ti n g a s a he rme t i c shi e l ds
No 1384, 8 22
3 PMOS P MIC
P‐ Cha nnel EPR OM t o ac t as a po s i t i ve mobile io n
con ta mi na ti o n se n sor wi th a p oly fl oa ti ng g ate are a
of 30 µm
2
a nd a c hannel of s ize 9 µm 3.6 µm
Yes 430, 15 8 4
4 PMOS F ET
Ident i ca l to D evice 3 but pi nned ou t as an FET for
calibra t ion of t he PMIC sens or
Yes 404, 15 8 4
5 PMOS P MIC
P‐ Cha nnel EPR OM t o ac t as a po s i t i ve mobile io n
con ta mi na ti o n se n sor wi th a p oly fl oa ti ng g ate are a
of 30 µm
2
a nd a c hannel of s ize 9 µm 3.6 µm
No 830, 15 8 4
6 PMOS F ET
Ident i ca l to D evice 5 but pi nned ou t as an FET for
calibra t ion of t he PMIC sens or
No 804, 15 8 4
7 PMOS P MIC
P‐ Cha nnel EPR OM t o ac t as a po s i t i ve mobile io n
con ta mi na ti o n se n sor wi th a p oly fl oa ti ng g ate are a
of 30 µm
2
, a Met a l 2 overla yer shi e ld, an d a ch anne l
of s ize 9 µm 3.6 µm
No 1230, 1 584
8 PMOS F ET
Ident i ca l to D evice 7 but pi nned ou t as an FET for
calibra t ion of t he PMIC sens or
No 1204, 1 584
9 PMOS P MIC
P‐ Cha nnel EPR OM t o ac t as a po s i t i ve mobile io n
con ta mi na ti o n se n sor wi th a p oly fl oa ti ng g ate are a
of 30 µm
2
, Me tal 2 and Metal 3 overlayer shields,
and a ch annel of s ize 9 µm 3.6 µm
No 1630, 1 584
10 PMOS F ET
Ident i ca l to D evice 9 but pi nned ou t as an FET for
calibra t ion of t he PMIC sens or
No 1604, 1 584
11 PMOS P MIC
P‐ Cha nnel EPR OM t o ac t as a po s i t i ve mobile io n
con ta mi na ti o n se n sor wi th a p oly fl oa ti ng g ate are a
of 17 µm
2
a nd a c hannel of s ize 6 µm 2.4 µm
No 233, 33 9
12 PMOS P MIC
P‐ Cha nnel EPR OM t o ac t as a po s i t i ve mobile io n
con ta mi na ti o n se n sor wi th a p oly fl oa ti ng g ate are a
of 5 µ m
2
a n d a channel of siz e 3 µm 1.2 µ m
No 1766, 3 39
159

A close‐up photomicrograph of the reg i on i n th e top left corner of the chip
is shown in Figure 64, including two matched pairs of PMIC sens ors and
corresponding PFETs. On the left of this photo is a region whe re a passivation
oxide cut has been placed, w hich i s expect ed t o extend d own pas t the Metal 2
layer as d etermined by t he HCTC‐1 layer thic kness test s tructure. In each PMIC
sensor design for HCTC‐2, the floating gate is fabricated in th e polysilicon layer,
and so i s no t ev er e xposed d ir ectly t o t he e x t er n a l env i ronm ent. By inco rporating
this oxide cut, more ion contamination should reach the sensor with the cut as
opposed to t he a djoining s enso r without the cut, s o that a r ela tive positive mobile
ion accumulation rate c a n be measured.
Figure 64. Optical micrograph of two as‐fabricated EPROM PMIC s ensors w ith correspon ding
cali bra t i o n PF ETs on t he s e c on d ge ne rati on he rme ti c c o a t i ng te s t ch ip ( HCTC‐2), s howing o ne
with a supervening passivation oxide cut region (Left), and one wit h a f ully i nt act pa ssivat io n
oxi d e re gi on ( Ri ght ) .
160

To the right of the oxide cut PMIC experiment is another region w ith all
oxides intact, including the passivation oxide. Further to the right on the chip
(not shown) is a region that is covered with a Metal 2 shield layer in addition to
all of the oxides. Lastly, to the far right of the chip (also not shown) i s a region
with yet another PMIC sensor and corresponding PFET that is in this c ase covered
with b oth Metal 2 and Metal 3 shield layers in addition to all of the oxides. In this
way, the relative rates of ion a ccumulation should decrease t o the rig h t as m ore
protection layers are added. Indeed, initial tests showed that the initial charges
on the PMIC sensors that were protected with additional metal a nd di electri c
layers decayed at slower rates as compared to the decay rates for u nprotected
PMIC sensors when the chips were contaminated. Contamination wa s induced by
placing a drop o f saline s olution on t he s urface o f the chip, w hich was then baked
at a h igh tem p eratu re ( 180 °C) for over 200 hours to acce lerate ion diffusio n and
thereby in it iat e a n accelerat e d lifet i me t est. T his tempera t ur e was chosen t o be
below the maxi mum recommended t em perature f or t he c hip after MO SIS
processing, as well as for its ceramic package. The overall de cay of t he f loating
gate voltage measured was not as large as desired over the time t ha t the chips
were a t elevated t emperatur e , a n d t he n ex t g e nerat i on H CTC des ign w as a ltered
to elevate the floating gate closer to the surface so that the effects of i onic
contamination could be mo re eas ily observed .
The int e rdigita t ed e lec t rodes wer e o nce again unfortun ately bur ied too
deep ( at t he l evel o f the polysilicon l ayer ) to s ee any effects of moisture
161

contamination. This problem with th e d es ign w as re al iz e d to o l ate in th e t esti ng
of HCTC‐1 to be incorporated into the design of HCTC‐2. However, the larger
capacitances were in fact measureable, with values within 10% o f their designed
value. The I‐V curves of the capacitors were taken after each soak t est and were
determined t o no t be a ffected by t h e s oak i ng. T his observ ation may also be due
to t he f ac t that t he c ap acitor e lectrodes ar e buried t oo d eep, and all future c hip
desig n s inc o rporated h igher metal layers fo r nearly every test stru cture.
4.4.2 CMOS Clock Test Chip (CCTC)
The CMOS c lock t est chip ( CCTC) was fabricated concurrently wit h HCTC‐
2 using the MOSIS service and th e ON S emiconductor C 5 process ( run T9BM). A
full set of m easured pro c ess parameters after fabrication is provided i n Append ix
A. A s described previously, the firs t hermetic c oatin g t es t ch ip ( HCTC‐1) was key
to proving the viability of a low‐cost, simple ring oscillator that could b e designed
to be tolerant to fabrication processing variations. However, HCTC‐1 h ad t hree
shortcomings in the CMOS clock designs that were addressed in t he s econd
generation test chip:
1. Insufficient output buffering caused the measured output clock
waveforms to be dis t or ted when t h e outputs w ere load ed w ith the
high c apacitance o f the oscillo scope probe tip. As a consequen ce,
additional o utput buffering w as included at all clock o utputs.
162

2. In g eneral, the frequen c ies produc ed b y CMO S r ing oscillators a re
too high f or t he c lock rates c ha racteristic of m iniature C MOS i mage
sensors (e.g. , 5 to 10 MHz). B ecause i t was shown in t he H CTC‐1
CMOS clock designs that division by 2 using a D‐flip‐flop is ef fect iv e
in halving the clock rate, and also does not apparently affect the
oscillator performance, a ddition al d ividers were a dded to p rodu ce
the option of even lower clock f requencies, d o wn to 6.25 MHz.
3. The clock o scillators o n HCTC‐1 w ere d e sig n ed a nd s imulated a t
room t emperatur e ; op erat ion at a h igher temperature (37 °C)
when i mplanted i n the body w ill cause a mea s ureable shift in t h e
clock f requency. A s a consequence, the cl ock c ircuits were
redes i gn ed f or higher t e mperatur e operatio n.
In o rde r to b e ab l e to t est im p ro ve m ents i n th e c l o c k desi gns m ore easily,
the clock c ircuits were i solated onto a separate chip design wi th t he d esig nation
“CMOS Clock T est Ch ip” (CCTC) a s s h own in F ig ure 65, w ith a sum mary o f the key
CMOS clock devices that we re tested in Table 16.

163

F i g ure 65. T he C ad ence l a yout (L eft ) a nd a n opt i ca l microg ra ph (Right) of t he C MOS Clock Test
Chip (CC TC).
164

Ta ble 16. Summa ry o f key devices t h at were t e st e d o n CMOS c lock t est chip C C TC, i ncludin g s ize,
loca tion, an d presence o f a pas s iva t ion ox ide cut f o r ea c h devi ce.
Device
No.
Device Type Description / Dimensions
Oxide
Cut?
Location
(x, y in µm
from bottom ‐
left corner)
1 CMOS Cl o ck
Seven sta g e 1 60 µm 60 µm ring oscilla tor
de si gne d wi t h an o ut put fre q ue ncy of 2 5 MHz at an
a mbient t e mpera t ure of 25 °C a nd wi t h two D‐F l ip‐
Flop o u t put s t ages f or a ddit i o na l 12.5 MHz a n d
6.25 MHz output s
No 560, 72 5
2 CMOS Cl o ck
Simila r to Device 1 but m odi fied f or a 10% slower
outpu t f re que n cy wi t h o ut c hangi n g the si ze
No 780, 47 7
3 CMOS Cl o ck
Simila r to Device 1 but m odi fied f or a 10% fa s ter
outpu t f re que n cy wi t h o ut c hangi n g the si ze
No 1337, 4 77
4 CMOS Cl o ck
Seven sta g e 1 60 µm 60 µm ring oscilla tor
de si gne d wi t h an o ut put fre q ue ncy of 2 5 MHz at an
a mbient t e mpera t ure of 37 °C a nd wi t h two D‐F l ip‐
Flop o u t put s t ages f or a ddit i o na l 12.5 MHz a n d
6.25 MHz output s
No 560, 12 7 3
5 CMOS Cl o ck
Simila r to Device 4 but m odi fied f or a 10% slower
outpu t f re que n cy wi t h o ut c hangi n g the si ze
No 780, 15 3 2
6 CMOS Cl o ck
Simila r to Device 4 but m odi fied f or a 10% fa s ter
outpu t f re que n cy wi t h o ut c hangi n g the si ze
No 1337, 1 554
7 CMOS Cl o ck
Seven sta g e 1 14 µm 30 µm ring oscilla tor simila r
to D e v i c e 1 b u t to t e s t mi ni mum p o we r d i ssi pa ti on
No 1745, 1 446
To a ddress the issue of i nsuffic ient o utput buf f e ring, addition al b uffe r i ng
was add e d in t he f orm of i nc reasingly larger i nverter gates, su ch t hat the input
capacitance is increased at each stage following the ring oscillator, but not so
much as to cause the waveform to be corrupted. Each inverter t hen reproduces
th e invert ed signal, and passes th e wave form to th e next larger i nv er ter until the
165

output resistance is sufficiently low to drive a large (external, in the case of the
oscilloscope probe) c apacitance s uch that t he R C time constan t is l ow, and fas t
switching of t he o scillator output c an occur. T he i ncreased c a pacity o utput
buffers are shown on the CCTC rin g oscillator schematic in Figu re 66.
Fi gure 6 6 . S che m ati c d i a gra m o f the ri ng‐ o s c i llat o r‐ b a se d clo c k d esign con t aini ng a 7 ‐stage r i n g
oscilla tor, wit h inverters an d delay cells d esigned t o s lo w d o w n the fr equency of oscillation. The
2 5 MHz out p u t i s pa sse d thr o ugh a tw o‐ st age D ‐ fli p ‐ f lop to h alve t h e f r equency t w ice, a nd o ut pu t
ta ps a re pr o vi ded for each f requency division s t ep.
The labels “ INVX1” a nd “ INVX4” i ndicate increasingly l arger inverter
desig n s (gr e ater c apac itanc e b ut a lso greater drive curr ent) a s compared with
the inverters employed in the ring oscillator. To help alleviate the capacitive
loading when measuring the output with an oscilloscope, an acti ve p robe w ith a
capacitance of less than 1 pF was used. In order to broaden th e use o f s uch a ring
oscillator to l ower f requency a ppl ications, for exam ple to d riv e th e oscillator
inputs of a CMOS image sensor, additional D‐flip‐flops (DFFs) w ere added to a ct
as divide by two elements, and are also shown in Figure 66 as b locks labeled
“ D F F N E G X 1” . Th e b asel ine 2 5 M H z ring o sc i l l at o r c i rc u i t fr o m H CTC‐1 was used
166

as t he core oscillator, a nd t he t wo D FFs s erve t o produce two a dd itio nal outputs
at 12.5 MHz and 6.25 M H z.
One such candidate image sensor to be driven by this ring oscil lator design
is the OmniVision (Santa Clara, CA) OV6920 that has an oscillat or d esign
frequency o f 6.14 MHz. Although a d vised by the manufacturer th at t he use of an
external clock was not possible, the crystal oscillator that dr ove the Pierc e
oscillator circuit within the image sensor was desoldered and repl a ced with a
bench‐top function generator. Both sine and square waveforms w ere injec t ed
into t h e c rystal o scillator terminal s, and the image sensor con tinued t o functio n
normally. A s the voltages a nd f r e quencies w ere chan ged, t he i m age senso r
continued t o function o ver a subs tantial rang e for each var iabl e. F igure 67 shows
a summary o f the experimentally v erified o p erat ing ra nges f or b oth sine a nd
square w aveforms. Interestingl y, the image sensor i s operatio n al o ver a
surprisingl y w ide range of voltages and frequencies. A lthough t h e maxi mu m
voltage shown on the figure is 2.50 V, the sensor continued to operate normally
up t o a maxi mum tested voltage o f 10.0 V. Because the C C TC r i n g oscillator is
de sig ned to b e a sq u are w ave o p e rati ng at 6. 25 M H z, and th e im a ge s ensor clock
input voltage can apparently be adjusted over a wide range of v alues, the OV6920
image sensor proved to be a good candidate for testing in conju nction w ith the
CCT C CM OS clocks.
167

F i g u re 67. Allowa ble f re q uency va ri a t ions f or b ot h squa re w a v e i n put si gn als ( Blue Bars ) an d
sine wave input signals (Red Bars) centered at a nominal input f requency of 6 .14 MHz, a s a
func ti on o f th e pe ak‐ t o‐ pe a k e x t e rnal clo c k si gn al a m p li tude ( V pp ) f o r t h e OmniVision O V6920
CMOS im a ge sensor.
Lastly, to a ddress the issue of a r oom t emperature o scillator desi gn
changing o utput frequency signif icantly when o perated at b ody t emperature, a
second complete set of r ing oscillators w as d esigned, s imulated , and itera t iv ely
redes i gn ed f or a n oper ation a l tem p erature of 37 °C, which is the widely accepted
normal h uman b ody temperature. T able 1 7 summarizes the res u lts of the
simulation for each of the three design variants, each operated a t their respective
design tem peratures, at all three cl ock output frequencies.
168

Table 17. Simulated frequencies for the six variants of the rin g‐os cillat o r‐based C M OS c l o c k
o sc i llat o rs, a s implement ed in a 0.5 µm CMOS pr o c e ss a nd o pe ra t e d a t 5 V.
Design
Temperature
Design
Variant
Simulated
25 MHz Output
Simulated
12.5 MHz Output
Simulated
6.25 MHz Output
25 °C
Slow 22.52 MHz 11.26 MHz 5.630 MHz
Nomin a l 24.56 MHz 12.28 MHz 6.139 MHz
Fa st 26.60 MHz 13.30 MHz 6.650 MHz
37 °C
Slow 22.52 MHz 11.26 MHz 5.630 MHz
Nomin a l 24.59 MHz 12.29 MHz 6.147 MHz
Fa st 26.78 MHz 13.39 MHz 6.695 MHz
After fabrication o f t he C CTC chip w as completed, the LabV iew
measurement of c lock j itter was perf ormed agai n, t hi s time w i t h significantly
improved r esults. N ote that a lthough the nominal lowest c lock fre q uency was
designed t o be 6.25 MHz, the simulated frequency was 6.14 M Hz d ue t o updated
MOSIS SPICE parameters. Because the OV6920 image sensor is des i gne d f o r a
clock f requency o f 6.14 M Hz, the desig n i s s e rend ipitous l y idea l for this sensor .
Figure 68 shows the output summary screen capture from the LabV ie w
experiment, showing the output w aveform (top), t he d eviation o f the period of
the wav e for m f rom the aver age per i od i n n a no seconds a s a funct ion of time ove r
an extended collection period (bottom left), and the histogram of w aveform
period deviations over the co llection period ( bottom right).
The measured d evi c e was the as‐f abricated no minal 6.2 5 M Hz f req uency
output f rom the room t emperature c lock, which was measured t o h ave a mean
frequency of 6 .01 M H z and a dete rm inis t i c period j itt e r of 1 . 2 4 ns. This
corresponds t o a deterministic va riation betw een 5.99 M H z and 6 .03 MHz, which
169

is w ell within t he o perational b and for the OV6920 image sens or. The only
obvious negative effec t of the added buffering is the apparent ringing that c an b e
seen in Figure 68, likely caused by the high capacity drive current coupled with
the low capacitanc e probe tip. This p heno menon was not obser ve d during
simulation and was an unanticipated resu lt.
F i g u re 68. Screen c apt u re o f t he L a b V iew GUI a pplica t ion for t h e mea s urement of clock j itt e r f o r
the CMOS Clock Test Chip. The waveform is shown (Top) in addit i o n to p e r iod deviati o n
mea s urements o ver t i me w i t h resp ec t to the a verag e pe rio d (Bott om Left) and a h is to gram of all
period deviati o n mea s urem ents (B o tt om R ight ).
With the confidence of a jitter, mean frequency, and voltage that are all
compatible with the OV6920, the b ench‐top f unction generator wa s replac ed
with the 6.25 MHz CCTC ring osci llator output i n order to s uccessfully drive the
170

image s e ns or i nternal oscillator circuitry. Figure 6 9 includes a video frame
capture of t he output from t he O V6920 image sensor t aken d ir ect ly f rom an
OV6920 development board, which shows the successful imaging of a color chart
in front of a test target. Col or operation was not possible due to the strict timing
requirements set forth by the NTSC standard, but there is no re aso n t o doubt that
color imaging would be possible with a digital image sensor. A lso of n otable
importance is that the captured i mage is der ived f rom the OV692 0 image sens or
used in conjunction with a custom polymer lens that was designe d and fabric ated
for the intr aocular came ra (IOC) previously.
Fi gure 6 9 . ( Le f t ) Ex pe ri me ntal set up co n t a ining a MOSIS‐fa brica ted CMOS di g it al clo c k b ase d o n
a mult i‐stag e ring o scilla tor d e si g n. T he IOC l ens a n d i m ag e s ensor a re held i n a mount t o t he
right of the power supply on the optical table. (Right) Acquired image from an OmniVision
OV6920 ima g e sensor a rray coupled with a n IOC custom p olymer l ens optical system, with the
OV6920 ima g e sensor a rra y d riven by t he CMOS d i g i ta l clock .
A lt houg h t h e pr i ma ry goa l of t he CC T C chi p wa s t o show tha t a C MOS ring
oscillator capable of b eing i nteg rated with b ioimplantable devi ces could be m ade
very simply, final tweaks to the design to improve the ringing caus ed b y over‐
buffering was explored with the HCTC‐3 design described next, i n wh ich the clock
was again included with other te st sensors on the same integrat ed c ir cuit chip.
171

4.5 Third Generation Hermetic Coating Test Chips
As a key issue with both HCTC‐1 and HCTC‐2 chip designs is the low
observed s ensitiv i ty o f the PM IC d evic es w ith polysilicon f loat ing gates to
positiv e m obile ion c o ntam in ation, t he t hir d g en er atio n herm et i c coating t est
chips (HCTC‐3 an d HCTC‐4) we re d eveloped t o explore the des i gn sp ace in t erm s
of floating gate size and vertical layer placement. The overal l goal o f the th ir d
generation d esig ns w as t o find a n optimal s e nsor d esig n that p rovides good
sensitiv ity t o p ositiv e m o bile i on contamination, and to a lso g ather data o n the
relat i ve d ec ay r a t es o f v a rious thick n esses and numbers of s par e ox ide and met a l
thickness p r otectio n l ayers. I n ad ditio n , HCTC‐4 ( fabric ated j ust a f ter HCTC‐3
and still considered to be a “third generation” design) explore s a range o f
interdigitated e lectrodes and c a pacitors t hat are placed o n the s urface o f the chip
in order to finally observe the e ffec ts of moisture accumulatio n.
4.5.1 Hermetic Coating Test Chip 3 (HCTC ‐3)
The third h e rmet ic coating t est ch ip ( HCTC‐3 ) features a p arallel d esign
approach t o determin e which of n ine differ en t PMIC s enso r geome tries will prove
to b e the most s ensitiv e t o ion con tamina tion. T he f loating gates of the various
sensors ar e placed at differen t v ert i cal lev e ls w ithin the process metal stack, and
several different floating gate s iz es are i ncorp o rated as w ell.
172

Fi gure 7 0 . The Cade n ce l ayou t (T op ) a n d an o p t i c al mi c ro gr aph ( Bot t o m ) of H ermetic C o a t ing
Test C hip 3 (HCTC‐3).
173

In addition, this chip features an experiment in which lateral diffusion of
mobile i on contamination is test ed b y placi n g one PMIC sensor a t th e corner of a
lateral d i ffusion interface and a second PMIC s ensor at t he c en te r, i n hope s of
observing a different i al d ecay r at e between the two and t h ereby d etermining t he
rate of lateral diffusion. Finally, a revised CMOS clock circu it i s also i ncluded on
HCTC‐3. T h e d es ign was fabric ated o n October 1, 2011 (run n umber V15L), again
using the M OSIS service and th e ON S emiconductor C 5 0.5 µm p roc e ss. A fu l l se t
of m easured process parameters a ft er f abrication i s provided i n Appendix A. The
layout and photomicrograph of the fabric ated chip are shown in Figu re 70.
The ring oscillator design on HCTC‐3 was again improved over it s
predecesso rs i n terms of w avefo r m quality and output b uffer i ng. The clock was
designed to have its own separate power and ground rails to red uce the effect s
of any other devices on the same chip, and the output buffering was slightly
reduced in s ize and m o ved farther away f ro m the clock circuit to redu ce the
unwanted ringing effects and excess noise. After the fabrication of HCTC‐3 was
completed, a LabView measurement of c lock j itter was performed again, a nd
signific antl y better r es ults w ere o b tained a s compared with t he j i tte r obse rved
with the clock circuits on the C CTC chip. O nl y the nominal clo ck c ircui t vari ant
was fabricated o n HCTC‐3, and no t ransistor sizing w as c hanged. Because of
process var i at ions, the frequency (desig ned to b e approximately 6.25 MHz) w as
signific antl y faster o n HCTC‐3 a t 6.79 MHz. F igure 71 sho w s th e output summar y
screen c ap t u re f rom the LabView exper i ment, showin g the output waveform
174

(top), the deviation of the period of the waveform from the ave rag e p eriod in
nanoseconds as a f unction of t ime over a n extended collection p eriod (bottom
left), and the histogram of waveform period deviations over the collection period
(bottom right).
F i g u re 71. S creen ca p t u re o f t h e La b V iew GUI a pplica t ion f o r t h e mea s urement of c l o ck j itt e r for
the thi rd he rme t i c c o a ti n g t e s t chi p ( HC T C‐ 3 ). The w av e f orm i s shown (Top) in addition to period
deviation measurements over time with respect to the average pe riod (B ot to m Lef t ) and a
histo g r a m of a ll period devia t ion me asure m ents ( Bo t t o m Rig ht).
The measured device here is the nominal 6.25 MHz frequency outp ut f rom
a clock generator d esigned to o pe r a te a t room t empera ture, whic h was measured
to h ave a m e an f requency o f 6.79 M H z and a deterministic period jitter of 2.20 ns.
This corresponds t o a determinis tic variation between 6. 74 MHz and 6.84 MHz,
175

which is well within the operational band for the OV6920 image sensor. This
amount of period and frequency jitter may be unacceptable for m any applications
that require high‐precision coh erent clocks, but CMOS image sen sors apparently
work well under these conditions. In this case, the clock cent ral frequency is
outside the band o f operation fo r the OmniVision O V69 2 0 C MOS im age senso r
array. As mentioned previously w ith regard t o the clock circui ts f a b ricated on
the CCTC chip, a certain amount of variation is to be expected due to f abrica tion
tolerances, and this can be compensated for by the incorporation of slow, normal,
and fast clock circuits, the most optimal of which can be selec ted post‐fabrica tion
and then s witched in b y a multiplexer circuit.
The overall waveform s hape i s greatly impro ved a nd m uch closer to a
square w aveform as compared with t he c lock c ircuits on t he CCTC chip, although
there ar e s t ill some r ingin g a rt ifa c ts, and the highest voltage a t the waveform
leading edge reaches nearly 7 volts. When looking at the period histogram for
this c lock c ircuit, two distinct Gaussian distrib u tions can be clearly s een, and th e
differ ence b etween t h e p eaks is v er y easy t o discern. T his ind ica t es t hat m u ch o f
the nond et ermin i stic c lock j itter, s uch as i nter fer e nce from o t her devices or
additional power supply noise, has been removed. Further refinement of t he
clock gener ator w ill be t abled for future des i g n s, as an ad diti o n al r ed esig n based
on the results of this set of experiments is expected to produc e an e ven mor e
stable output.
176

Based on the experiments with varying levels of protection inco rporated
in HCTC‐2, a new set of PMIC sensors was designed to explore a wide range of
device configurations, a nd t o de ter m ine the effects of f loating g ate depth and size
on device sensitivity. A summary of all key PMIC devices that were t ested is
deta iled in Table 18.
Ta ble 18. S umma ry o f key PMIC s ensors t hat were t est e d on h erme t i c co at ing t e s t c hip H CTC‐3,
including size, loc a ti on, an d pr esence of a pa ssivat ion oxide c ut f or ea c h device.
PMIC
No.
Gate
Layer
Gate
Area
Channel Size
(W L in µm)
Oxide
Cut?
Experiment Set
Location
(x, y in µm from
bottom ‐left corner)
1 Metal 2 108 µm
2
9 µm 3 µ m Yes Varying G a te/Protecti o n 750, 12 0 0
2 Metal 1 108 µm
2
9 µm 3 µ m Yes Varying G a te/Protecti o n 850, 12 0 0
3 Pol y 108 µm
2
9 µm 3 µ m Yes Varying G a te/Protecti o n 950, 12 0 0
4 Metal 3 30 µm
2
9 µm 3 µ m Yes Varying G a te/Protecti o n 750, 11 0 0
5 Metal 3 108 µm
2
9 µm 3 µ m Yes Varying G a te/Protecti o n 850, 11 0 0
6 Metal 3 432 µm
2
9 µm 3 µ m Yes Varying G a te/Protecti o n 950, 11 0 0
7 Metal 3 108 µm
2
9 µm 3 µ m No Varying G a te/Protecti o n 750, 90 0
8 Metal 2 108 µm
2
9 µm 3 µ m No Varying G a te/Protecti o n 850, 90 0
9 Metal 1 108 µm
2
9 µm 3 µ m No Varying G a te/Protecti o n 950, 90 0
10 Metal 3 108 µm
2
9 µm 3 µ m No La te ral D i f f usi o n
1150, 1 200
(100 µm f rom cut)
11 Metal 3 108 µm
2
9 µm 3 µ m No La te ral D i f f usi o n
1250, 1 300
(200 µm f rom cut)
12 Metal 1 108 µm
2
9 µm 3 µ m No
La te ral D i f f usi o n
Met a l 1 + Me ta l 2 Shield
1150, 8 00
(100 µm f rom cut)
13 Metal 1 108 µm
2
9 µm 3 µ m No
La te ral D i f f usi o n
Met a l 1 + Me ta l 2 Shield
1250, 7 00
(200 µm f rom cut)

177

A suit e of n ine d e vic e s was a rra ng ed i n a gr id n ear the center of t he c hip,
with various sensors designed to have floating gates fabricated in different metal
layers, and with varying gate sizes for the top‐ most s enso rs ( PMICs 1 through 9).
A summary o f the experiments fo r thes e n i n e k ey P MI C sensor s is shown i n a
cross‐sectio n illustration i n Figure 72. For the three sizes o f floa ting g ate s
indicated in the diagram, the actual design sizes are 3 by 9 mi cr ons for the
smallest sensor, 6 by 18 microns for the medium sensor, and 12 by 36 microns
for the larg est sensor. A m ore comprehensiv e m odel f or s ensiti vity scaling wil l
be presented in Chapter 5, but the basic idea is that as th e fl oating g ate becomes
larger, the c o llection ar ea t o d e tect p ositive mobile i ons incr eases prop ortionally;
the over all capacita nc e also i nc r e ases, but to a l esser ext e nt. Several PFET
calibration devices were a lso in cluded w ithin the pass ivatio n c ut a nd u ncut
regio n s. K eepin g in m ind the previous r es ults f rom the layer t hickness test
structure a n d the D e ktak s ca ns, it i s expected t ha t t h e sensor s with the
passivation o xide c ut covering th e area w ill likely h ave Metal 3 and Metal 2 layers
exposed to the external environment, while other metal layers, or a ny l ayers with
the passivation layer intact, will not be directly exposed. A photomi c rograph of
the as‐fabr i cated H C TC‐3 t est region s chematically d epicted in Figure 72 is
shown in Figure 73. The boundaries of each test area defining the various
exper i ment s are outlin ed w ith Metal 3, a nd c an b e s e en a s whit e borders in the
photograph. The top six sensors in this case each have an oxid e cut region o ver
the area, w h ile the bottom three s e nsors do n ot.
178

Fi gure 7 2 . C ros s s e c ti on i llustrati on o f the ni ne P MIC se nso r e xperi ment s o n t he t hird hermet i c
coating test chip (HCTC‐3). The top and middle groups have passivation oxide cuts, while the
bottom group has the passivation layer intact. The numbers abo ve e ach sensor s h o w the
e s ti ma te d thi c kne s s of o xi de c ove ri n g t h e se nsor f l o a t i n g g a te . Th e flo a ti n g g a t e me talli za ti o n
layers are Metal 1 (dark blue), Metal 2 (maroon), and Metal 3 ( lig h t blue). Via s w ere used t o
int e rconnect the va rious me t a llizat i on levels f o r a g i ven PMIC se ns or f l o a t i n g g a te .
179

Fi gure 7 3 . P ho tomi cro g r a ph o f the ni ne P MIC se ns ors on t he t hi r d hermetic c oa ting t es t ch ip
(HCT C‐3) t ha t were incorp o rate d to mea s ure the p osi t ive mobile i o n con ta mi na ti o n se n si ti vi ti e s
of EPR OM P M I C se ns ors wi th v ari o us f l o ati n g ga te d e p ths (i.e. me t alli zati on l aye rs) and si ze s.
PMIC sensors 1 through 9 from Table 18 are a rra ng ed l e f t ‐ to‐right and top‐to‐bottom. The larges t
floa ting g a t e s h own in t he m iddl e r o w a n d righ tm os t column is 12 by 36 microns. The top two
rows h ave a p a ssi v ati o n o x i d e cu t and the b o t t o m r ow h as t he p a ssivat ion oxide inta ct .
180

Looking for a moment only at the top half of the photograph, re call that
both Metal 3 and Metal 2 layers are expected to be exposed by t he o xide c ut.
However, f rom th e ph o tomic ro graph, it is somewhat unclear if this is th e case. A
scanning e l e ctron microscope was used t o capture a more d etail e d micrograph
to l ook a t these metal layer exposures (as shown below in F igure 74). The area
shown is o f the oxid e cut “m edium ” s ized P M I C sensor w ith a Met al 3 f loating gate
as c an b e seen i n the central experimental r egion of F igur e 73, adjacent to a PFET
calibration device to its right. The floating gate structure c an c learl y b e seen t o
be above the surface o f the top field oxide r e m a in ing a f t e r the passiv a tion etch.
Figure 74. S cannin g e lectron micr o s co pe ( SEM) m ic ro gra p h of a n expose d HCTC‐3 M etal 3
floating gate. This “medium” sized PMIC sensor is located in t he c entra l e xperiment a l re g i on o f
Figure 73, just to the left of its c o m p a nion P FET c a librati o n device.
181

The small dimples in th e metal s urface are via contacts to conn ect the top
Metal 3 floating gate surface to the Metal 2, Metal 1, and poly silicon f loating gates
below. The Metal 2 traces that a re used to connect the so u rce and drain areas to
the bonding pads are also shown to be visible just above the re m a in ing top field
oxide, which is consistent with the findings from the Dektak pr ofilo m eter s can s .
Some edge roughness can also be seen around the floating gate w here i t is
exposed, and there i s a slight taper to the edges.
A long‐term, accelerated lifetime test system for batch testing of CMOS
VLSI c hips containin g m o isture a nd p ositive mobile i on contamin at i o n (P MIC )
sensors has been constructed. T h e a ccelerated lifetime t est sy stem i s centered
around a B inder FD‐53 high‐temperature, prec ision oven w ith a t hrough‐port for
instrumentation, a s well as a n ES PEC SH‐241 temperature and hum idity cham ber
with 2‐inch cable ports on both the left and right sides. The long‐term testing
system also includes a number of K eithley digital multimeters, multiple K eithley
1 40 computer controlled multiplexers, Agilent computer controlled h igh
precision p o wer supplies, and a data a cquisition compute r, a ll integ r a t ed t hrough
a custom LabView graphical user interface. A custom printed ci rcuit board using
high‐temperature materials and a high‐temperature zero insertio n force (ZI F )
socket allow both rapid wiring of new designs as well as rapid in sert ion and
removal of chips from the oven. A photograph of two custom hig h‐temperature
PCBs with HCTC‐3 chips mounted in ZIF sockets inside of the oven is shown in
Figure 75.
182

Fi gure 7 5 . P ho to gra p h of t w o c us to m h i gh t e mpe ra tu re P CB s wi th HCTC‐3 chips mounted in high
t e mperat ure zero‐insert i on‐f orce sock e t s in t h e Binder FD‐53 pr ecision oven.
This system allows for up to 80 individual PMIC sensors to be f irs t c har ged
and then r ead out continuously for months at a time in a high t emperatur e
environment. H igh‐temperature ce ramic zero‐insertion‐ f orce ( ZI F) s ockets ar e
mounted internal t o th e oven, and are connected t o the electron ic m easuring
equipment through specializ e d f iberglass coated high t emperature wires. A
photograph of the chip in the ZI F socket is sho w n in Figure 76.

Figure 76. P ho to graph of an HCT C ‐3 ch i p in a hi g h temperature ze ro‐insert ion‐force socket .
183

The custom L abView d ata acquisit i o n and analysi s soft w are (Fi gure 77) is
capable of m anaging the substantial amount o f data t hat is coll ected during e ach
experimental run, which can accumulate well over 1,000 hours per chip. The
software can display and store all relevant data in a customiza ble format, and can
continuously monitor a ll electric al m easurement e qui p ment t o sa fegu ard agains t
abnormal PMIC operating conditions. In this way, long term, ac celerated lifetim e
test data can be obtained with much higher reliability and prec ision than w it h
m o re conve ntion a l m e a n s, and in a f ract ion of t he exper im ent e r’ s time.
F i g u re 77. S creen ca p t u re of t h e PMIC t est i ng dat a co nt ro l a n d a c quisit ion La b V iew prog ram.
184

Using the n e w automated tes t s tation, very l on g term e xperiment s began
with an HCTC‐3 chip that was intentionally contaminated with po sitive m obile
io ns ( w i th a dro p o f b u ffer e d sal i ne so l u tio n ad de d to th e c h i p su rfac e ) . B ec au se
the chip was to be electrically measured in the oven at high te mperature, t he
saline s oak technique c o uld not be continuous. Rather, a micro‐pipette was used
to drop a small amount of saline solution onto the center of th e c h ip, and then
after the droplet dried, t he c hi p was inserted i nto the oven, b rought u p to 180 ° C,
and then charged for subsequent testing. A photomicrograph of a co ntaminated
chip after d rying is shown in Figure 78.
Fi gure 7 8 . P ho tomi cro g r a ph o f an H CT C‐ 3 chi p a f t e r con ta mi na ti on with a saline drop that was
subsequent ly dried.
185

First, d a t a was collected f rom th e contamin a ted PMIC s ensors t h at h ad
Metal 2, M etal 1, and polysilicon floating gate layers, as well as a cut in the
passivation oxide layer over the entire test area (Table 18, PMIC 1, 2, and 3). In
this case, the poly and Metal 1 gates remained buried under oxide protection, but
the Metal 2 gate w as e xposed t o t h e exterior e nvironment a s a c onsequence o f
the passivation cut. Figure 79 shows a plot of the measured dr ain current a s a
function o f time d uring the bake a t 180 °C. Normalized drain current is plotted
to m ore directly compare t he c urves, because not every PMIC s en so r charges up
to the same gate voltage and pro d u ces the same initial d rain cu rrent.
Figure 79. P lot of t he n ormalize d dr ain current a s a fu nction o f time f or t hree con t a mina ted PMI C
sensors on HCTC‐3, all with an oxide passivation cut. Cases in cludi n g t h a t o f a M e tal 2 fl oa ti n g
gate (ex po sed to air ), a Metal 1 flo ati n g g a t e , and a poly f loa ti ng g ate ar e shown.
As c an b e seen f rom the plot, it i s obvious that t he P MIC senso r with t he
floating gat e fabrica t ed i n the Metal 2 layer had a nearly i nst antaneo u s discharge
of the floating gate voltage. This phenomenon of immediate dis charge i s
observed even in the case of a PMIC sensor with a Metal 2 gate exposed to air and
186

not contaminated. This i s lik e ly d ue t o the fact t hat th ere ar e a pp roximately
200,000 po sitiv e i ons per cubic ce ntimeter i n the ambient air, as m easured with
an A lphaLab Air Ion Co unter. Th e free i ons will nearly ins tant ly b e drawn to t he
floating gate and neutralize the charge, thereby causing the dr ain c u rrent t o fall
to zero (the normally uncharged state of the PMOS EPROM device). The PMIC
sensor with a Metal 1 gate shows a significant positive ion acc umulation rate,
resulting in a gradual decrease in the normalized drain current caused by the loss
of g at e volt age. T he d ecreas e is m ore rapid initially, w i th t h e ra te o f decreas e
diminishing with time. The more deeply buried polysilicon gate PMIC sensor
shows a further r ed uced p ositiv e ion contamination a ccumulation rate by
comparison, as expected.
Although this normalized drain c urrent p lot closely resemb les a n inv e r t ed
plot o f ion accumulation, and will b e used a s the primary way t o display these
curves, care must be taken to eventually convert the drain curr ent to g ate volta g e
by applying the appropriate calibration I‐V curves extracted fr om the
accompanying PFET d evices, and then t o accumulated charge b y ta king i nto
account t he gate capacitance, w hi ch i s calculated b y multiplyin g the gate a reas
f o r each metal layer by th e corresp onding cap ac itance per unit area values given
for the particular run as listed in Appendix A. Figure 80 show s a plot o f
accumulated charge for the case d i scussed above. Note that at th e measurement
resolution currently being used, ions can be detected at the level of a few hundred
at a time.
187

Fi gure 8 0 . P lot o f t he n umb e r of a c c u mulate d ch a rge s as a f un ct ion of t ime f o r t he case o f
cont a minat e d PMIC s ensors w it h a Metal 1 f l oat i ng g at e (Blue) a nd a p oly fl oa ting g a t e ( R ed),
under condit i o ns of accelera t e d lif e t i me test i ng at 180 °C.
An uncontaminated HCTC‐3 chip (with no saline contamination applied),
was also subjected to accelerated lifetime testing under nearly identical
conditions, with results as shown in (Figure 81). Surprisingly , the number o f
accumulated charges does n ot a pp ear to b e to o different from t h e contaminate d
curves previously shown. This result suggests that there may b e a si gnif i c ant
intr ins i c de nsity o f p ositiv e m o bile i ons w i th in t he f ield a nd passivation o xide
layers that are inherently embedded as part of the CMOS fabrica tion p rocess.
Even i n the pres ence o f add i tio n al s urfac e contamin a t ion, a m aj ority of t h e
overall positive m obile ion accumulated charge is due to the in tr insic
contamina t i on, with a n enhanced c harge diffusio n r at e due to t h e extr insic
contamination. T his does n ot i mmediately exp l ain why the conta mi nated versi o n
of the poly gate curve is lower than the uncontaminated curve, but it w ill be
188

shown in C hapter 5 t h a t the chip‐to‐chip variation i n these mea surements could
easily a ccount f or t he o bserved le vel of d ifferential between t he c urves, especially
in consideration of the fact that the deeply buried poly gate i s far removed fro m
the extrinsic contamination located at the surface.
Fi gure 8 1 . P lot o f t he n umb e r of a c c u mulate d ch a rge s as a f un ct ion of t ime f o r t he case o f
con ta mina ted (D ashe d) a n d u nc ont a min a ted (S olid) P M IC s ens o rs w i t h a Me tal 1 floa ti ng g at e
(Blue) an d a p oly flo a ting g ate (Re d ).
Let us n ow consider the two cases discussed above in comparison with
results from PMIC sensors with Metal 2 and Metal 3 floating gat es f rom an a rea
that did not have a passivation oxide cut (Table 18, PMIC 7 and 8). It has already
been shown that when the floating gate is exposed to the air, a c harge canno t be
held, but with the passivation layer intact, all floating gates will be buried
regardless of the metal layer height. Figure 82 shows a plot o f the norm alized
drain current for the previous cases, but also a dds comparable curves f or t he
PMIC sensors with Metal 2 and Metal 3 gates.
189

Fi gure 8 2 . P lot of t he n o r mali ze d dr ai n current a s a functi on o f ti me f or c on ta minate d (D as hed)
and unc o nt a m inate d ( Soli d) P MIC devices on H CTC‐3, c omp a ring t h e d e ca y of t he d ra in c urren t
for a nu mb e r o f flo a ti n g g a t e con fi gur a ti o n s.
As the top set of four curves has previously been discussed, le t us n ow
focus on t he b ottom set of f our curves, and th eir relation t o the top set of four.
The difference between the bottom set (Metal 2 and Metal 3 gates) and the top
set (Metal 1 and poly gates) is that the top set are passivation oxide cut, but the
bottom are not. In i tially it cou l d b e th o u gh t t h at with o u t th e passivation cut, the
floating gate at the Metal 3 layer should be approximately 2 mi crons below the
surface, which is also approximately the case for the Metal 1 gate with the oxide
cut. However, these two devices behave much differently, with the Metal 3 gate
decaying s ignifican t ly f aster, a nd w ith the differ ence b et ween contami nate d and
190

uncontaminated cases for both o f the senso r s with t he p assiva ti on l ayer i ntact
seemingly statistically insignificant. The answer to the latte r observation could
lie in the hypothesis that the intact passivation layer provide s a quite good
hermetic barrier, and that the effects of surface contamination cannot be seen
with in the stati stical variations of the PMIC devices. It may also b e hy pothesized
th at the pro cess of spinning on the passivation oxide creates e ither a n in ter f acial
or b ulk lev e l of p ositive mobile i on contamination t hat contrib utes t o a muc h
higher intrinsic decay of the uncontaminated sensors with the p assivatio n l ayer
intact. These hypotheses together can explain the curves seen in F ig ure 82, a nd
will be e xplored further in consi deratio n o f results obtained from testi n g
hermetic co a ting test c h ips HCTC‐ 6 and HCTC‐7.
An additional PMIC experiment designed into HCTC‐3 is the lateral
diffusio n t est structure, consisting of one PMIC sensor placed in the center of an
oxide‐covered area (Table 18, PMIC 11) and a second PMIC sensor placed at the
corner nearer to the edge (Table 18, PMIC 10), as shown in Figu re 83 . N ear the
edge, a passivation oxide cut is made to create a trough surrou nding the
experiment area, thereby creating access to the passivation oxide / field oxide
interface that could allow for faster diffusion of positive mob ile ion s l ate r ally
toward th e c enter. I f th is is th e cas e , then a dr o p in dra in c urrent sho uld be seen
more prominently by the PMIC sensor closest to the edge of the area, followed by
the sensor i n the cen t er .
191

Fi gure 8 3 . P ho tomi cro g r a ph o f the P M IC l ate r al d i ffusi on e xpe ri me nt t e s t are a , showi n g o n e
PMIC s e n s o r at t he c orne r and another at t he c enter of t he t est region. Both PMIC sensors are
fab ri c a te d wi t h Me t al 3 g ate s .
Figure 84 additionally shows an SEM micrograph of one edge of this
int e rface w h ere a meta l trace emerges from the interface of the top and bottom
oxide regions (top‐left‐most m etal trace in Figure 83, rotated by 180 °). The
interfaces between the passivation oxide and the Metal 2 – Meta l 3 field oxide, a s
well as b etween t he t wo f ield o xides, can be s een ; these int e rf aces m ay give r i se
to a fas ter p o sitiv e mobile ion d i ffusion ra te.
192

Figure 84. (Top) SEM micrograph of one area of the outer rim of t he t est a rea , showing a meta l
trace passing between the Metal 2 layer and Metal 3 layer oxide interface. (Bottom) An
illustrati on to help vis ualize this geome t ry.
The resulting normalized drain current decay plot is shown in F igu r e 8 5
for the cases of contaminated an d uncontaminated senso r s both at the corner of
the test area and in the center, for a total of 4 curves. Although there may appear
to b e some s epara t ion among the four c urves, due to d evice to d ev ice variatio ns
193

within the fabrication process, the curves are not significantl y differ en t.
Apparently, with the oxide passivation layer in place contamina tion d iffuses
neither from the top nor the sides in a way or at a rate that i s mea s ureable w i th
these PMI C s ensors w it h gates fabr icated i n M e tal 3.
Figure 85. P lot of n orm a l i zed drain current as a function of ti me f or t wo s ets of P MIC sensor s a t
t h e corner ( Red) a nd a t t h e cent er ( Blue) of t he l a t era l d iff u s i o n test re g ion of H CTC‐3.
Uncon ta mina t e d curves a re s hown as solid lines, while the conta min a ted curves a re s hown a s
dashed lines.
Three very important conclusions can be drawn from the results of t esting
HCTC‐3 tha t are funda m ental to t he way that future chips should be desig n ed:
1. The value of e xposed f loating ga tes for PMIC sensors i s twofold .
They can be directly overcoated with hermetic thin and thick fi lm
coatings, an d then u sed to e valuate the hermeticity per f or mance of
the coatings. E xposed f loating ga tes are also l ess subject to the
194

effects of intrinsic field oxide and passivation oxide contamin ation.
However, f l o atin g ga tes that p ro tr ude through the oxid e and ar e
exposed to the air cannot hold a charge due to ambient positive
ions in the air. In future research, exposed floating gate PMI C
sensors could be evaluated in a vacuum oven to eliminate
atmospheric positive mobile ion contamination. Alternatively, the
floating gate metal layer can be c hosen to r eside below one or more
oxide layers.
2. Exter n al c o n tamin a tion d oes no t appear t o signific antly affect
PMIC sensors when the passivation layer is present, even if the
floating gate is fabricated at the top metal layer. To mitigat e this,
the PMIC sensors must be passivation oxide cut in order to prod uce
an observa ble effec t .
3. The closer the floating gate is to the surface of the chip, the better
the sensitivity that i s observed in response to external ion
contamination.
Analyz ing t h ese three conclusions a s a whole pushed t he d esig n of f u t ure
PMIC sensors in a very specific direction. Namely, when consid er ing the firs t a n d
second conclusions t ogether, t here must be an oxide cut present , and the f l oati ng
gate must not be exposed to the air. Therefore, the only feasi ble g a te l ayers ar e
poly or Metal 1 layers. When considering the third conclusion, the higher layer is
preferred. As a conseq u ence, the fourth g ener ation HCTC d esig n s all incorporate
195

PM IC s enso rs w ith fl o a ting gat es i n th e M e tal 1 l ayer. B e c au se th e e f f ec ts o f ga te
size could not be tested in HCTC‐3, as the PMIC sensors with va rying gate s izes
were fabricated in Metal 3 under a passivation cut and therefor e ex posed to a ir ,
this will also be addressed by the fourth generation hermetic c oating t es t chip
desig n s (HCTC‐5, HCT C‐6, and HCT C‐7).
4.5.2 Hermetic Coating Test Chip 4 (HCTC ‐4)
The fourth h ermetic coatin g t es t chip ( HCTC‐4) departs from t he past set
of experiments primarily dealing with PMIC sensors, and was spe cifically
desig n ed t o test a n ar ra y of m ois t ure‐sensitiv e elements. T wo types of m oisture‐
sensitive fu nctional devices were included. First, two very la rge poly‐oxide‐poly
(PoP) capacitors w ere designed and fabricated to allow for much e asier I‐V curve
measurements due to the larger capacitance. These devices also feature two
differ ent levels o f per f orat ion of t he t op p oly electrode to a l low for better
penet r at ion of positiv e m o bile ions into the d ie lectric r e gio n .
Second, two suites of nine interdigitated electrodes were desig ned that
are placed a t the top‐ most m etal layer of the chip (Metal 3) with a passivation
oxide cut t o e xpose t h e electrod es t o the a i r. T he d esign was fa b r icated o n
November 23, 2011, again using the MOSIS service and the ON Sem iconductor C 5
process (run n umber V17L). A f ull set of m easured process para meters a fter
fabrica t ion is p rovide d in A ppe n d ix A . T h e C ad ence V irtuoso layout and a
196

photomicrograph of t he a s‐fabric a t ed c hip ar e shown in F igure 8 6. A summary
of all key devices that w ere tested follows in Table 19.

Fi gure 8 6 . The Cade n ce l ayou t (T op ) a n d an o p t i c al mi c ro gr aph ( Bot t o m ) of H ermetic C o a t ing
Test C hip 4 (HCTC‐4).
197

Ta ble 19. Summa ry o f k e y devices tha t w ere t e sted o n herme t ic c oat i ng t est chip HCTC‐ 4
including size, loc a ti on, an d pr esence of a pa ssivat ion oxide c ut f or ea c h device.
Device
No.
Device Type Description / Dimensions
Oxide
Cut?
Location
(x, y in µm
from bottom ‐
left corner)
1 Cap a ci tor
Poly2 over Poly1 ca p a c i t or occupying ne a rly the
entire to p hal f of the c hi p, w it h 25% of t h e a rea of
t h e Poly2 la yer perforat ed a nd a desi g n capaci tan c e
of 500 pF
Yes 1000, 1 375
2 Cap a ci tor
Poly2 over Poly1 ca p a c i t or occupying ne a rly the
e n ti re b ot tom half o f the chi p , wi th 8 % o f t he are a
of t he P oly2 layer perfor a te d an d a de si gn
ca pacita nce of 550 pF
Yes 1000, 5 00
3
Interdigi t ated
Elect rodes
A set of nine int e rdig i t at ed Met a l 3 elect rodes wi t h
f i ng ers of size 8.4 µm 100 µm a nd wi t h e v ery
pos s i b le pe rmuta ti on o f g a ps in t h e set [2 µm, 4 µm,
8 µm] a n d number of f i ng er s in t he set [2, 8, 32]
Yes 936, 13 0 2
4
Interdigi t ated
Elect rodes
A set of nine int e rdig i t at ed Met a l 3 elect rodes wi t h
f i ng ers of size 8.4 µm 20 µm a nd wit h every
pos s i b le pe rmuta ti on o f g a ps in t h e set [2 µm, 4 µm,
8 µm] a n d number of f i ng er s in t he set [2, 8, 32]
Yes 936, 60 0
The gr een areas of t h e m icrogr a p h are the two poly‐to‐poly c apa citors,
each d es ign e d to h a ve a per f orat ed u pper electrode to a l l ow f or p os itiv e m o bile
ions to pass through into the oxide between the poly capacitor elec trodes. T he
smaller capacitor on the bottom of the chip has 8% of the area perfo r ated, w a s
desig n ed t o have a c ap acitance o f 500 pF, was simulated to h ave a capacitance of
518 pF, and was measured to have a capacitance of 524 pF. The larger capacitor
on the top of the chip has 25% of the area perforated, was designed to have a
capacitance of 550 pF, was simula ted to h av e a capacitance of 570 pF, and was
measured to have a capacitance of 618 pF. As of the time of th is w r i t i ng,
198

extr actio n o f the I‐V curves f ro m these two capacitors a nd m eas uring their
response to ion contam i nation is ongoing.
The whit e s e rpen tin e ‐like ar eas in t he t op a nd b ottom center r e gions of
the chip a r e 18 differ ent g e omet ries o f interdigitated e lectrod es t hat were
desig n ed t o be u sed for moisture d etect i on. A ll electrodes i n this r egion we re
fabricated i n the Metal 3 layer and have a n oxide cut to e xpose the electrodes to
the air. The devices were separated into two groups, one group w ith 100 µm
fing er l en gt hs, and the other group with 20 µm f ing e r len g t h s. Finger widths were
kept constant for all of the devices, but the gaps between fing ers v a ried f rom 2 µm
to 8 µm, and the number of fingers on each device varied from 2 to 32. A
photom icrograph showing all nine 20 µm f ing e r length d ev ices i s shown in Figure
87.
Fi gure 8 7 . P ho tomi cro g r a ph s h o wi ng t he fab ri ca te d i n te rdi g i t a t e d elect r ode s with 20 µm f ing e r
le ngths.
199

Because all of t he i nt er digit a t e d el ectrodes w ere now ex posed t o air in t his
design, a simple breath test wit h visible cond e nsation r e gister ed a D C res i sta nc e
measurem ent as low as 25 MΩ that quickly returned to an open ci rcuit condition
once t he condensation evaporated w ithin approximately 3 seconds. Distilled
water droplets w ere also p laced on t he s urface o f the chip, and resistance
measurements as low as 4 MΩ were registered. Lastly, saline dr o p lets w er e
placed o n t h e surface, b ut r eadin g s were l a r g e ly i ncons i s t ent d ue t o the rapid
evaporatio n of t he s ol ution; however, m easur e ments wer e a s low as 1 MΩ in this
case a cross all sets o f electrodes. T able 20 contains a summary of the measured
resistanc e values for both the b reath test and the distilled wa ter tes t .
Ta ble 20. R esista nce mea s urement s o f key int e rdi g ita t ed e lect ro des f o r tw o different m ois t ure
source s.
Device Description
(Number of Electrodes
Electrode Length, with
8 µm Finger Width)
Visible Condensation From
Breath at 1” for 15 seconds
Drop of Distilled Water
2 100 µm w it h 8 µm ga p s > 120 MΩ (Open) 8 .8 MΩ
8 100 µm w it h 4 µm ga p s 52 MΩ 7.6 MΩ
32 100 µm wit h 2 µm ga p s 28 MΩ 3.5 MΩ
2 20 µm with 8 µm ga p s > 120 MΩ (Open) 12 MΩ
8 20 µm with 4 µm ga ps 48 MΩ 9.3 MΩ
32 20 µm w it h 2 µm ga p s 25 MΩ 4.1 MΩ
With t hese r esults, it i s clear that i n t erd i gitat e d electrod es on the ord er of
20 microns on a side can provide enough sensitivity to even dis tilled water to be
measured w ith on‐chip electronic s. T he m ea sured res i st ance d id n ot a ppear t o
200

depend significantly on the finger length, but rather only on the number of fingers
and the fin ger gap s iz e. T his a g r ees w ith the hypothesis t hat add i tional f ing e r s
effectively add more resistive elements in parallel, while a decreased gap size
decreases the resistance o f each element. Testing such devices can prove
somewhat cumbersome, but the new ESPEC humidity oven is current ly b eing
i nte grate d w i th acce le rate d lif e ti me te s ti ng i nstrume n tati on, p ro viding t he
capability f or s tabilized temper ature and humidity f or h igher p rec i sion t esting,
as discussed further in C hapter 6.
4.6 Summary of Key Findings
In this chapter, designs of five different hermeticity test chi ps s panning a
time period of over 2 years of development and testing have bee n pres ent e d, w ith
numerous lessons learned that can be carried forward to future designs. T h e
following summary attempts to capture a brief description of each chip, along
with key lessons learned and conclusions that are important to future w ork in
this ar e a.
Hermetic Coating Test Chip 1 (HCTC‐1) was designed with a shotg un
approach to incorporate several sensors potentially sensitive t o moisture a nd
positive mobile ion contamination. Some of the key devices fab ricated and tested
included a n EPROM‐based positive m obile ion contamination ( PM IC ) sensor,
interdigitated e lectrodes for detecting moisture, and cl ock gen erati o n circui ts
based on a r ing oscillator to e xplore l ow‐power c locking of b io im plantable im a g e
201

sensors. The ring oscillator designs were shown to work near the desig ned
frequency and all variants fabricated were operational. One ke y les s on l earned
from HCTC‐1 was that there was insufficient output buffering to a deq u ately drive
an oscilloscope probe for measurement, and the fr equency was to o high to clock
a commercial image sensor. T he P MIC senso r s showed a s ensitivi ty t o posi ti ve
mobile ions, and were proven to be viable for this application. However the
magnitude of the observed voltage drops were not large enough t o comfortably
measure th e differenc e s over a r ea sonable amount of tim e .
The issues w ith the HCTC‐1 c lock gen era t ion circuits w ere ad dre ssed in
the CCTC chip design. Additional output buffering was added an d an active scope
probe was used to rest ore the sha r p edges of th e sq u are w a ve o u tput. However,
the additional b uffering a ctually c aused ringing to occur near the rising e dge,
with a high degree of noise during the signal high period. Despite this, the
frequency and w a ve form w ere suffi ciently stable t o clock a n OV6 920 C M OS
image sens or a nd t o capture images f ro m a devel o pment board use d in
conjunction with the IOC polymer lens.
The interdigitated e lectrodes f r o m H CT C‐ 1 we re r edes ign e d i n th e H C TC‐
2 chip t o be l arger for easier m eas u rement; h o wever, t hey were plac ed t oo deep
to d etec t a n y effect s of m oistur e condensation near t he s urface. The PMIC
sensors were essentially identical to those of HCTC‐1, but now included
experiments with various overlayers of metals and oxides. Some e ffec t s wer e
seen d ue t o applied ex t e rnal con tami nati on, but i t w as dete rmi n ed t h a t the PMI C
202

floating gates w ere located too deep t o see a reasonable s ensit iv ity t o p ositive
mobile ions provided b y external contamination.
The third‐generation HCTC‐3 chip expanded on the experiments with
varying levels o f protection f ro m HCTC‐2, and i ncorporated a su ite of n ine PMIC
sensors at various floating gate depths within the oxide stack. It was quickly
discovered that when a PMIC sensor floating gate is exposed to air (placed at
M e tal 2 o r M etal 3 w it h a p assiva t i o n o xid e c u t ) , th e sen so r c a nno t h o l d a c h arge
in air and therefore cannot be used to measure ion accumulation . H owever,
several viable Metal 1 and poly gate PMIC sensors remained oper atio nal on t his
chip, and w e re u s e d to d educe that e ven w i th n o ex ter n al contam ina t ion applie d,
the floating gate voltages still drop substantially, likely due t o intrinsic
contamination w ithin the field and passivation o xides. F urther more, the sensors
located below a passiv a tion o x i de c ut w ere those most l ikely to s ee a sign ific an t
increase i n decay rate i n the presence o f additional i on contam in at ion. Therefor e ,
the PMIC sensors can only be tested with external contamination in the case of
an oxide cut over the sensor. Lastly, PMIC sensors placed clos er t o t h e top‐most
oxide wer e m ost sens it ive to add e d ex t er nal c o ntam in ation. The se o bservations
led to t he conclusion that s ubse quent designs should p la ce a n oxid e cut over the
sensor r egion, and that the floa ting gate should be placed in t he M etal 1 layer.
With two years o f kno w ledge and experience gained with positive mobile
ion contamination s ensing, a set of f ourth‐generation c hips w as designed,
fabricated, and tested, as will be presented i n Chapter 5.
203

CHAPTER 4 REFERENCES
[1] J. M. Andrews, Jr., N. Lifshitz, and G. Smolinsky, "Method of Manufacturing
Semiconductor D evices Involving t he D etect i on o f Impurities," U .S. Patent
4,978,915 A, issued D ecember 18, 1990.
[2] J. C . Mitros, "Method for Detecting Mobile Ions in a S emico nductor Device",
U. S. Patent 5,394,101 A, issued February 28, 1995.
[3] F. Shone, H. Liou, C. Pan, B. Woo, and M. Holler, “A novel method t o
characterize and screen mobile i on contaminated nonvolatile mem ory
products,” i n 1991 International Symposium on VLSI Technology, Systems,
and Applications , pp. 224–226, 1991.
[4] J. G arcia and M. M cBride, "A ppar a tus for Mea s uring Co nt am in ant M o bile
Ions i n Diel ectric M aterials," U .S. Patent 6,699,436 B1, issued March 2,
2004.
[5] G. R . Gilliam, "On‐Chip M obil e Ion Contamination Tes t C ir cu it", U .S. Patent
5,790,463 A, issued August 4, 1998.
[6] C.‐S. Hsu and M.‐D. Liu, "Method of Monitoring Ion Contamin ation in
Integ r ated C ircuits " , U. S. Patent 6, 025,734 A, issued February 15, 2000.
[7] J. J.‐W. Liou, C.‐J. Huang, H.‐H. Chen, and G. Hong, “Characterization of
process‐induced mobile ions on the data retention in flash memo ry,” IEEE
Transactions on Electron Devices , vol. 50, no. 4, pp. 995–1000, 2003.
[8] P. H orowitz and W. H ill, The Art of Electronics, 2nd Edition , Cambridge
Univers ity Press, Cam b ridge, United Kingdo m, 1989.
[9] J. C. Mitros, “Single Gate EPROM Cell for the End‐of‐line I onic
Contamination Test,” i n 1993 International Integrated Reliability
Workshop Final Report , pp. 40–44, 1993.
[10] W . C. T ang, T .‐C. H . Nguyen, and R. T . Howe, “L aterally D r iven P olysilicon
Resonant M icrostructures,” i n Proceedings of 1989 IEEE Micro
Electromechanical Systems Workshop , pp. 53–59, 1989.
[11] G. K. Fedder, R. T. Howe, T.‐J. K. Liu, and E. P. Quevy, “ Technologies f or
Cofabricating MEMS a nd E lectronics,” Proceedings of the IEEE, vol. 9 6, no.
2, pp. 306–322, 2008.
204

[12] K . Sundaresan, K. C . Brouse, K. U ‐ Y en, F. A yazi, and P. E . Allen, “A 7‐MHz
process, temperature and supply compensated c lock o scillator in 0 . 25 μm
CMOS,” i n Proceedings of the 2003 International Symposium on Circuits and
Systems , vol. 1, p. I‐693, 2003.
[13] X . Zhao, R. C hebli, a nd M . Sawan, “ A wide t uning range vo l tage‐contr olled
ring o scill a tor dedic a ted to u ltrasound transmitter,” in The 16th
International Conference on Microelectronics , pp. 313–316, 2004.
[14] R. H. Nixon, S. E. Kemeny, B. Pain, C. O. Staller, and E. R. F ossum, “ 256×256
CMOS a ctive pixel sensor c amera‐on‐a‐chip,” IEEE Journal of Solid ‐State
Circuits , vol. 31, no. 12, pp. 2046–2050, 1996.
[15] E. R. Fossum, “CMOS Image Sensors: Electronic Camera‐On‐A‐ Chip,” IEEE
Transactions on Electron Devices , vol. 44, no. 10, pp. 1689–1698, 1997.
[16] K . B. C ho, A. I. Krymski, an d E. R . Fossum, “ A 1.5‐V 550‐μ W 176 144
Autonomous CMOS Active Pixel Image Sensor,” IEEE Transactions on
Electron Devices , vol. 50, no. 1, pp. 96–105, 2003.
[17] E . R. F ossum, “ Digital Came ra S ystem‐on‐a‐Chip,” IEEE Micr o, vol. 18, n o.
3, pp. 8–15, May/June, 1998.
[18] T. H. Lee and A. Hajimiri, “Oscillato r phase noise: A tuto rial,” IEEE Journal
of Solid ‐State Circuits , vol. 35, no. 3, pp. 326–336, 2000.
[19] A. Hajimiri, S. Limotyrakis, and T. Lee, “Jitter and phase noise in ring
oscillators,” IEEE Journal of Solid ‐State Circuits , vol. 34, n o. 6, pp. 790–804,
1999.

205

CHAPTER 5
FOURTH GENERATION HERMETIC COATING TEST CHIPS
5.1 Introduction
This c hapter i ntroduces the curr ent fourth g eneration h e rmetic coating
test chips, which build upon the lessons learned from the third generation
hermetic coating test chip designs presented in Chapter 4. The primary goal of
this fourth design generation is to further refine the sensitiv ity, reliability, and
accuracy o f the PMI C s ensors. S eco n dary goals a re t o further t he u nd erstanding
of how the passivation oxide cut affects the sensitivity of the P MIC sensors, to
further the understand ing establ is hed by t he e xper im ent s o n HC T C‐ 3 that u se
spare metal layers f rom the CMOS p rocess as a n over‐shield to p rotect s ubvening
circuitry ag ains t ion co ntaminat ion, a nd t o develop a platform suitable for the
testing of post‐deposited hermetic coatings. These goals play an important role
in d etecting i on contamination in situ and in real time, as both microcase and
micropackaging t echnologies have been focused recently on hermeticity, but
with no pra c tical way t o validat e herm et icity a f te r im plant a tio n [ 1–4].

206

5.2 Hermetic Coating Test Chip 5 (HCTC ‐5)
The fifth hermet ic coatin g t est chip ( HCTC‐5) contains essentia lly a
superset o f the most u seful structures f ro m HCTC‐1 a nd H CTC‐3. H CTC‐5
includes a series of PMIC sensors that are distributed across t he c hip and around
the chip p erimet er w it h the idea o f tes t ing th e unifo r mit y o f a h ermetic coating
that could be deposited in a post processing step. The intent was to d eposit
hermetic c o a tings d i rec t ly o n the sensors exposed to a ir a nd t h er eby directly t es t
the hermet icity of t h e coatin g . H owever, when t his chip w as s u bmitted for
fabrication, it was still unknown that a PMIC sensor with a flo ating gate t hat is
exposed to air w ill not hold a charge.
Since the PMIC sensors placed around the chip were in the Metal 3 layer
and have a passivation oxide cut , t h ey were u n fortun atel y unusa ble. However, a
number of test structures were included, as well as a suite of PMIC s ensors, that
duplicated many of the HCTC‐3 experiments, but this time placed under the
passivation oxide to prevent exposure to the air. Lastly, an oxide t hickness test
structure, c lock generators, and interd igitated e lectrodes were a lso included i n
the design. HCTC‐5 was fabricated on June 15, 2012, again usin g the MOSIS
service and the ON S emiconductor C5 process (run number V23R). A full set of
measured process parameters after fabrication is provided in Appendix A. The
layout and photomicrograph of the fabricated chip are shown in Figure 88. A
summary of all key devices that were tested f ollows in Tab l e 21.
207

Fi gure 8 8 . The Cade n ce l ayou t (T op ) a n d an o p t i c al mi c ro gr aph ( Bot t o m ) of H ermetic C o a t ing
Test C hip 5 (HCTC‐5).
208

Ta ble 21. Summa ry o f k e y devices tha t w ere t e sted o n hermet ic c o a ti n g t e s t chi p H CT C ‐ 5 ,
including size, loc a ti on, an d pr esence of a pa ssivat ion oxide c ut f or ea c h device.
Device
No.
Device Type Description / Dimensions
Oxide
Cut?
Location
(x, y in µm
from bottom ‐
left corner)
1 PMOS P MICs
A se t o f ni n e P ‐ Channe l EPR O Ms t o ac t as p osi t i v e
mobile ion c o n taminati on s ensors, all with Metal 3
floa ti ng g a t e s w i t h are a s r a ngi n g fro m 2 7 µm
2
t o
2110 µm
2
and channel sizes of 9 µ m 3 µ m an d
9 µm 36 µ m . Devices did n ot h old a ch arge due t o
exposure to a i r a n d t h e s e nine a re redesigned a n d
det a iled f urt h er in Sect ion 5.3.
Yes
1013, 9 79
(all nine in
appr oxim ately
the s a me
center
loca tion )
2 PMOS P MICs
A set o f eigh t i dentic al P‐Ch a nnel EPROMs t o act as
posi t i ve mobile ion cont ami n ati o n se ns or s wi th
Met a l 3 f l oa t i ng g at es wit h a n a rea of 324 µm
2
a nd a
channel o f s iz e 9 µm 3.6 µ m, po s itioned aroun d
the chi p f or a s pa ti ally d i s tri b ute d s am pli n g o f
con ta mina tio n ; als o d etaile d in Sec tion 5.3.
Yes
400, 40 0
550, 10 0 0
400, 16 0 0
1000, 5 50
1000, 1 450
1450, 1 000
1600, 1 600
1600, 4 00
3 PMOS P MICs
Id ent i ca l s e t of n ine H C TC‐3 PMIC devices placed
for redundancy on HCTC‐5
No
780, 72 0
to
780, 11 4 6
4 CMOS Cl o ck
Seven sta g e 2 45 µm 70 µm ring oscilla tor
de si gne d wi t h an o ut put fre q ue ncy of 2 5 MHz and
with tw o D‐Flip‐Flop o u t put st ages f or a d d i tional
12.5 MHz a n d 6.25 MHz output s
No 1070, 7 38
5 CMOS Cl o ck
Ident i ca l to D evice 4 but pl a c ed under a pa s sivat i on
oxide cut
Yes 1482, 7 88
6 Oxide We dge
Pas s iva t ion ox ide cut t a pere d fr om a widt h o f
12 µm d own to 1.2 µm over a length of 20 0 µm to
form a wed ge‐shape d cu t
Yes 1220, 8 64
7
Interdigi t ated
Elect rodes
Interdigi t ated Metal 3 ele c t r ode s wi th 8 f i n ge rs o f
size 4.2 µm 55 µm an d 2.1 µm gaps for m ois t ure
detec t ion
No 1220, 1 060
8
Interdigi t ated
Elect rodes
Ident i ca l to D evice 7 but pl a c ed under a pa s sivat i on
oxide cut
Yes 1404, 1 144
9
Interdigi t ated
Elect rodes
Ident i ca l to D evice 7 but pl a c ed under a ta pered
pa ssivat ion oxide cut
Yes 1220, 1 128
10 Surfa c e Prof ile
Prof ilomet er test st ruct ure to me a sure t hick ness of
various met a l and oxi d e laye rs
Yes 1032, 1 275
209

Because this desig n wa s the firs t that was primarily designed t o be tested
with p ost‐fabrication deposited coatings, a number o f test s tru ctures w ere
included t o specific a lly t est the cover age capability a nd f ea si bility o f an
int e rdigit at ed e l e ctrod e t o det e ct m oisture with a coating a ppl ied. For example,
an oxide cut was placed with a diminishing width along its leng th, which resulted
in a wedge‐shaped indentation in the passivation oxide narrowin g from 10
microns down to less than 1 micron. It was thought that a depo sited c o ating may
bridge the gap at some point along the tapered oxide cut and cr eat e a n air pocket ,
which is t ho ught t o be e ssent ial for the de tect io n of m oisture on a n int e rdig it ated
electrode.
In o rder t o optim i ze t h e d ev ice se nsit ivity to p ositiv e m o bile ions, PMIC
sensors with various gate and channel ar eas w e re d esigned with expo sed floa tin g
gates and placed i n t h e cent er o f the chip w ith an o xide p a ssiv ation cut. A
photomicrograph of t hese P MIC se nsors (top), t he w edge‐shaped o xide c ut
(center), and a version of the wedge placed over an interdigita ted electrode
(right) is shown in Figure 89. Although several thin film coat ings ( disc ussed later
in this chapter) were applied, none showed any visible indicati on t hat the gap
was bridged at any point.
210

Figure 89. Photomicrograph showing a close‐up view of the PMIC sens o rs wit h v a rio us g at e si zes
(Top ), t he w e d ge ‐ s h a pe d o x i d e cut (Ce n te r), an d the se t o f i nte rdi g ita t ed e lect rodes (Rig ht )
included in t h e HCTC‐5 desig n.
As was previously mentioned, nearly all of the PMIC sensors were placed
in the Metal 3 layer with a passivation oxide cut, and were the refo re exposed t o
air and unusable in a majority of tests. In fact, even with a dielec tric coat i ng
applied post‐fabricatio n , the su rfa c e float i ng g ates s till did not hold a charge
longer than about 1 minute at room temperature. However, sever al P M I C sensor
desig n s duplicated f r o m the HCTC‐3 c hip design w ere placed u nder the
passivation oxide, and remained viable. Long term data from a number o f
duplicate experiments on uncontaminated HCTC‐5 chips were run t o begin to
understa nd t he s tatis t ical varia tio n s from c hip‐to‐chip for ide n t ical e xperim e n ta l
conditions.
211

Fi gure 9 0 . P lot o f t he a v e rage n orm a li ze d drai n cur re nt as a f u n c ti on o f ti me f or four di ffer e n t
HCTC‐5 chips, with error bars indi c ati n g ± 1 st a nd ard de vi ati o n a t se l e c te d ti me s.
Figure 90 shows a plot of normalized drain current as a function o f t ime
for PMIC sensors placed below a passivation oxide layer on an uncontamin a t e d
HCTC‐5 chip, showing the natural decay due to intrinsic contami nation w ithin the
field oxides and passivation oxides. Error bars are shown that indicate one
standard deviation above and below the mean at selected times f or s ets of d ata
from four different chips. This plot indicates that a variation of about 6 percent
in n ormaliz e d drain current can b e expec t ed a t 100 hours due to chip‐to‐chip
variatio ns, and that t h e magnitude of the variat ion incr eases w ith time. T his
result means that any two curves for identical devices fabricat ed on separat e
chips that appear closer than about 6 percent may not actually be s ignificantly
212

differ ent. B ecause m ost experime nts invol v ing the he rm etic coa ti ng t e s t chi p s
and ion contamin a t i on d etect i on a re very l o ng t erm (some exceed ing 1,000
hours), a full statistic a l analys is o f ever y exper i ment i s not feasible. N onetheless,
this d ata suggests the magnitude of c hip‐to‐chip varia t io n that m ay b e expect ed
i n general.
As mentioned previously in this chapter, HCTC‐5 was designed wi th a goal
to test coatings applied externally after the chip is fabricate d and delive red by
MOSIS. Some HCTC‐5 chips were used to test this process in coo peration w ith
Morgan T echnical C er amics (Allentown, P A) , a company t hat speci alizes i n
dielectric “ diamond‐like” coatings that could b e used t o hermetically seal VLSI
chips against ion contamination and moisture. A process was cr ea ted to p rotect
the ceramic DIP package that h olds the HCTC‐5 microchip against p rocesses
consistent with coating d eposition, i ncluding h igh temperatures and voltages. An
aluminum foil wrap was applied such that all pins of the packag e w e re s horted,
with a small window cut out to allow for the coating to be appl ied on t he
microchip area only. A photograph of an HCTC‐5 chip ready to b e coated (left )
and a close‐ up of the chip surfa ce after coating (right) are sh own in Fig ure 91.
Two differe n t coat i ng s were a pplie d for in it ial test ing of t he PMI C sensor
charge a ccumulation f or contamina ted and uncontaminated cases: (1) a
dielectric multilayer diamond‐like coating (DLC) and (2) a mult ilayer d ielectric
DLC and m e tal s t ack. In the case o f the coa ting contain ing met al l aye rs, all pi ns
were determined to be shorted, likely due to the fact that the wire b onds
213

themselves w er e not masked fro m t he coat i ng d epos ition. H owev er, the pure
dielectric coating was successfully tested for h e rmeticity as d escribed b elow.

Figure 91. H CTC‐5 chip p repared f o r co ati n g wi th a n alumi n um f o il w ra p ( Top) a nd a macro
phot og raph o f t he chip aft er c oa t i ng (Bot t om).
214

Initial experiments with PMIC sensors used the regions of the c hip
covered with both the passivation oxide and a diamond‐like coating. The use of
a saline contamination drop was again applied as a source of ex tern al
contamination, and th en t he c hip was meas ured a t an e levated te mperatur e
(180 °C) t o perform accelerated lifetime testing. The drain current was
measured, and the PFET c alibration c urve w as u sed to convert t h e plot t o the
number of accumulated charge s as shown in Figure 92.
Fi gure 9 2 . P lot of t he a c c umula t ed n u m ber of p osi t ive ions a s a function of time for both
uncon ta minat e d an d c o nt a m inate d u nc o a ted chip s, a n d a c ont a mina t e d chip wit h a mult ilayer
dielect ric diamond‐like coat ing (DLC).
The results of these initial tests were quite promising, with t he coated and
contaminated chip s howing n early the exac t same r at e of a ccumul ation of
positive mobile ion charges as a chip that was not coated, but also n o t
contaminated. C ompared with t h e b aseline chip t hat contained a PMIC sensor
that w as n o t coated but contamin a ted, i t appears that t h e coati ng m itig ate s t he
215

effects of t he e xternal contamina tion. I t wa s later disc overed t hat other post ‐
fabrica t ion processing s teps r ela t ed t o the deposition o f the coating may also
either generate part ial l y hermetic l ayers o r r emove surface con tami nati on that
influence th e rate o f charge a ccu mulation. C ontinuing tests wi th p ost‐fabrication
herm etic coatin g d eposition is o ngoing w it h HCTC‐7, and is a n a rea of great
research int eres t for th e future.
5.3 Hermetic Coating Test Chip 6 (HCTC ‐6)
The shortcomings o f th e exposed fl oatin g g at es o f the HCT C ‐5 d e sign w ere
address e d with h ermetic coating t est chip H CTC‐6, a nd t here fore t his chip i s
essentially identical to HCTC‐5, but with all floating gates th at w er e prev iously
fabricated in the Metal 3 layer lowered to the Metal 1 layer, t hereby p rotecting
them with a sufficient thickness of field oxide such that they can hold a charge
irrespective of whether they are below a passivation oxide cut or n ot. W ith HCTC‐
6, the full suite of PMIC sensors with various floating gate an d channel sizes can
now be t es ted to f urth er o ptimiz e the gate g eometries fo r futur e d e signs. T his
desig n was fabrica ted o n December 18, 20 12, again using the MOS IS s ervic e a nd
the ON Semiconductor C5 process (run number V29J). A full set of m easured
process parameters after fabrication is provided in Appendix A. The layout and
photomicrograph of the fabricated chip are shown in Figure 93. Because this chip
design is largely the same as HCTC‐5 but with previously air‐ex posed PMIC
216

sensors now buried under oxides, all of the key PMIC devices that are different
from HCTC‐ 5 and that are the focu s of this chip are summarized in T able 22.

Fi gure 9 3 . The Cade n ce l ayou t (T op ) a n d an o p t i c al mi c ro gr aph ( Bot t o m ) of H ermetic C o a t ing
Test C hip 6 (HCTC‐6).
217

Ta ble 22. S umma ry o f key PMIC d evices t ha t were t est e d on h erme ti c co a t i n g te st c hi p H CT C‐ 6 ,
including size, loc a ti on, an d pr esence of a pa ssivat ion oxide c ut f or ea c h device.
PMIC
No.
Gate
Layer
Gate Area
Channel Size
(W L in
µm)
Oxide
Cut?
Experiment Set
Location
(x, y in µm from
bottom ‐left corner)
1 Metal 1 324 µm
2
9 µm 36 µm Yes Varying Gate/Channel 944, 10 2 4
2 Metal 1 324 µm
2
9 µm 3 µ m Yes Varying Gate/Channel 1006, 1 024
3 Metal 1 324 µm
2
9 µm 3 µ m Yes
Varying Gate/Channel
(Calibration PFET)
1071, 1 024
4 Metal 1 27 µm
2
9 µm 36 µm Yes Varying Gate/Channel 944, 97 3
5 Metal 1 27 µm
2
9 µm 3 µ m Yes Varying Gate/Channel 1006, 9 73
6 Metal 1 640 µm
2
9 µm 3 µ m Yes Varying Gate/Channel 1071, 9 73
7 Metal 1 2110 µm
2
9 µm 3 µ m Yes Varying Gate/Channel 978, 92 1
8 Metal 1 1126 µm
2
9 µm 3 µ m Yes Varying Gate/Channel 1078, 9 21
9 Metal 1 324 µm
2
9 µm 3 µ m Yes Spat ia lly Distribut ed 400, 40 0
10 Metal 1 324 µm
2
9 µm 3 µ m Yes Spat ia lly Distribut ed 550, 10 0 0
11 Metal 1 324 µm
2
9 µm 3 µ m Yes Spat ia lly Distribut ed 400, 16 0 0
12 Metal 1 324 µm
2
9 µm 3 µ m Yes Spat ia lly Distribut ed 1000, 1 450
13 Metal 1 324 µm
2
9 µm 3 µ m Yes Spat ia lly Distribut ed 1600, 1 600
14 Metal 1 324 µm
2
9 µm 3 µ m Yes Spat ia lly Distribut ed 1450, 1 000
15 Metal 1 324 µm
2
9 µm 3 µ m Yes Spat ia lly Distribut ed 1600, 1 600
16 Metal 1 324 µm
2
9 µm 3 µ m Yes Spat ia lly Distribut ed 1000, 5 50

218

Because the floating gates on HCTC‐6 are now buried below field oxide,
the sensors originally p laced around the edges of the chip to t est coa tin g
uniformity can be tested. Plots of chip‐to‐chip variation were p resented in
Section 5.2 by testing one sensor on four different chips, but now, i n this c ase,
four sensors on the same chip with a minimum of 1.5 mm of later al s eparation
can be t est e d in a s ing l e experiment t o d e t e r m ine th e va riat ion from device‐to‐
device on a single chip. Figure 94 shows a plot of the normali zed drain current
as a function of time for four PMIC sensors (Table 22, Devices 9, 11, 13, a nd 15)
placed below a passivation oxide cut on a single HCTC‐6 chip, f or b oth
contaminated and unc o ntaminated cases.
Note that in this plot each of the four curves is shown so that var iations of
both the decay rates and curvature can be seen. Also note that the variance
within each set of four decay curves is very similar to the statistics collected i n
Chapter 4, a nd t he a pproximate 6 percen t v a ria t ion bet w een c urv es s eems to
hold t rue even t hough in t his case, the P M IC s ensor s w ere co ver ed b y a
passivation oxide and were all fabricated on the same chip, but i n differ ent
locations within the chip.
It c an b e o b served t h a t ther e is m ore variability in g eneral w i th t he
contaminated curves t han with t h e u ncontam i nated curves. T his is l ikely due to
slight i nconsistencies i n how th e contamination l ayer i s applie d, a lthough it i s
believed t h a t th e surfa c e is e ss ent i ally s a t ura t ed w ith ion con tami nati on give n
the concentration o f saline a pplied. It can also b e observed t hat after only 50
219

hours of testing, there is a clear separation between the group ings of
uncontaminated and c ontaminated curves. This indicates that an im minent chip
failure may be d etecta ble early in t he d ecay process, a lthough the acceleratio n
factor a t th e temper at ure used has n ot y et b een d eter mined a n d is a n a r ea of
future res earch.
F i g u re 94. P lot of t h e norma l ized d ra in c urrent a s a funct i on o f ti me for o ne uncon ta mi nate d an d
one contamin ated chip, e ach with four P M IC sens o rs t ested simul taneously.
The suite o f s ensors i n t he cent er of the chip i s composed o f P MIC sensors
with a range of floating gate and FET channel areas. Figure 95 shows a plot of the
normalized d rain current a s a fu nction of time f or f our such P M IC s ensors: (1) a
large channel area and a s mall ga te ar e a (Table 22, D evice 4), (2) a large channel
area a nd a l arge g a t e a r ea ( Table 22, Device 1), (3) a small ch annel area a nd a
220

large g a te a rea (Table 2 2, D evice 2) , and (4) a s m all channel a rea and a ver y l arg e
gate area (Table 22, Device 8). Note that the labels correspon ding t o the ga te a n d
channel ar eas in t he p lot are ap proximate numbers fo r clarity, and t h at t he
precise gate and channe l sizes are given in Tab le 22. Th es e de vic e s are arrang ed
in a grid within a central area o f the chip that is approximate ly 200 µm on a sid e.
Fi gure 9 5 . P lot of t he n orm a li ze d dr ain current a s a func tion of ti me f or a n uncon ta mi nate d
HCTC‐6 c hip with four PMI C sens o rs o f d i fferent g ate and ch anne l areas.
The firs t observation that c an b e deriv e d fro m t his data i s tha t the two
sensors with large channel areas at first seem to have a much slow e r decay rate
compared with the sensors with small channel areas; h owever, du e to the larg er
channel len g th, there is m ore resistance b et ween s ourc e and drain and less hot
carrier injection of the electrons onto the gate. In other wor ds, the floating g a t e
221

does not charge up to a voltage that is as high as a PMIC sensor with a smaller
channel width, and theref o re sta rt s muc h low er on th e decay cur ve. Becau se the
drain currents ar e nor m alized, it l ooks as if th e over all decay i s much smaller.
The second observat io n is t hat a l a rger g a t e area e n a bles a m uc h more
rapid d e cay curve. In other word s, a larg er g ate area p roduces a PMIC sensor
with a much larger response to the same amount of contamination , and is
therefore more s ensitive. T his observation seems intuitive in the sense that a
larger gate area allows for the capture of a larger number of ions if the diffusion
density per unit a rea is constan t for a given level of contamin ation. However, i t
also s eems counterin tuitiv e in t he s ense t hat a larger g a t e ar e a also p roduces a
larger overall device capacitance and would therefore produce a s maller gate
voltage change f or e ac h positiv e m obile ion that n eutralizes a n e lectron charge a t
the gate. In order to resolve which factor is more dominant, a m odel i s proposed
that attempts to predic t this sc alin g of PMIC s e nsitiv ity w i th gate siz e.
Two quantities of interest can be defined that have opposing
proportionalities with r espect to the gate a r e a of a n EPROM dev ic e. F irst, the
amount of charge accumulated over a given time is given by Equa tion 10, i n whic h
the change i n charg e c an b e expressed as the product of the ion c o n tamina tion
rate (number per unit time) per unit area, σ
q
, the charge o f each i on, q
e
, and the
area o f th e collection area, in t his case a ssuming t hat the ion s are primarily
collected by a large M e tal 1 gate above a polysilicon gate, A
M1 ‐g ate
.
222

1





qe M gate
Q
qA
t

(10)
The EPROM device w ith a floating g ate can fun d ament a lly b e thou ght of a s
a capacitor that s tores a given amount o f electrical c har g e on the fl oatin g gate
after charg i ng. A s char ges are neu t raliz e d by i ncoming po sitive mobile ions, the
change of the gate voltage in re sp onse defines the sensitivity of the PMIC sensor.
From fundamental electrical engi n eer ing pr in ciples, the change in voltage given
a change in charge is simply proportional to the inverse of the t otal c apacitance,
as given by Equation 11. The total capacitance here is dominat ed b y the
polysilicon gate in close proximity to the channel, with a smal l correction due to
the Metal 1 gate.
1 


gate
total
V
QC

(11)
Because of the f act that Ctotal is d o m i nat e d b y th e p o l y ga te, th e se nsi t iv ity
(Equation 1 1) is e ss entially constant with i ncr e asing Metal 1 g at e a r ea, while the
charge a ccumulation r ate (Equat io n 10) i ncr e ases p roportionally w ith increas i ng
Metal 1 gate a r e a. This i s the fun d amental r e ason t hat the ove rall decay of t h e
gate voltage is so much faster with a larger Metal 1 gate area. The overall gate
voltage change o ve r a gi ve n ti me i nterval can be o btained by f o rming the product
of Equation 10 and Equation 11, and is giv e n b y Equation 12.
1





gate q e M gate
total
VqA
tC

(12)
223

Until n ow, plots have b een shown f or n orm a lized drain current a s a
function o f time, beca use this i s the quantity m easured, a nd t he floating gat e
voltage must be inferred from the calibration curves. Although this calibration
has been m easured to b e a logar i t h mic fit, i t is a pproxim ately linear o ver a large
rang e of g a t e voltages. P er forming a simulatio n u sing t he m easured SPICE device
parameters f rom this p articular M O SIS fabric ation run o f H CTC‐6 , and assuming
a constant ion collection rate p er u nit area, σ
q
, this model predicts that the
differ ence i n decay ra t e s between the floatin g g ate of s ize 324 µm
2
and the
floating gate of siz e 1126 µm
2
should be appr oximately a factor of 2.6.
Looking ag ain at F igure 95, a t aro u nd 100 hours into t he e xperi ment, the
difference i n decay rates for the two corresponding s ensors i s approximately a
factor o f 2.1. O bviously s ome other phenom enon i s at p lay, s ince the measured
floating gate voltage d e cay rate is about 20% slower than expected when the gate
area i s scaled f rom 324 to 1126 µm
2
, but t h e simple collection rate v ersus
sensitiv ity model can explain the ordering o f the curves a nd c an show why the
increase i n charge a ccumulation r ate (Equatio n 10) d ominates o v er t he r eductio n
in sensitivity (Equation 11) as the floating g a t e ar ea i ncrease s. T he s ame set o f
curves was also measured for the c ase of a contaminated chip, and this plot is
shown in Figure 96. Th e ordering of the curves and the relativ e sensitivities are
consistent with t hose o f the uncon taminated case, and the effec t o f contamin a tion
results in f aster ov erall decay rates as expec ted .
224

Figure 96. Plot of the normalized drain current as a function o f ti me for a c ont a mi nate d HCT C ‐ 6
chi p wi t h f o ur PMIC se nsor s of di f fe re nt g ate and ch ann e l are a s.
Finally, with the HCTC‐6 design all PMIC sensors work as intend ed, and
there a r e no major r emain i ng d esig n issues t hat need t o be a d d r essed. In
addit i on, an understa n ding has be e n ga ined with respec t to floa ting gate scaling,
which represents the last major unknown in the design process. Moving forward,
the emphasis can now turn to optimization of the hermeticity of i mplantable
chips using spare CMOS processing l ayers and post‐processing co atings, as w ell
as more advanced designs for on‐chip threshold detection of cri tical io n
contamina t ion levels.
225

5.4 Hermetic Coating Test Chip 7 (HCTC ‐7)
The seventh hermetic coating test chip (HCTC‐7) is the most rec ently
fabricated and tested design as of the writing of this thesis. T his design s eeks to
expand u pon the exp e r i ments firs t initia ted w i th H CTC‐2 to a nal yz e t h e vary ing
levels of protection afforded by using spare CMOS fabrication p rocessing l ayers,
but with several key improvements. Namely, the suite of experi ments has be en
expanded to include PMIC sensors with six variations of ion contamin a tion
protectio n , including o x ides w ith a passiva tion o xide c ut, a fu ll oxide stack, al l
oxides w ith one metal, a ll oxides w ith two metals, and the latter two also with
passivation oxide cuts. This represents nearly every combinati on o f protection
over a PMIC senso r th a t can be d e s ign e d usin g this fabr i cation process.
In addition, the experimental area on the chip was centralized to a
1 millimeter diameter circle in the center of the chip to make uniform
contamination with a s aline drop much easier to apply. Lastly, the entire suite of
six s e nsors was mir r or ed o n the b o ttom half of the chip with is olated p ower and
ground lines, in order to allow a du plicate of the experiment t o be run i n parallel.
The HCTC‐ 7 c hip was fabrica t ed o n December 2 1, 2 012 , aga i n us in g the M O SIS
service and the ON S emiconductor C5 process (run V29J). A full set o f m e asu red
process parameters after fabrication is provided in Appendix A. The layout and
photomicrograph of t he f abricate d chip i s shown in F igur e 97. A summary of t he
key PMI C d evices t hat were t est e d follows in Table 23.
226

Fi gure 9 7 . The Cade n ce l ayou t (T op ) a n d an o p t i c al mi c ro gr aph ( Bot t o m ) of H ermetic C o a t ing
Test C hip 7 (HCTC‐7).
227

Ta ble 23. S umma ry o f key PMIC d evices t ha t were t est e d on h erme ti c co a t i n g te st c hi p H CT C‐ 7 ,
including size, loca t i on, a n d pr esence o f a pa ssivat ion oxide c ut f or e ach device. A ll six P M IC
sensors were mirrored to c omprise a n a ddit iona l set of s ix s ens ors on t he b o t t o m h a lf o f the c h ip.
PMIC
No.
Gate
Layer
Gate
Area
Channel Size
(W L in µm)
Oxide
Cut?
Metal Overlayer Shield
Location
(x, y in µm from
bottom ‐left corner)
1 Metal 1 108 µm
2
9 µm 3 µ m Yes None 550, 11 3 0
2 Metal 1 108 µm
2
9 µm 3 µ m No None 840, 11 3 0
3 Metal 1 108 µm
2
9 µm 3 µ m Yes Metal 2 Shield 840, 14 2 0
4 Metal 1 30 µm
2
9 µm 3 µ m No Metal 2 Shield 1130, 1 130
5 Metal 1 108 µm
2
9 µm 3 µ m Yes Met a l 2 + Me ta l 3 Shield 1130, 1 420
6 Metal 1 432 µm
2
9 µm 3 µ m No Met a l 2 + Me ta l 3 Shield 1420, 1 130
The first set of d ata coll e cted w as s imply for an H CTC‐7 chip w ith e x t e r n al
contamination applied to the top surface in the same way as all previous
contamination experiments to date. The cases of PMIC sensors w it h field ox id e
only protection (with a passivation oxide cut), with all oxides ( no p assivation
cut), and with Metal 2 + Metal 3 protection layers were compared, and a plot of
normalized d rain c urrent a s a fu nction o f time i s shown in F igu re 98. As e xp ected ,
the Metal 2 + Metal 3 p rotection layers provided the most prote ction to e x t er nal
contamina tion, and r e s u lted i n the least floating gate voltage decay of t he t hree
cases. The next most sensitive PMIC sensor is the case for which all oxides
including the passivation layer are above the floating gate, although a departure
from the decay curve for the PMIC sensor with a passivation oxide cut i s not seen
until a f ter 1 00 hours, a n d even t hen may be w ithin the pro c ess variations seen in
the HCTC‐5 statistical a nalysis.
228

Fi gure 9 8 . P lot of t he n o r mali ze d dr ai n current a s a func tion o f ti me f or a c ont a minate d HCT C‐7
chip cont a ining PMIC sensors wit h v a rying am ount s of p rotec t ion agai n st i on c ontamination.
Perhaps even more interesting is the comparison of these three cases to
the corresp onding unc ontaminated versio ns, as shown in Figure 99. In this case,
it can be seen that when either the passivation layer is intact o r the metal layer s
are pres en t , o r both, the effects of contamina t i on a r e e v i dent but mi ni mal. T he
largest effects of e xternally a pplied contamination a re o bserved for the case o f
field oxide only protection (with a passivation oxide cut). Fu rthermore, t he
uncontaminated curves a lso show t hat the metal‐protected PMIC s ensors d ecay
slower t han those only p rotected w ith oxide layers, indicating that the metal
layers protect against intrinsic ion contamination that is pres ent in t h e
passivation layer or top field oxides after normal fabrication processing, a s well
229

as in or on the top field oxide layers following a passivation oxide c u t. T his has
potent ially r em arkable im plication s, in t hat it c an b e curre ntl y postulated t hat by
adding spare metal overlayers to any VLSI chip design, the herm eticity can be
im proved a gains t ev e n int e rn al contamination effects.
Also worthy of note is the fact that the effects of contaminati on are only
primarily seen by the PMIC sensors that have no metal protectio n layers a nd a n
oxide cut above the sensor. T his observation is consistent wit h previous
observations regarding HCTC‐3, and reinforce the fact that the PM IC s ensors
seem t o bes t d et ect pos i tiv e m obile ions i n thes e exp e riments w hen a passivation
oxide cut is present.
F i g u re 99. P lo t of t he n orma lize d dr ain current a s a func tion of time for both contaminated and
uncon ta minat e d HCTC‐7 c hips con t a ining PMIC s ens o rs w ith v a ry in g am ount s of pr o te cti o n
against ion contaminati o n.
230

5.5 Summary of Key Findings
In this chapter, designs of three additional hermetic coating t es t chips
were presented, sp anning a time period o f an additional 2 years of development
and testing. A number of the additional lessons learned can be c arried f orw a rd
to f uture designs, and t h is s et o f ex perimen t s nearly completes the picture of how
device g eometry and spare processi ng l ay er p rotectio n affects PMI C sens it ivit y.
The following summary attempts to capture a brief description of each chip, along
with key lessons learned and conclusions that are important to future w ork in
this ar e a.
Hermetic Coating Test Chip 5 (HCTC‐5) was the first design inte nd ed t o be
coated w i th a n addition al h ermeti c coating a fter M OSIS f abricat ion. T o t his e n d,
a distributed set o f P M I C sensor s w a s placed aro und the perimet er of t h e chip to
test both PMIC device variation across a single chip, and also the uniformity o f a n
applied coating. W hile some coatings were a pplied to t h i s chip to demonstrate
the feasibility of a post‐processing application of the coating , the key sensors
were designed to be exposed to the air prior to coating (thereb y allowing t he
hermeticity of the coating to be evaluated directly), which was shown to diminish
their ability to e ffectively h ol d a charge, ev en w hen o v ercoat e d aft e r in it ial
fabrication. However, this chip was used to characterize devic e var i ations a mong
oxide covered P MIC sensors on f ou r different uncontaminated chi ps, and th e
statistical variation of t he n or malized drain current d ec ay r at e was found to b e
about ±6% after 100 hou rs in the oven a t 180 °C.
231

The final two test chips, HCTC‐6 and HCTC‐7, represent the state of the art
in PMIC sensor design, and all major design issues in previous HCTC c hip designs
have b een corrected. H CTC‐6 was used t o characterize t he d evic e‐to‐device
variatio n o n a s ingle chip b y anal yzing four i dent ical P M I C se n sors t ested on t h e
same chip th at were separated by a dista nce of approxim atel y 1. 5 m m f rom each
other. The variation in this case was also found to be about ± 6% a fter 100 hours,
and it w as a lso deter m ined t hat the variatio n is s lightly highe r at a round ±9%
after 10 0 hours for a chip that wa s contamin a ted w i th a s aline drop.
The key experiment conducted with HCTC‐6 was a comparison of th e
floating gate voltage decay rate s of PMIC sensors with differen t floating gate a n d
FET channel areas. A m odel w as p r o posed that p redic t s th at a n increase i n decay
rate should be observed with an increase in floating gate area, which was verified
experimentally. The key experiment conducted with HCTC‐7 was a comparison
of the floating gate voltage decay rates of PMIC sensors with different l ev els of
protectio n above the floating ga te using spare oxide and metal layers o f the VLSI
fabrication process. T he p resence of m etal l ayers was shown to improve the
hermeticity of the chip even against intrinsic ion contaminatio n within t he
passivation and top field oxide layers. In addition, the PM IC sensors that h ad n o
metal protection layers and that had a passivation oxide cut were the only ones
that s howed a signific ant increase i n d e cay rate i n response t o e xternally a pplied
saline contamination.
232

Both H CTC‐6 and HCT C ‐7 d esigns w ere also i nt ended to b e test ed with
post‐processing coatings applied in an attempt to enhance the her meticit y, in
order to a llow for the assessmen t of such coatings. T hese c hip s are the focus of
continuing research o n hermetic coatings as a n extens ion of t he w ork already
published on h ermetic coatings wi th h ermetic coating t est chip HCTC‐5 [5]. Since
this research was published, the HCTC‐6 and HCTC‐7 designs have s uccessfully
been coated and tested, showing similar results in t hat that a m u ltilayer d iel e ctric
coating applied after chip fabrication demonstrated a reduction i n the rate o f
positive ion contamination charge accumulation.
However, i t was also d iscovered t h a t some p rocesses asso ciated with the
hermetic coating application may create a quasi‐hermetic layer, o r ma y ev en a lter
the device properties in some cases. One such process is a pre ‐cleaning step t hat
attempts t o enhance the adhesion o f the coatin g to t he s ubstrat e by sli ghtly
etching the surface with a plasma. Figure 100 shows a set of P FET I‐V calibration
curves f or u ncontaminated HCTC‐6 a nd H CTC‐7 chips, several of which h ave been
exposed to the pre‐cleaning step only, as if they are being pre pared fo r a hermetic
coating.
233

F i g u re 100. P F ET I‐V ca li bra t ion curve s f or a n umbe r of u ncont a minate d HCTC‐6 a nd H CTC‐7
chips, several of which have undergone a pre‐cleaning process b efore a hermet i c coa t i ng w as
applied. In se veral c a ses, t h e p as siva tion o xide c ut a p p a rentl y cause s a si g nific a nt altering o f t he
PF ET device cha ract e rist ics.
The calibration curves of measured d rain c urrent as a fu nction of a pplied
gate voltage f or t he P FETs a re r emarkably consistent across dif fe rent d ev ice
geometries and chip designs, and even before and after accelera ted lifetime
testing of the chips. In addition, the pre‐cleaning step does not s i gnificantly
change the I‐V calibration curve for a majority of the devices. H owever, two of
the HCTC‐7 P FET devices that b ot h had passivation o xide c uts above them
showed a n altered calibration curve, a nd i n one case e ven fail e d u p on t esting.
This is a strong indication that care must be taken to verify t ha t the dev i ce
characterist ics have not been chang ed after the oxide cut step, p re‐cleaning step,
234

and before further accelerated testing of the devices, and that l ess aggressive p re‐
processing s teps may be needed w he n working with PMIC devices.
Now, with over four years of knowledge and experience gained wi th
positiv e m obile ion c o ntam in ation sensing, a set of fifth‐gener at ion chips is
envision ed f or future r e sea r ch, as will be discussed in Ch a pter 6.

235

CHAPTER 5 REFERENCES
[1] T. S tieglitz, “Manufacturing, assembling a nd p ackaging o f m iniatu rized
neural i mplants,” Microsystem Technologies , vol. 16, n o. 5 , pp. 723–734,
2010.
[2] E. C . Gill, J . Antalek, F . M. K i m ock, P. J . Nasiatka, B. P . M cI ntosh,
A. R. Tanguay, Jr, and J. D. Weiland, “High‐Density Feedthrough
Technology f or H ermetic Biomedical M icropackaging,” Materials Research
Society Proceedings , v o l. 1572, C ambridge Univers ity Press, p. m rss13‐
1572‐ss05‐ 08, 2013.
[3] K. N ajafi, “ Micropackaging T echnologies for Integrated M icr osystems:
Applications to MEMS and MOEMS,” in SPIE Proceedings on
Micromachining and Microfabrication Process Technology VIII, vol. 1, pp.
1–19, 2003.
[4] D. C . Rodger, A. J . Fong, W. L i, H . Ameri, A. K. Ahuja, C. Gutierrez, I. Lavrov,
H. Zhong, P. Menon, an d E. Meng, “Flexib le par yl ene‐ base d multi elect r ode
array technology f or h igh‐densit y neural s timulation a nd r ecord ing,”
Sensors and Actuators B: Chemical , vol. 132, no. 2, pp. 449–460, 2008.
[5] J. D. Weiland, F. M. Kimock, J. E. Yehoda, E. Gill, B. P. McIntosh, P. J. Nasiatka,
and A. R. Tanguay, Jr., “Chip‐Sc ale Packaging for Bioelectronic s Implants,”
in 2013 International IEEE/EMBS Conference on Neural Engineering , pp.
931–936, 2013.

236

CHAPTER 6
ADVANCED INTRAOCULAR CAMERA LENS DESIGNS
6.1 Modelling of the Biological Cornea for Intraocular Camera Testing
As discussed previously in Section 1.4, the current fourth gene ration
intraocular camera design has been fabricated and tested for us e in conjunction
with retinal prostheses. The current design, shown in Figure 101, has a lens
diameter of 2.8 mm and a total optical system length of 3.51 mm [ 1–10]. T he
custom polymer aspherical lens itself is only 2.2 mm thick and has a mass of
9.0 mg. The focal length of this lens is approximately 2 mm, a nd t he f ‐number is
approximately 1.0.

F i g u re 101. T he fourt h g e nera t i on int raocula r ca m era ra y d i ag ra m, f rom [3].
This fourth generation IOC lens design was optimized using the cornea
and aqueous humor portions of the Liou and Brennan eye model as i mplemented
237

in CODE V [11]. The design was also optimized for an image plane without a cover
glass betw een t h e l ens and the sensor. B oth of t hese o ptimizat ions p os e
problems when attempting to validate the design on a lab bench, because most
image sens ors do in fact have a gl ass layer protecting the sens or surface, and the
images are typically taken in air and not proximity‐coupled to a biological cornea.
In order to simulate the camera performance when coupled with a
com m e rcial im age s e ns or a rra y, t h i s ex ist i ng d esig n was used a s the basis for a
new design that included the actual dimensions of the OV6920 im age s e ns o r
array, including the sensor cover glass. The focal distance wa s the n o ptim ize d
using the CODE V lens design sof tware (Synopsis, Pasad e na, CA). When the IOC
is coupled with this particular image sensor, the design has an effective
resolution o f 50 by 50 pixels, derived from a 24 micron s pot si ze a veraged acros s
all field angles up to 20 degrees and assuming a square active image pl ane reg i on
of 1.2 mm on a sid e.
In order to solve the issue of imaging performance measured on a lab
bench, a n equivalent c ornea mod el was designed f rom commerci a ll y available
lenses t o pr oduce performance eq uivalent t o that s imula t ed w ith with the Liou
and Brennan cornea and aqueous humor models. Through an iterativ e des i g n
approach, it w as f ound t hat two commercial lenses p laced in seq uence (Thorlabs
LF1510‐A and LA1472‐A) could approx imately replicate the MTF p e rformance
and spot sizes a cross the fields of interest, and could therefore act as a suitable
benchtop eye model when testing intraocular camera designs. Th e ray diagram
238

of this equivalent eye model in conjunction with the fourth gen e rati o n IOC le ns
and the OV6920 image sensor is s h own in Fig ure 102.
F i g u re 102. The f ourt h g e nera t i on i nt raocula r camera coupled w i th t he OV6920 imag e sensor
array, a n d t he e quivalent biologic al cornea model co mp o s ed o f t w o c ommercially a vailable l enses.
The MTF p e rformance of t his desi gn, which can be realized on a lab bench,
was compared with the MTF performance of the original fourth ge neration I O C
design including the original Liou and Brennan eye model compon ents, as
modified to include the OV6920 im age sensor. The two M TF plots are compared
in Figure 103, and only small differences between the plots can be observed. A
similar co mparison w as m ade for the RMS spot s ize at e ach field a ngle i n th e
rang e of 0 t o 20 degr ees in 5 d eg ree incr emen ts. T hese r esults are summarized
in Table 24 below.
239

F i g u re 103. M TF plot s o f the orig ina l i ntra ocula r c a me ra desig n with the Liou and Brennan eye
model a n d t h e OmniVision O V6920 ima g e sensor ( Top) c ompa red wit h t h e modi f i ed d esig n using
the biolo g ic al cornea m odel (Bo t t o m). O n l y minor vari a t ions c a n be o bserved when c ompa ring
the tw o se ts o f plo t s.
240

Table 24. Summary of RMS spot sizes for various field angles when comparing the original
int raocula r camera d esig n using t he L i ou a nd Brennan eye model and t h e OmniVision O V6920
i mage se ns or, the s a me c am e ra de si gn i m a gi n g i n ai r (n o corne a/ aqu e ous hu mor m odel), and the
same c amera design u sing t he cornea repl acement m o d e l.
Field
Angle
RMS Spot Size with Liou and
Brennan Cornea Model
RMS Spot Size with no
Cornea Model (in Air)
RMS Spot Size with Cornea
Replacement Model
20° 35 µm 37 µm 32 µm
15° 29 µm 33 µm 28 µm
10° 27 µm 28 µm 25 µm
5° 18 µm 19 µm 18 µm
0° 10 µm 15 µm 10 µm
In a ddition t o the comparison o f MTF plot s and spot s izes, a se t of
simulated images was compiled to visually illustrate the benefit of adding the
cornea r e placement m o del when compared with i maging o nly in a ir in a lab
setting. A s expected f rom the spot s izes, an i mprovement c an easily be seen when
comparing fields close to the center of the image, and modest i mprovement c an
even be s e en toward the edg e s.
To valida t e these simulations, a lab setup was creat e d to t est a fabric ated
IOC lens i n conjunction w ith the cornea replacement m o del and t h e OV6920
image sens or a rray. The experime ntally c aptured images c losely match the
simulated images, and show the same expected improvement. The simulated and
captured images are summarized in Figure 104. The success of t his experiment
not only validates the simulation of the fourth generation intr aocular camera
used i n conjunction w ith a sma ll c o mmercially a vailable i mage sensor, but also
241

validates the use of the replacement cornea model in the lab when testing the IOC
performance.
CODE V Simulated Im ages Experimentally Captured Im ages
Wi thou t C o rn e a Re p l a ce me nt Wit h Cornea Repla c ement
F i g u re 104. Compa rison of t he f ourt h genera t i on i nt ra ocula r c am era imag ed i n a i r wit hout the
cornea repl ac ement mo del (Top R ow ), a nd a gain i ma ged in a ir w it h t h e corne a repla c emen t
model (Bottom). CODE V simulated images are shown (Left Column ), a s well as e xperimentall y
cap t ured images ( R igh t C ol umn).
A focus on f urther r edu c ing the siz e a s well as i ncreasing t h e perform ance
of the IOC lens design itself is pr e s e nted in the following s ec tion.
6.2 Next Generation Intraocular Camera Lens Designs
Efforts to further reduce the size and mass of the existing int raoc ular
camera lens design have produced promising fundamental results for future
biomedical c ameras. D esig n res e a r ch h as b een completed t o show that as the size
242

of t hese o p t ical s ystems i s decreased, t he s pot sizes al so d ecr eas e s ignificantl y
while m a in tain ing an e quivalent f‐number. Because new commercial image
sensors have pixel sizes on the order of 3 µm, the light can be focused onto a
smaller area than previously designed, allowing for the preserv at ion of l ight
throughput. Figure 105 shows one such “fifth generation” intraocular camera
desig n t ha t is s im ila r t o the fourth g ene r at ion desig n , but wit h a lens t hat is o nly
1.1 mm i n d i ameter and 1.5 mm thi c k and wi th a n f‐number of 1.18. The total
field‐of‐vi e w is 4 0 ° with spot sizes down to 14 µm on‐axis. The quartz window
in front of the lens is still required in order to maintain an air gap th at e nables
enough refr a cting pow e r at t he f ro nt surfac e of the Zeonex E48R polymer lens.
A second alternative d e sig n i s shown in Figure 106. This design fea tures
a single a spherical lens m ade from E CO‐550, a h igh‐index biocom patible glass
that is only 0.44 mm in diameter and 0.56 mm thick with an f ‐number of 1.30.
The en tir e o ptical s yst e m length, including th e image sen s or c o ver gl ass, is a bout
1.1 mm and covers a total field‐of‐view of 4 0 ° with spot sizes down to 2 µm on‐
axis. Although the reduction in spot size is significant compared to the fourth
generation design, the image projected onto the image sensor array i s also
reduced in s ize, y ieldin g an e ffective resolutio n of about 50 b y 50 pixels.
243

F i g u re 105. One pot ent i a l n ext ‐ g e nerat i on i nt raocular ca mera i m a g i n g syst em l ens de sig n t o be
used w it h t h e OV6920 ima g e sensor a rra y. The l ens ra y diag ram (T op ), a s i m ulate d i m a ge
(Bo t tom, Le f t) , and a s p o t di a gr am ( Bo tto m , Right) are s hown. The f i eld of view in t he simulat ed
ima g e is ±20°, wit h o n a x is s pot sizes of a pproximat ely 14 µm. This design retains the windo w
place d i n fr on t o f t he le n s to mai n t ai n an a i r ga p i n f ron t of the le ns.

244

F i g ure 106. A p ot ent i a l ne x t ‐ g e nerat i on b ioimpla nta ble ima g ing sys t em l ens desi gn b ased o n the
OV6920 ima g e sensor. T he l ens ra y d i a g ra m (Top), a s i m ula t e d i m a ge (Bo tt om, Le f t ), an d a sp o t
di a g ra m (B o t t o m, R i g h t ) are shown. The fi eld of vi e w i n t he s i mulat e d ima g e is ±20°, wit h o n axis
spot sizes of approximately 2 µm. This design has no window in f r o nt o f the l e ns, but the i nde x
difference is large en o ugh to all ow for g ood im aging.

245

Because th e lens i n th e smallest d esig n is m a d e of b ioco mpatibl e glass, it
can be s afel y placed d ir ectly in contact w ith the aqueous humor o r other bodily
fluids. E ven though t his glass has a higher d ensity t han the p olymer m aterial used
in the fourth generation IOC design, the mass of this ultramini a t ure lens i s ev en
lower than t hat of t he c ustom po ly mer lens. A lthough the optic al system shown
in Figure 106 is designed for imaging of relatively distant obj ects, furth e r d e signs
are being e v aluated fo r proxim ity im agin g of o bjects s uch as t u mors, organs, and
tissues. One approach is to design a built‐in standoff distance, w hereby good
im agin g pe rform a nce can st ill be a chieved ev en i f t i ssues come in c ontact w ith
the front window of the camera system. This can be achieved ei th er b y increas i n g
the power of t he a sph e rical lens b y modification o f th e surface profile, or by
adding a s econd lens. A dditional future research d irections ar e d i scussed in t h e
following chapter.

246

CHAPTER 6 REFERENCES
[1] P. J . Nasiatka, M. C . Hauer, N . R. B . Stiles, A. R . Tanguay , Jr., and
M. S. Humayun, "Intraocular Camera for Retinal Prostheses", U.S. Patent
8,197,539 B 2, issued June 12, 2012.
[2] M. C . Hau e r, P . J. N asiatka, N . R. B . Stil es, J.‐C. Lue, R . Agrawal,
J. D . Weiland, M . S. H umayun, and A. R . Tanguay, J r., “Intraocu lar Camera
for Retinal Prostheses : Optical Desig n ,” Annual Meeting of the Optical
Society of America, San Jose, California, 2007; FiO/LS/OMD Conference
Program , Paper FThT1, p. 142, 2007.
[3] M . C. H auer, “Intraocular Camera for Retinal Prostheses: R efract iv e and
Diffractive Lens Systems,” Dissert ation, University o f Southern Califo r nia,
Los Angeles , Califo r nia, 2009.
[4] P. J . Nasiatka, A. A huja, N. R . B. S tiles, M. C . Hauer, R . N. A grawal, R. F reda,
D. G üven, M . S . Humayun, J . D. W eil a nd, and A. R . Tanguay, J r., “Intraocular
Camera f or R etinal P rostheses”, Investigative Ophthalmology and Visual
Science , vol. 46, no. 5, ARVO Abstract 5277, P aper B480, 2005.
[5] P. J . Nasiatka, A. A huja, N. R . B. S tiles, M. C . Hauer, R . N. A grawal, R. F reda,
D. G üven, M . S . Humayun, J . D. W eil a nd, and A. R . Tanguay, J r., “Intraocular
Camer a D esign for Retinal Prostheses”, Annual Meeting of the Optical
Society of America, Tuscon, AZ, 2005; FiO/LS Conference Program , Paper
FThI4, 124, 2005.
[6] P. J . Nasiatka, M. C . Hauer, N . R. B . Stiles, L. L ue, S. T a kahash i, R . N. A grawal,
R. F reda, M. S . Humayun, J . D. W eiland, and A. R. Tanguay, J r., “ Intraocular
Camera f or R etinal P rostheses”, Investigative Ophthalmology and Visual
Science , vol. 47, no. 5, ARVO Abstract 3180, P aper B554, 2006.
[7] P. J . Nasiatka, M. C . Hauer, N . R. B . Stiles, L. L ue, S. T a kahash i, R . N. A grawal,
J. D . Weiland, M . S. H umayun, a n d A. R . T a nguay, J r., “An Intraocular
Camera f or R etinal P ro stheses”, Proceedings of BioMed , Irvine, Califo rnia,
7‐8 June, 2006.
[8] P. J . Nasiatka, B. P . McIntosh, N. R . B. S tiles, M. C . Hauer, J. D. Weiland,
M. S . Humayun, a nd A . R. T anguay, Jr., “An Intraocular Camera f or
Provision of N atural F oveation i n Retinal Prostheses”, Proceedings of the
2010 Neural Interfaces Conference , Long B each, California, 21‐23 J une,
2010.
247

[9] N. R . B. S tiles, B. P . McI n tosh, P. J . Nasiatka, J. D . Weiland, M. S. Humayun,
and A. R . Tanguay, J r., “Intraoc ular C amer a for Retinal Prosth e ses:
Psychophysical Analysis of Image Sampling and Filtering”, Proceedings of
the 2010 Neural Interfaces Conference , Long Beach, Califo rnia, 21‐23 June,
2010.
[10] N . R. B . Stiles, B. P . McIntosh, P. J . Nasiatka, M. C . Hau er, J. D . Weiland,
M. S. Humayun, and A. R. Tanguay, Jr., “An Intraocular Camera f or R etinal
Prosthes es: Restorin g Sight to t he B lind”, i n Optical Processes in
Microparticles and Nanostructures: A Festschrift Dedicated to Richard
Kounai Chang on His Retirement from Yale University , World Scientific
Publishing Company, Singapore, ch. 20, pp. 385–430, 2010.
[11] H. L. Liou and N. A. Brennan, “Anatomically accurate, finite model eye for
optical modeling,” Journal of the Optical Society of America A: Optics, Image
Science, and Vision , vol. 14, no. 8, p p . 1684–1695, 1997.

248

CHAPTER 7
SUMMARY AND FUTURE RESEARCH DIRECTIONS
7.1 Summary
Although blindness is widespread and largely incurable, advance s in
visual p rostheses have enabled e ntirely new areas o f r esearch i n ophthalmology
and biomed ical e ng ineering. In this t hesis, key advanc es i n vi sual p rostheses in
general, and i n intraoc u lar retinal prostheses w ith extraocular a nd i ntr a ocular
cameras in p articular, w ere presented along with a ssociated r es ea r c h to f urther
improve th ese a n d ot her implan table biomedical d evices. S ever a l unique
challenges w ere pres ented in t er ms o f both p hysical size a nd i m aging constraints,
particularly in the low pixellation limit. Additional challeng es w ere discussed
that are faced by any microelectronic device that is implanted in the body, such
as contaminatio n from positiv e mobile ions.
The firs t half o f the thesis e xp lored advances i n image pr ocess ing th at c an
be i mplemented i n a realti me system to si m ulate such vi sual pro stheses, in o rder
to better experience and understand what patients with visual p rostheses may
be capable of, and to aid in the design of optimized intraocula r retinal prostheses.
A nov e l har d ware a nd s oftwar e implementat i on w as p r e s e nt ed t ha t can not only
simulate p rosthetic vision, but can do so f o r both R etinitis P igmentosa and Age‐
Related Macular Degeneration, an d in b oth eye‐pointed a n d head‐ pointed camera
modes. A set of visual psychophysics experiments was presented using the visual
249

prosth esis sim ulator in order to quantify th e bene fits of fo vea tion affo rded by an
intraocular camera (IOC) in several ways. A well‐known visual search t ask was
performed for the firs t time w it h prosthetic vision s i m u lation and compare d to
historical visual search d ata. The results from t his visual s e arch t ask showed t hat
provision o f f ov eation e ngenders s ignific a nt i mprovements in t h e time t o f i nd a
targ et o bject amongst a field o f d is tractor objects in t he c as e of both RP and AMD,
and fo r sev e ral pixellation l ev el s c h aracteristic o f current a n d future i ntraocular
retinal pros theses.
A fu nc tio n a l reac h and grasp task r e l ying o n m o no c u l ar de p t h p e rception
was des i gn ed t o further a n alyze the capabilities o f a vis u al p r osthesis i n both
head‐point ed a nd e y e ‐pointed m odes, and with var ying l ev els of bl ur and
pixellatio n , for both R P and AM D simulated conditions. A compa rison of e ye‐
pointed and head‐pointed m odes w hile p erforming a reach and gra sp t ask
showed t hat the eye‐pointed mode o ffer s a s ignifica nt i mproveme nt o ver the
head‐pointed m ode for all cases in a 24 40 resolution simulation, and in 3 of 4
cases in a 12 20 resolution simulation. Improvement was observed when
comparing b lurred with u nblurred percepts o nly for the case o f 12 20
resolution simulation in the RP mode with an eye‐pointed camera . T his suggests
that w hile a l ower r es olution elec trode arr a y would ben e fit mos t gr eatly from
optimal electrode blurring f or t h i s particular t ask, add i ng f o v eation almost
always affo rds a sig n ificant impro vement.
250

A significant improvement was also observed in both the reducti on o f
obstacle collisions and an i ncre as e in w al ki ng sp eed w h e n u sing th e e ye‐ p o i nte d
camera mode instead of the head‐pointed camera mode during a na vigatio n t ask.
The improv ement in w alking speed and o b sta c l e avo i d an c e is ap p r oximately the
same in both full‐resolution and 24 40 array resolutio n s. T h i s indicates that
the improvement is largely driven by the added benefit of fovea tion, a l though t he
effects o f b lurring a nd e ven lower electrode resolutions have n ot y et b een t est e d .
The second half o f the thesis f ocu sed on i mprovements to i mplan ta ble
devices in g eneral, inc l uding the design a nd t esting o f a posit iv e m o bile i on
contamination senso r that may be integrated with future implant able d evices t o
provide early warning of l eaks or c ontamin a tion t hat may jeopardize t he i mplant.
A compact, low power integrated C MOS cl ock c ircuit c apable o f d ri ving a
commercial image sensor array was also presented, along with se veral key
exper i m e nt s proving th e utility o f s uch devices.
The first hermetic coating t est chip e xplored several sensors p otentially
sensitive to m oisture and positi ve m obile ion contamination. S ome of t he k ey
devices fab r icat ed a nd t ested inc l ude an E P R OM‐based p ositive m obile ion
contamination ( PMIC) sensor, inte rdigitated e lectrodes for dete cting moisture,
and clock generation circuits based on a ring oscillator to explore l ow‐power
clocking of b ioimplantable image sensors. T he r ing oscillator designs were
shown to work near the designed frequency, and all variants fab ricated were
operational. The PMIC sensors showed a sensitivity to positive m obile ions, a n d
251

were proven to be viable for this application. Both the freque ncy and waveform
in the second CMOS clock test chip were sufficiently stable to drive the clock
circuit of an OmniVision OV6920 CMOS image sensor, and to captu re i mages from
a development board u s ed in conju nction w ith the IOC pol ymer le ns.
The third‐g e ne rat i on H CTC‐3 chip e xpanded on p revious experimen ts
varying the levels of protection afforded by spare dielectric C MOS processing
layers a gainst p ositive mobile i ons, and thes e exper i ments wer e e ven further
developed in f ourth‐generation H CTC chip d esigns. H ermetic coa ting t e st chi p
HCTC‐5 was used to characterize the device variations among fou r differ e n t
uncontaminated chips, and th e statistical v a ria t ions i n the nor malized drain
current d ecay r a t es w ere found to b e about ±6% a ft er 1 00 hours in t he o ven at
180 °C. H ermetic coating t est chip H CTC‐6 was used t o characterize the de vi ce ‐
to‐device v a ria t ion on a s ingle chip w ith th e a n alysis o f four ide n tical PMIC
sensors tested on the same chip, each separated by a distance o f app r oximately
1.5 mm from the others. The variation in this case was also fo und to b e about
±6% af t er 1 00 h o u rs. It w as al so dete rm i ned t h at th e var i atio n is slightly higher
at a round ± 9 % after 10 0 hours for an H CTC‐6 chip t hat was conta minated wi th a
saline d rop.
Another key experiment conducted with HCTC‐6 was a comparison o f the
floating gate voltage decay rate s of PMIC sensors with differen t floating gate a n d
FET channel areas. A m odel w as p r o posed that p redic t s th at a n increase i n decay
rate should be observed with an increase in floating gate area, which was verified
252

exper i ment ally. A k ey e xper imen t conducted w ith her m etic coat i ng t est chip
HCTC‐7 was a comparison of the floating gate voltage decay rate s of P MIC sensors
with d iffer e nt l evels of p rotectio n above the floatin g g ate using spare oxide and
metal layers of the VLSI fabrication process. It was shown tha t the presence o f
metal layers improved the hermeticity of the chip even against intrinsic ion
contamination within the passivation and top field oxide layers. Alth ough many
signific ant find ings w ere pr esen ted tha t a r e a pplicable t o a w ide range of
intr aocular retinal pros theses, visual p rostheses, and even i mplantable d evices
in g en eral, a number o f future r es earch direc t ions h a v e been e nvisioned and are
presented in the following sec tion.
7.2 Future Research Directions
Even at the time of this thesis writing, a number of research a ctivities in
the areas of human subject testing with the visual prosthesis s imulator, the
desig n and t esting o f next‐gener ation hermetic coating test chi ps, and the design
of next generation implantable imaging systems are currently underway. Some
of t he p lan n ed a c t iv ities ar e pres ent e d her e , particularl y t hos e tha t d ir ectly
extend this thesis work to further improve visual prostheses an d implantable
devices. F irst, experiments in further h uman s ubject t esting w it h the visua l
prosthesis simulator are currently underway. In order to fully quantify the
benefits o f foveation, a n additional s et o f reach and gr asp exp eriments i s plann e d
to t est even l ower r es olution elec trode a rray s, such a s a 6 10 co nfigur atio n
253

sim i l ar to t h at o f th e A rgu s II de v i c e th at is c u rren tl y c o m m e r cially a vailable. T o
expand u pon the pilot experiment f or n avigation and m obility, a f ull‐scale
exper i ment i s planned that w ill collect d ata for many m ore than the 4 participants
presented in Section 3.4. In addition, a much wider range of e xperimental
configurations is p lanned f or t he se e xper im ents, includ ing both RP and AMD
modes, as w ell as s everal l ower r esolution array configurations with varying
amounts of blur.
Second, m iniatur i za tio n o f the vi sual p rosthesis simulator shou ld b e
continued i n an e ffort t o constru ct a compact VPS t hat can be fully worn by the
subject and have e nhanced mobili ty, esp e cially o utdoor s. N ot o nly will this
enable a new class of outdoor navigation and mobility experimen ts, but it a lso
may lead t o a new ther apeutic treatment option i n which an e ye tracke r can be
added to e xisting commercial visual p rostheses so t hat foveatio n c a pability c an
be restored . Research has been p ub l i sh ed sh o w i ng th at e ven th o ugh some b lind
patients have not had conscious eye movements due to foveation for many years,
some r etain good conscious contr ol, and oth e rs c an b e retrained t o have m ore
predict a ble eye movements [1–6 ]. T his is a n importa n t ar ea o f research t o
explore in c onjunction w ith ey e t r a c king d evic es a n d t he a bilit y for blind patients
to perform a reasonably accurate eye tracking calibration if an eye‐tracked EOC
is to be feasible. In addition, automated slip correction algorithms and auto‐
calibration routines w ould b e needed, a s w ell as a dapt ability f or a n umber of
lighting conditions for all‐day wear, each o f which is a n area of a ctive research
254

[7–10]. A set of human subject experiments can also be envisioned that m akes
use of many of the settings and parameters of visual prosthesis s imulation th at
were not yet tested, including varying amounts of latency, elec trode dropout,
varying fill factors, and even e lectr o de shape.
Third, n ex t gen e ra tio n h ermetic coating c hip design s are alr e ad y
envision ed t hat build upon t he l es sons l ear n ed w ith the first four g e nerations o f
PMIC and humidity sensors. An integrated threshold detection c ircuit f or
breaches i n hermeticit y or t he d etectio n o f deteriora t ing herme ticity m ay b e
implement e d, such as that shown in Figure 107, which is a low b attery detect i on
circuit from t he Intersil ISL21440 microchip. T his comparator circuit is c urrently
being used o n an i nteg rated circuit to d etect low battery volta ges, but can be
adapted to d etect low PMIC f loating gate vol tages witho u t much modification.
Having s uch a circuit integ r ated w ith a PMI C c ircuit w ill enabl e a new class o f
stand a rd VLSI cells that can be integra ted w i th all future impl antable devices.

F i g u re 107. Example sche mat i c diag ram of a l ow b at t e ry v oltag e detection circuit including a
comparator that can be repurposed as a low gate voltage detecti on c ircuit i nteg ra t e d with P MIC
sensors. From Int e rs il ISL21440 da t a sheet .
255

Fourth, alt h ough P MI C sensors have s hown great p romise, it i s a lso
desir e d to c ontinue the development of i ntegr a ted s e nsor s that a r e s e nsitiv e to
moisture. Interdigitated electrodes have been shown to be sens it iv e to s urface
moisture condensatio n, but it ha s proven difficult to calibr ate these devices with
precise lev e ls o f condensation. A n ew E SPEC S H‐241 precision h um idity oven h as
been installed in conju nction w it h a set of P eltier coolers tha t can be mounted t o
the HCTC p ackages. T his combination w ill al low for precise con trol o f the local
tempera t ure differen tial of the chip within th e oven r elat ive to the oven ambient
temperature, and therefore control of the amount of condensation on the surface
of t he c hip. A photograph o f on e such P eltier t hermoelectric c ooler m ounted o n
the underside of HCTC‐4 is shown in Figure 108. An aluminum he atsink i s needed
to m aintain a temper ature differential between the ambient oven e nviro n ment
and the sur f ace of t he c hip, a nd a s mall thermistor c an b e seen o n the right s i de
to provide f eedback to a thermoelectric contro ller module.

F i g u re 108. P hot o g raph of a n HCTC ‐4 chip wit h a Pel t i er cooler m o u n te d to t he u n d e rsi de o f the
ceramic package, including an al uminum h eat sink and thermistor f or f eedbac k con t r ol.
256

Lastly, sinc e a small integrated C MOS clock generation circuit has been
demonstrated to successfully drive an OmniVision OV6920 image s ens o r array, a
new s e t of f ifth‐gener ation intr aocular lens d esigns h as b een e xplored, a nd w ill
continue to b e optimized for n e xt‐ generation implantable camera s. I ndeed, the
com binatio n o f a polym e r lens o ptim ized f or t he O m n iVis ion OV69 30 ( an
advanced version of the OV6920 with lower power dissipation), a small VLSI
CMOS clock generation circuit, and a hermetic seal detection ci rcuit featuring a
PMIC sensor would form the basis of a complete package for adva nced
implantable cameras. U ltraminiat ure cameras such a s t h is h ave a number o f
potent ial a pplications for b iom e dical im aging, i ncluding e ndosc opes, chronically
implantable diagnos t ic c ameras, and intr a ocular cameras.

257

CHAPTER 7 REFERENCES
[1] R. J . Leigh, S . E. T hurston, R . L. T o m sak, G. E . Grossman, and D. J . Lanska,
“Effect o f monocular visual l oss upon s tability o f gaze,” Investigative
Ophthalmology and Visual Science , vol. 30, no. 2, pp. 288–292, 1989.
[2] R. J. Leigh and D. S. Zee, “Eye movements of the blind,” Investigative
Ophthalmology and Visual Science , vol. 19, no. 3, pp. 328–331, 1980.
[3] K. R. Sherman and E. L. Keller, “Vestibulo‐ocular reflexes of a dven titiously
and congenitally b lind a dults,” Investigative Ophthalmology and Visual
Science , vol. 27, no. 7, p p . 1154–1159, 1986.
[4] E. D. Cohen, “Prosthetic interfaces with the visual system: Biological
issues,” Journal of Neural Engineering , vol. 4, no. 2, pp. R14–R31, 2007.
[5] E. Kasten, U. Bunzenthal, and B. A. Sabel, “Visual field re covery after vision
restor atio n therapy (VRT) is i n d epend e nt o f eye mov e ments : A n eye
tracker study,” Behavioural Brain Research , vol. 175, n o. 1, pp. 1 8–26,
2006.
[6] D. Kömpf and H.‐F. Piper, “Eye movements and vestibulo‐ocul ar r eflex in
the blind,” Journal of Neurology , vol. 234, no. 5, pp. 337–341, 1987.
[7] W. J . Ryan, A. T . Duchowski, and S. T . Birchfiel d , “Limbus/ pupil switching
for wearable eye tracking under va riable l ightin g conditio n s,” in
Proceedings of the 2008 Symposium on Eye Tracking Research and
Applications , pp. 61–64, 2008.
[8] K. P . White, J r., T. E . Hutchinson, and J. M . Carley, “Spat ially d yn am ic
calibration of a n eye‐tracking system,” IEEE Transactions on Systems, Man,
and Cybernetics , vol. 23, no. 4, pp. 1162–1168, 1993.
[9] L. Dongheng, D. Winfield, and D. J. Parkhurst, “Starburst: A hybrid
algorithm for video‐based ey e trackin g combining f eature‐based and
model‐based approaches,” i n 2005 IEEE Computer Society Conference on
Computer Vision and Pattern Recognition Workshops , p. 79, 2005.
[10] A. J. Hornof and T. Halverson, “Cleaning up systematic error i n eye‐
tracking data by using required fixation locations,” Behavior Research
Methods, Instruments, and Computers , vol. 34, no. 4, pp. 592–604, 2002.

258

BIBLIOGRAPHY
J. M . Andrews, Jr., N. L ifshitz, and G. S molinsky, "M ethod of M anufacturing
Semiconductor Devices Involving the Detection of Impurities," U .S. Patent
4,978,915 A, issued D ecember 18, 1990.
M. P . Barry a nd G . Dagnelie, “ U se o f the Argu s II retinal prost hesis to i mprove
visual guida nce of f in e hand m ovements,” Investigative Ophthalmology and Visual
Science , vol. 53, no. 9, p p . 5095–5101, 2012.
W. D . A. B eggs, “Psyc h ological correlates o f w alking s peed i n t he visually
im paired,” Ergonomics , vol. 34, no. 1, pp. 91–102, 1991.
K. M . B. B ennett and U. C astiello, Insights into the reach to grasp movement ,
vol. 105, Elsevier, Amsterdam, The Netherland s, 1994.
R. B ez, E. C amerlenghi, A. M odelli, and A. Visconti, “Introduct ion to f lash m em ory,”
Proceedings of the IEEE , vol. 91, no. 4, pp. 489–502, 2003.
C. B rauer‐B u rchardt and K. Voss, “ A n ew a lgorithm t o correct f i sh‐eye and strong
wide‐angle‐lens‐distortion from single images ,” in 2001 International Conference
on Image Processing , vol. 1, pp. 225–228, 2001.
N. M . Bressler, S. B . Bressler, and S. L . Fine, “Age‐related m a cular degeneration,”
Survey of Ophthalmology , vol. 32, no. 6, pp. 375–413, 1988.
G. S. Brindley, “Number of information channels needed for efficient reading,” The
Journal of Physiology , vol. 177, no. 2, p. P44, 1965.
G. S . Brindley a nd W . S. L ewin, “The s ensations produced b y ele ctrical stimulation
of t he visual cortex,” The Journal of Physiology, vol. 196, n o. 2, pp. 479–493, 1968.
K. Canola, B. Angénieux, M. Tekaya, A. Quiambao, M. I. Naash, F. L. Munier, D. F.
Schorderet, and Y. A rsenijevic, “Retinal S tem Cells T rans plante d into M odels of
Late Stages of Retinitis Pigmentosa Preferentially Adopt a Glial or a Retinal
Ganglion C ell Fate,” Investigative Ophthalmology and Visual Science, vol. 48, no. 1,
pp. 446–454, 2007.
A. Caspi, A. Roy, G. Consedai, R. Greenberg, A. Safran, and J.‐ A. S ahel, “Retinal
Prosthes is ‐ S teering t h e Line o f Sight with E ye M ovements”, in 36
th
Annual
International Conference of the IEEE Engineering in Medicine and Biology Society,
2014.

259

K. Cha, K. Horch, and R. A. Normann, “Simulation of a phosphene ‐based visual
field: Visual acuity in a pixelized vision system,” Annals of Biomedical Engineering ,
vol. 20, no. 4, pp. 439–449, 1992.
K. Cha, K. W. Horch, and R. A. Normann, “Mobility performance w ith a pixelized
vision s ystem,” Vision Research , vol. 32, no. 7, p p . 1367–1372, 1992.
K. Cha, K. W. Horch, R. A. Normann, and D. K. Boman, “Reading s peed w ith a
pixeliz e d vision s ystem , ” Journal of the Optical Society of America A: Optics, Image
Science, and Vision , vol. 9, no. 5, pp. 673–677, 1992.
D.‐R. Chebat, F. C . Schneider, R . Ku pers, and M . P tito, “Naviga ti on w i t h a se nsory
substitution d evice in congenitall y blind individuals,” Neuroreport, vol. 22, no. 7,
pp. 342–347, 2011.
C. Y . Chen, T. Y . Wong, and W. J . Heriot, “Intravitreal Bevacizumab ( Avastin) f o r
Neovascular Age‐related Macular Degen e ra tion: A Short‐term S tud y”, American
Journal of Ophthalmology, vol. 143, no. 3, pp. 510–512, 2007.
S. C. Chen, L. E. Hallum, N. H. Lovell, and G. J. Suaning, “Vis ual acuity m easurement
of p rosthetic vision: A virtual‐reality simulation study,” Journal of Neural
Engineering , vol. 2, no. 1, pp. S135–S145, 2005.
S. C . Chen, L . E . Hallum, G . J. S uaning, and N. H . Lovell, “ A q uantitative analys is o f
head movement behaviour during visual acuity assessment under p rosthetic
vision s im ulation,” Journal of Neural Engineering , vol. 4, no. 1, pp. S 108–S123,
2007.
K. B. Cho, A. I. Krymski, and E. R. Fossum, “A 1.5‐V 550‐μW 176 144
Autonomous CMOS Active Pixel Im a ge Sensor,” IEEE Transactions on Electron
Devices , vol. 50, no. 1, p p . 96–105, 2003.
D. D . Clark‐Carter, A. D . Heyes, and C. I. Howarth, “ The efficiency and walking
speed o f visually i mpaired people,” Ergonomics , vol. 29, n o. 6, pp. 779–789, 1986.
E. D. Cohen, “Prosthetic interfaces with the visual system: Bio logical issues,”
Journal of Neural Engineering , vol. 4, no. 2, pp. R14–R31, 2007.
G. Dagnelie, “Retinal Implants: Emergence of a multidisciplinar y field,” Current
Opinio n in N eurology, vol. 25, no. 1, pp. 67–75, 2012.

260

G. D agnelie, D . Barnett, M . S. H umayun, and R. W . Thompson, “Pa ragraph text
read ing using a pixeliz e d prosthet ic v ision sim u lator: P aram ete r dependenc e a nd
task lear ning i n free‐viewing conditions,” Investigative Ophthalmology and Visual
Science , vol. 47, no. 3, p p . 1241–1250, 2006.
W. H . Dobelle, M. G . Mladejovsky, and J. P . Gir v in, “A rtific ial V is ion fo r the Blind :
Electrical S timulation o f Visual C o r tex O ffe rs H ope for a Funct ional Prosthes is,”
Science , vol. 183, no. 4123, pp. 440–444, 1974.
J. D ongarra, J. L . Martin, and J. W orlton, “Computer benchmarki ng: P at h s and
pitfalls: The most popular way of rating computer performance can confuse a s
well as inform; avoid misunderstanding by asking just what the benchmark is
measuri ng,” IEEE Spectrum, vol. 24, no. 7, pp. 38–43, 1987.
L. Dongheng, D. Winfield, and D. J. Parkhurst, “Starburst: A hy brid a lgorithm f or
video‐based eye tr acking c o m bining f eature‐based a nd m odel‐base d
approaches,” i n 2005 IEEE Computer Society Conference on Computer Vision and
Pattern Recognition Workshops , p. 79, 2005.
A. Doshi and M. M. Triv edi, “Head and ey e gaz e dynamics d uring visual a ttention
shifts in complex envir onments,” Journal of Vision , vol. 12, no. 2, p. 9, 2012.
J. D owling, “Mobility Enhancemen t Us ing Simulated Ar tificial H u man Vi si on,”
Dissertatio n, Queensland Universi ty of Technology, Brisbane, Qu ee nsland, 2007.
J. Dowling, W. Boles, A. Maeder, “Mobility assessment using sim ulated A r t ificia l
Human Vision,” i n 2005 IEEE Computer Society Conference on Computer Vision and
Pattern Recognition Workshops , p. 32, 2005.
M. P . Eckstein, “ Visual s earch: A r etrospective,” Journal of Vision, vol. 11, no. 5,
p. 14, 2011.
J. R . Evans, “Risk factors for ag e‐related macular degeneration ,” Progress in
Retinal and Eye Research , vol. 20, no. 2, pp. 227–253, 2001.
G. J. Farrar, P. F. Kenna, and P. Humphries, “On the genetics o f retinitis pigmentosa
and on m utatio n‐ind e pendent appr oaches t o therap eutic intervent ion,” The
EMBO Journal , vol. 21, no. 5, pp. 857–864, 2002.
G. K . Fedd er, R. T . Howe, T.‐J. K. L iu, and E. P . Quevy, “ Techn ologies for
Cofabricating MEMS a nd E lectronis,” Proceedings of the IEEE , vol. 96, n o. 2,
pp. 306–322, 2008.

261

A. P. Fornos, J. Sommerhalder, B. Rappaz, A. B. Safran, and M. Pelizzo ne,
“Simulation of artificial vision, III: Do the spatial or temporal characteristics of
stim ulus p ixeliza t ion r e ally m atte r?” Investigative Ophthalmology and Visual
Science , vol. 46, no. 10, pp. 3906–3912, 2005.
A. P. Fornos, J. Sommerhalder, A. Pittard, A. B. Safran, and M. P elizzo ne,
“Sim ulation of a r t ificial visio n , IV : Visual i n f or m a tion r eq uired t o achiev e simple
pointing and m anipulation tasks,” Vision Research , vol. 48, n o. 16, pp. 1705–1718,
2008.
E. R. Fossum, “CMOS Image Sensors: Electronic Camera‐On‐A‐Chip, ” IEEE
Transactions on Electron Devices , vol. 44, no. 10, pp. 1689–1698, 1997.
E. R. Fossum, “Digital Camera Sys tem‐on‐a‐Chip,” IEEE Micro, vo l. 18, no. 3, p p .
8–15, May/ J une, 1998.
D. K . Freem an, J. F . R i zz o III, a nd S . I. F ried, “Enc oding visu al i nfo r matio n i n r e tinal
ganglion c ells w ith prosthetic s tim u lation,” Journal of Neural Engineering , vol. 8,
no. 3, p. 035005, 2011.
D. S . Friedman, B. J . O'Co lmain, B . Muñoz, S. C . Tomany, C. M cCarty, de Jong, Paulus
T V M, B . Nemesure, P. M itchell, a nd J . Kempen, “Prevalence of Age‐Related
Macular Degeneration i n the United S tates,” Archives of Ophthalmology , vol. 122,
no. 4, p. 564, 2004.
J. Garcia and M. McBride, "Appara tus for Me a s uring Cont am inan t Mobile Ions in
Dielectric Materials," U .S. Pate nt 6,699,436 B 1, issued M arch 2 , 2004.
E. C . Gill, J . Antal e k, F. M . Kimock, P. J . Nasiatka, B. P . McI ntosh, A . R. T anguay, Jr. ,
and J. D. Weiland, “High‐Density Feedthrough Technology for Her meti c
Biom edical M icropacka ging,” Materials Research Society Proceedings , vol. 1572,
Cambridge Univers ity Press, p. mrss13‐1572‐ ss05‐08, 2013.
G. R . Gilliam, "On‐Chip M obile I o n Contam inat ion T e s t C ircuit ", U .S. Patent
5,790,463 A, issued August 4, 1998.
A. H aj im iri, S . L i m o tyrakis, and T. L ee, “ J i tt er a nd p h ase no i s e in r ing oscillators,”
IEEE Journal of Solid ‐State Circuits , vol. 34, no. 6, pp. 790–804, 1999.
D. T. Hartong, E. L. Berson, and T. P. Dryja, “Retinitis pigmen tosa,” The Lancet,
vol. 368, no. 9549, pp. 1795–1809, 2006.

262

M. C. Hauer, “Intraocular Camera for Retinal Prostheses: Refrac tiv e a n d
Diffractive L ens Systems,” D issert ation, U niversity o f S outhern C alifornia,
Los Angeles , Califo r nia, 2009.
M. C . Hauer, P . J. N asiatka, N . R. B . Stiles, J.‐C. Lue, R . Agr awal, J. D . Weiland, M . S.
Humayun, and A. R. Tanguay, Jr., “Intraocular Camera for Retina l Prosthes es:
Optical Des i gn,” Annual Meeting of the Optical Society of America, San Jose,
California, 2007; FiO/LS/OMD Conference Program, Paper F T hT1, p. 142, 2007.
J. S . Hayes, V. T . Yin, D . Piyathaisere, J. D . Weiland, M . S. H umayun, and G. D agnelie,
“Visually guided performance of simple tasks usin g simulated pr osth etic v ision, ”
Artificial Organs , vol. 27, no. 11, p p. 1016–1028, 2003.
R. W . Hock ney, The Science of Computer Benchmarking, vol. 2, t he S ociety for
Industrial and Applied Mathem atic s (SIAM), Philadelphia, Pen n sy lvania, 1996.
A. J . Hornof a nd T . Halv erson, “ Cleaning u p sys t ematic e r r o r in eye‐tracking data
by using required fixation locations,” Behavior Research Methods, Instruments,
and Computers , vol. 34, no. 4, pp. 592–604, 2002.
P. Horowitz and W. Hill, The Art of Electronics, 2nd Edition , Cambridg e Univ e rs ity
Press, Cam b ridge, United Kingdo m, 1989.
C.‐S. Hsu and M.‐D. Liu, " Method o f Monitor i ng Ion C ontaminatio n in Integrated
Circuits ", U. S . Patent 6, 025,73 4 A, issued February 15, 2000.
D. H . Hubel, Eye, Brain, and Vision , in S cient i fic Am er ican L ibrary S e r ies, vol. 22,
W. H. Freeman and Company, New York, New York, 1988.
H . L . H u dson, R . D . S tu l ting, J . S . H e ier, S . S . L an e, D . F . Ch ang, L. J. Singerman, C. A.
Bradford, and R. E . L e onard, “ Im plantable telescope fo r end‐sta ge age‐rel at ed
macular degeneration: Long‐term visual a cuity and safety o utcom es,” American
Journal of Ophthalmology , vol. 146, no. 5, pp. 664–673, 2008.
M. S . Humayun, “ Intraocular retinal prosthesis,” Transactions of the American
Ophthalmological Society , vol. 99, pp. 271–300, 2001.
M. S. Huma yu n, E. de Juan, J. D. Weiland, G. Dagnelie, S. Kato n a, R . Greenberg, and
S. S uzuki, “ Patter n electrical s timulation o f the human retina, ” Vision Research ,
vol. 39, no. 15, pp. 2569–2576, 1999.

263

M. S. Humayun, J. D. Dorn, L. da Cruz, G. Dagnelie, J.‐A. Sahel , P. E . Stanga, A. V .
Cideciyan, J. L. Duncan, D. Eliott, E. Filley, A. C. Ho, A. San to s, A. B . Safran, A. A rditi,
L. V. Del Priore, and R. J . Greenberg, “Interim r esults f ro m th e inte r n atio nal tr ial
of S econd Sight's visual p rosthesis,” Ophthalmology , vol. 1 19, n o. 4, pp. 779–788,
2012.
M. S . Humayun, E . de J uan, J r., G. D agnelie, R . J. G reenberg, R. H. Propst, and D. H.
Phillips, “Visual Perception E li cited by E lectrical Stimulat ion of Retina in Blind
Humans,” Archives of Ophthalmology , vol. 114, no. 1, p. 40, 1996.
M. S. Humayun, J. D. Weiland, G. Y. Fujii, R. J. Greenberg, R . Williamson, J. Little, B.
Mech, V. C immarusti, G . van Boemel, G. D ag nelie, a nd E . de J uan , Jr., “Visual
perception in a blind subject with a chronic microelectronic re tin a l prosthesis,”
Vision Research , vol. 43, no. 24, p p. 2573–2581, 2003.
R. D. Jager, W. F. Mieler, and J. W. Miller, “Age‐Related Macul ar D egenerat ion, ”
New England Journal of Medicine , vol. 358, no. 24, pp. 2606–2617, 2008.
H. J osh, C . Mann, L . Kl eeman, and W. L . D. L ui, “Psychophysics Tes t ing of B ion i c
Vision Image Processing Algorithms Using an FPGA Hatpack,” in IEEE
International Conference on Image Processing 2013 , pp. 1550–1554, 2013.
E. Kasten, U. Bunzenthal, and B. A. Sabel, “Visual field recove ry a ft er v ision
restor atio n therapy (VRT) is i n d epend e nt o f ey e mov e ments : a n e ye tracker
study,” Behavioural Brain Research , vol. 175, no. 1, pp. 18–26, 2006.
D. Kömpf and H.‐F. Piper, “Eye movements and vestibulo‐ocular r eflex in t he
blind,” Journal of Neurology , vol. 234, no. 5, p p . 337–341, 1987.
R. R . Lakhanpal, D . Yanai, J . D. W eil a nd, G. Y . F u jii, S . Caffey, R. J. Greenberg, E. de
Juan, Jr., and M. S. Humayun, “Advances in the development of v isual prostheses, ”
Current Opinion in Ophthalmology , vol. 14, no. 3, pp. 122–127, 2003.
T. H. Lee and A. Hajimiri, “Oscillator phase noise: A tutorial, ” IEEE Journal of Solid ‐
State Circuits , vol. 35, no. 3, pp. 326–336, 2000.
R. J . Leigh, S . E. T hurston, R . L. T omsak, G. E . Gr ossman, and D. J . Lanska, “Effect o f
monocular visual l oss upon s tability o f gaze,” Investigative Ophthalmology and
Visual Science , vol. 30, no. 2, pp. 288–292, 1989.
R. J. Leigh and D. S. Zee, “Eye movements of the blind,” Investigative
Ophthalmology and Visual Science , vol. 19, no. 3, pp. 328–331, 1980.

264

H. L. Liou and N. A. Brennan, “Anatomically accurate, finite mo del eye for optica l
modeli ng,” Journal of the Optical Society of America A: Optics, Image Science, and
Vision , vol. 14, no. 8, p p . 1684–1695, 1997.
J. J.‐W. Liou, C.‐J. Huang, H.‐H. Chen, and G. Hong, “Character ization of p rocess‐
induced mobile ions on the data retention in flash memory,” IEEE Transactions
on Electron Devices , vol. 50, no. 4, p p . 995–1000, 2003.
R. E . MacLaren, R. A . Pearson, A. M a cNeil, R . H. D ouglas, T. E . Salt, M. Akimoto, A.
Swaroop, J. C . Sowden, and R. R . Ali, “ Retinal repair b y transp lantation o f
photoreceptor precurs o rs,” Nature , vol. 444, no. 7116, pp. 203–207, 2006.
M. M ahadevappa, J. D . Weiland, D . Yanai, I. Fine, R. J . Greenbe rg, and M. S .
Humayun, “ Perceptual t hresholds and electr ode imped a nce in t h r e e ret i n a l
prosthesis s ubjects,” IEEE Transactions on Neural Systems and Rehabilitation
Engineering , vol. 13, no. 2, pp. 201–206, 2005.
E. Margalit, M. Maia, J. D. Weiland, R. J. Greenberg, G. Y. Fujii, G. Torres, D. V.
Piyathaisere, T. M . O'Hearn, W . Liu, G . Lazzi, G. D agnelie, D . A. Scribner, E. de Juan,
Jr., and M. S. Humayun, “Retinal Prosthesis for the Blind,” Survey of
Ophthalmology , vol. 47, no. 4, pp. 335–356, 2002.
P. B. Matteucci, P. Byrnes‐Preston, S. C. Chen, N. H. Lovell, a nd G . J. S uaning, “ARM‐
based visual p rocessing s ys tem for prosthe t ic v ision,” in 2011 33rd Annual
International Conference of the IEEE Engineering in Medicine and Biology Society,
pp. 3921–3924, 2011.
H. M azyar, R . van d e n Berg, R. L . Seilheimer, and W. J . Ma, “In dependence is
elusive: S e t s ize e f fects on e ncodin g precisio n in visual se arc h,” Journal of Vision,
vol. 13, no. 5, p. 8, 2013.
B. P . McIntosh, P. J . N a siatka, N. R . B. S tiles, J . D. W eiland, M. S. Humayun, and
A. R. Tanguay, Jr., “The Importance of Natural Foveation in Ret inal Pro stheses:
Experiments with a Vis ual Prosthesis Simulator,” Proceedings of the 2010 Neural
Interfaces Conference , Long Beach, Califo rnia, 21‐23 June, 2010.
B. P . McIntosh, N. R . B. S tiles, M . S. H umayun, and A. R . Tangu ay, J r ., “Effects o f
Foveatio n on V isual Search T ask with V isual Prosthesis S im ulati on,” Journal of
Vision , vol. 13, no. 9, p. 685, 2013.
J. C . Mitros, “Single Ga te E PROM C ell for the End‐of‐lin e Ion i c C o n tamina tion
Test,” in 1993 International Integrated Reliability Workshop Final Report , pp. 40 –
44, 1993.

265

J. C. Mitros, "Method for Detecting Mobile Ions in a Semiconduc tor Device", U . S.
Patent 5, 394,101 A, i ss ued February 28, 1995.
P. M ora, S . Renard, G . B ossu, P. W altz, G. P ananakakis, and G. Ghibaudo,
“Reliability i ssues r ela t ed t o Fas t C harg e Loss Mechanism in E m bedded Non
Volatile M emories,” i n 2006 IEEE International Integrated Reliability Workshop
Final Report , pp. 68–72, 2006.
K. Najafi, “Micropackaging Techn ologies for Integrated M icrosystems:
Applications to MEMS and MOEMS,” in SPIE Proceedings on Micromachining and
Microfabrication Process Technology VIII, vol. 1, pp. 1–19, 2003.
P. J . Nasiatka, A. A huja, N. R . B. S tiles, M. C . Hauer, R . N. A grawal, R. F reda, D. G üven,
M. S . Huma yun, J . D. W eiland, and A. R . Tanguay, J r., “Intraocu lar Camera f o r
Retin a l Pro s theses”, Investigative Ophthalmology and Visual Science, vol. 46, no. 5,
ARVO Abstract 5277, Paper B480, 2005.
P. J . Nasiatka, A. A huja, N. R . B. S tiles, M. C . Hauer, R . N. A grawal, R. F reda, D. G üven,
M. S . Huma yun, J . D. W eiland, and A. R . Tanguay, J r., “Intraocu lar Camera D esig n
for Ret i n a l Prosthes es” , Annual Meeting of the Optical Society of America, Tuscon,
AZ, 2005; FiO/LS Conference Program, Paper F T hI4, 124, 2005.
P. J . Nasiatka, M. C . Hauer, N . R. B . Stiles, L. L ue, S. T akaha shi, R . N. A grawal, R.
Freda, M. S. Humayun, J. D. Weiland, and A. R. Tanguay, Jr., “I ntraocular Camera
for Retinal Prosthes es” , Investigative Ophthalmology and Visual Science , vol. 47 ,
no. 5, ARVO Abstract 3180, Paper B 554, 2006.
P. J . Nasiatka, M. C . H a uer, N . R. B . Stiles, L. L ue, S. T akaha shi, R . N. A grawal,
J. D . Weiland, M . S. H umayun, and A. R . Tanguay, J r., “An Intra ocular Camera f or
Retin a l Pro s theses”, Proceedings of BioMed , Irvine, California, 7‐8 June, 2006.
P. J . Nasiatka, B. P . McIntosh, N. R . B. S t i les, M. C . Hauer, J . D. W eiland ,
M. S. Humayun, and A. R. Tanguay, Jr., “An Intraocular Camera f or P rovision o f
Natural Foveation in R etinal P rostheses”, Proceedings of the 2010 Neural
Interfaces Conference , Long Beach, Califo rnia, 21‐23 June, 2010.
P. J . Nasiatka, M. C . Hauer, N . R. B . Stiles, A. R . Tanguay, J r., and M. S. Humayun,
"Intraocular Camera f or R etinal P rosthes e s", U.S. P atent 8,197,539 B2, issued
June 12, 2012.
R. H. Nixon, S. E. Keme n y, B. Pa in, C. O. Staller, and E. R. Fo ssum, “256×256 CM OS
activ e p ixel s ensor cam e ra‐on‐a‐chip,” IEEE Journal of Solid ‐State Circuits , vol. 31,
no. 12, pp. 2046–2050, 1996.

266

V. Pradeep, G. Medioni, and J. D. Weiland, “Robot vision for th e visually i mpaired,”
in 2010 IEEE Computer Society Conference on Computer Vision and Pattern
Recognition Workshops , pp. 15–22, 2010.
M. P tito, S. M . Moesg aard, A . Gjedde, and R. K upers, “Cross‐mod al p lasticit y
revealed b y electrot act i le s timulat i on o f th e t o ngue i n the con genitally b lind,”
Brain , vol. 128, no. 3, p p . 606–614, 2005.
G. M . Redding, R. B . Mefferd, Jr., and B . A. W iel a nd, “Effect o f observer m ovement
on m onocular depth perception,” Perceptual and Motor Skills, vol. 24, no. 3, pp.
725–726, 1967.
L . R enier, O. Co l l i gno n , C. Po i ri er, D . Trand u y, A. Vanl i erd e , A. B ol, C . Ve r aart , and
A. G. De Volder, “Cross‐modal activ ation of vis ual cortex during dep t h perceptio n
using audit o ry s ubstitution of vis ion,” Neuroimage , vol. 26, n o. 2, pp. 573–580,
2005.
J. J . van Rheede, J. J oram, C. K ennard, and S. L . Hicks, “Simul ating pros thetic v isio n:
Optimizing t he i nformation content o f a li mi ted vi sual di s play, ” Journal of Vision ,
vol. 10, no. 14, p. 32, 2010.
D. C . Rodger, A. J . Fong, W. L i, H . Ameri, A . K . A huja, C. G uti errez, I. L avrov, H .
Zhong, P . Menon, and E . Meng, “Flexible parylene‐based m ultielectrode array
technology for high‐density neural stimulation and recording,” Sensors and
Actuators B: Chemical , vol. 132, no. 2, pp. 449–460, 2008.
B. Rogers and M. Graham, “Motion parallax as an independent cue for depth
perception,” Perception , vol. 8, no. 2, pp. 125–134, 1979.
B. Rogers and M. Graham, “Similarities between motion parallax an d stereops is
in human depth percep t ion,” Vision Research , vol. 22, no. 2, pp. 261–270, 1982.
W. J . Ryan, A. T . Duchowski, and S. T . Birchfield, “Limbus/pupi l switching fo r
wear able e ye t rack ing under var i a b le l ighting conditions, ” in Proceedings of the
2008 Symposium on Eye Tracking Research and Applications , pp. 61–64, 2008.
N. Sabbah, C. N. Authié, N. Sanda, S. Mohand‐Said, J.‐A. Sahel, A. B. Safran,
“Importance of E ye P osition on Sp atial L ocalization i n BlindSub je ct s We aring an
Argus II R e tinal Prosthesis”, Investigative Ophthalmology & Visual Science , vol. 55,
no. 12, pp. 8259–8266, 2014.

267

E. S akagami, N . Arai, H. T sunoda, H. E gawa, Y. Y amaguchi, E. K a mi ya, M.
Takebuchi, K . Yamada, K. Y oshikawa, and S. M ori, “ The impact o f intermetal
dielectric layer and high temperature bake test on the reliabil ity of n onvolatile
memory d evi c es,” i n Proceedings of 1994 IEEE International Reliability Physics
Symposium , pp. 359–367, 1994.
E. M. Schmidt, M. J. Bak, F. T. Hambrecht, C. V. Kufta, D. K. O 'Rourke, and P.
Vallabhanat h, “ Feasibility of a visual prosthesis f or t he b lind based on
intracortical micro stimulat ion of t he v isual cortex,” Brain , vol. 119, n o. 2,
pp. 507–522, 1996.
P. Servos, M. A. Goodale, and L. S. Jakobson, “The role of bino cular vision i n
prehensio n : a kine matic analysis,” Vision Research , vol. 32, no. 8, pp. 1513–1521,
1992.
K. R. Sherman and E. L. Keller, “Vestibulo‐ocular reflexes of a dventitiously and
congenitall y blind adults,” Investigative Ophthalmology and Visual Science , vol. 27,
no. 7, pp. 1154–1159, 1986.
K. Shintani, D. L. Shechtman, and A. S. Gurwood, “Review and up date : current
trea tment t r ends f or p atients wit h r etinitis p igmentos a,” Optometry ‐ Journal of
the American Optometric Association , vol. 80, no. 7, pp. 384–401, 2009.
F. Shone, H. Liou, C. Pan, B. Woo, and M. Holler, “A novel meth od t o characterize
and screen m obile ion contaminate d nonvo l atile mem o ry p roducts,” in 1991
International Symposium on VLSI Technology, Systems, and Applications , pp. 224–
226, 1991.
J. S ommerhalder, E . Oueghlani, M . Bagnoud, U . Leonard s, A. B . S afran, a nd M .
Pelizzon e , “ S im ulation of a rt ific ial vision : I. E ccentr i c rea d ing of isolated words,
and perc eptual learning , ” Vision Research , vol. 43, no. 3, pp. 269–283, 2003.
J. Sommerhalder, B. Rappaz, R. de Haller, A. P. Fornos, A. B. Safr a n, a nd M .
Pelizzone, “Simulation of a rtif icial vision: II. E ccentric read ing of f ull‐page t ext
and the learning o f this t ask,” Vision Research , vol. 44, n o. 14, pp. 1693–1706,
2004.
G. P. Soong, J. E. Lovie‐Kitchin, and B. Brown, “Preferred walk in g speed f o r
assessment of m obility performance: sighted guide versus non‐si ghted guid e
techniques, ” Clinical and Experimental Optometry , vol. 83, n o. 5 , p p . 279–282,
2000.
T. S tieglitz, “Manufacturing, ass e mbling a nd p ackaging o f minia tur i zed n e ural
implants,” Microsystem Technologies , vol. 16, no. 5, pp. 723–734, 2010.

268

N. R . B. S t i les, B. P . McIntosh, P. J . Nasiatka, M. C . Hauer, J . D. W eiland ,
M. S . Humayun, a nd A . R. T anguay, Jr., “An Intraocular Camera f or R etinal
Prosthes es: Restoring Sight to t he B lind”, i n Optical Processes in Microparticles
and Nanostructures: A Festschrift Dedicated to Richard Kounai Chang on His
Retirement from Yale University , World Scientific P ublishing Com p an y, S ingapor e ,
ch. 20, pp. 385–430, 2010.
N. R . B. S til e s, B. P . McIntosh, P. J . Nasiatka, J. D . Weiland , M. S. Humayun, and
A. R. Tanguay, Jr., “Intraocular Camera for Retinal Prostheses: P sychophysical
Analys is o f Image Sampling a nd F ilter i ng”, Proceedings of the 2010 Neural
Interfaces Conference , Long Beach, Califo rnia, 21‐23 June, 2010.
K. Sundares an, K. C. Bro use, K. U‐Yen, F. Ayazi, and P. E. Alle n, “ A 7‐MHz process,
tempera t ure and supply compensated c lock o scillator in 0.25 μm CMOS,” i n
Proceedings of the 2003 International Symposium on Circuits and Systems , vol. 1,
p. I‐693, 2003.
R. S zeliski, Computer Vision: Algorithms and Applications , Spring er‐Ver lag Londo n
Limited, London, United Kingdom, 2011.
W. C. Tang, T.‐C. H. Nguyen, and R. T. Howe, “Laterally Driven Polysilicon
Resonant M icrostructures,” i n Proceedings of 1989 IEEE Micro Electromechanical
Systems Workshop , pp. 53–59, 1989.
A. R . Tanguay, J r., N. R . B. S tiles, J. C risp, and B. P . McInto sh, “Persistenc e o f
Monocular Depth Perception i n the Low Resolution L imit,” Journal of Vision , vol.
12, no. 9, p. 1195, 2012.
A. R . Tanguay, J r., N. R . B. S tiles, B. P . McIntosh, and M. S . Humayun, “ Functional
Use of M onocular Depth Percep tion i n the Low Resolution L imit,” Journal of
Vision , vol. 13, no. 9, p. 1182, 2013.
R. W . Thompson, G. D . Barnett, M . S. H umayun, and G. D agnelie, “Facial
Recognition Using Simulated Pro sthetic Pixelized Vision,” Investigative
Ophthalmology and Visual Science , vol. 44, no. 11, pp. 5035–5042, 2003.
C. V eraart, M.‐C. Wanet‐Defalque, B. G érard, A . Vanlierde, a nd J. D elbe ke, “Pattern
Recognition w ith the Optic Nerve Visual P rosthesis,” Artificial Organs , vol. 27, n o.
11, pp. 996–1004, 2003.
J. D. Weiland, A. K. Cho, and M. S. Humayun, “Retinal prostheses: current clinical
results and future need s,” Ophthalmology , vol. 118, no. 11, pp. 2227–2237, 2011.

269

J. D. Weiland, F. M. Kimock, J. E. Yehoda, E. Gill, B. P. McIntosh, P. J. Nasiatka, and
A. R . Tangu a y, J r., “Ch i p‐Scale Packaging for Bioelectronics Im plants,” i n 2013
International IEEE/EMBS Conference on Neural Engineering , pp. 931–936, 2013.
J. D . Weiland, W . Liu, a nd M . S. H umayun, “Retinal p rosth e sis,” Annual Review of
Biomedical Engineering , vol. 7, pp. 361–401, 2005.
K. P . White, J r., T. E . Hutchinson, and J. M . Carley, “Spatiall y dynamic calibration of
an e ye‐trac k ing system,” IEEE Transactions on Systems, Man, and Cybernetics ,
vol. 23, no. 4, pp. 1162–1168, 1993.
S . A. W i nges, D . J . W e ber, and M . S ant el l o , “ T he ro l e o f visio n on hand preshaping
during reach to grasp,” Experimental Brain Research , vol. 1 52, n o. 4 , pp. 489–498,
2003.
J. M. Wolfe, “What Can 1 Million T rials Tell Us A bout Visual Se arch?” Psychological
Science , vol. 9, no. 1, pp. 33–39, 1998.
J. M. Wolfe and S. L. Franzel, “Binocularity and visual search, ” Perception and
Psychophysics , vol. 44, no. 1, pp. 81–93, 1988.
J. M. Wolfe, “Guided search 2.0 a revised model of visual searc h,” Psychonomic
Bulletin and Review , vol. 1, no. 2, pp. 202–238, 1994.
D. Y anai, J. D . Weiland, M . Mahadevappa, R . J . Greenb e rg, I. F i ne, and M. S .
Humayun, “Visual performance using a retinal prosthesis in thre e su bjects w ith
retinitis pigmentosa,” American Journal of Ophthalmology , vol. 143, n o. 5,
pp. 820–827, 2007.
X. Zhao, R. Chebli, and M. Sawan, “A wide tuning range voltage‐ controlled ring
oscillator dedica ted to u ltrasound transmitter,” in The 16th International
Conference on Microelectronics , pp. 313–316, 2004.
D. Zhou and E. Greenbaum, Implantable Neural Prostheses 2: Techniques and
Engineering Approaches , Springer S cience+Bu s iness Med i a LLC, New York, New
York, 2010.
E. Z renner, K . U. B artz‐ S chmidt, H. B enav, D. B esch, A. B ruckmann, V.‐P. Gabel, F.
Gekeler, U . Greppmaier, A. H arscher, S . Kibbel, J . Koch, A. K us nyer ik, T. P et ers, K.
Stingl, H. S achs, A. S tett, P . Szurman, B . Wilhelm, a nd R . Wilk e, " Subretina l
electronic chips allow blind patients to read letters and combine them to words,"
Proceedings of the Royal Society B: Biological Sciences , vol. 278, n o. 1711,
pp. 1489‐1497, 2011.

270

APPENDX A
HERMETIC COATING TEST CHIP CADENCE LAYOUTS AND
MOSIS SPICE DEVICE SIMULATION PARAMETERS
Hermetic Coating Test Chip 1 (HCTC ‐1) – MOSIS Run T95S

271

MOSIS WAFER ACCEPTANCE TESTS

RUN: T95S VENDOR: AMIS (ON‐SEMI)
  TECHNOLOGY: SCN05 FEATURE SIZE: 0.5 microns
  Run type: SKD

INTRODUCTION: This report contains the lot average results obtained by MOSIS
from measurements of MOSIS test structures on each wafer of
this fabrication lot. SPICE parameters obtained from similar
measurements on a selected wafer are also attached.

COMMENTS: SMSCN3ME06_ON‐SEMI

TRANSISTOR PARAMETERS   W/L   N‐CHANNEL P‐CHANNEL UNITS

MINIMUM 3.0/0.6
Vth   0.77 ‐0.91 volts

SHORT 20.0/0.6
Idss 465   ‐250   uA/um
Vth   0.66 ‐0.89 volts
Vpt   12.9 ‐11.7 volts

WIDE   20.0/0.6
Ids0 < 2.5     < 2.5      pA/um

LARGE 20.0/20.0
Vth   0.68 ‐0.94 volts
Vjbkd   10.8 ‐11.9 volts
Ijlk < 50.0    < 50.0     pA
Gamma   0.48 0.56   V^0.5

K' (Uo*Cox/2) 55.9 ‐19.0 uA/V^2
Low‐field Mobility   459.75 156.27   cm^2/V*s

PROCESS PARAMETERS N+   P+   POLY PLY2_HR   M1   UNITS
Sheet Resistance   87.0   110.4 23.1 1080 0.09   ohms/sq
Contact Resistance   60.7   146.2 18.5   ohms
Gate Oxide Thickness   142 angstroms

CAPACITANCE PARAMETERS N+   P+   POLY POLY2   M1   UNITS
Area (substrate)   437 733 94   aF/um^2
Area (N+active) 2438   aF/um^2
Area (P+active) 2357   aF/um^2
Area (poly) 876   69   aF/um^2

CIRCUIT PARAMETERS UNITS
Ring Oscillator Freq.
DIV256 (31‐stg,5.0V) 103.00 MHz
Ring Oscillator Power
DIV256 (31‐stg,5.0V) 0.48 uW/MHz/gate

COMMENTS: SUBMICRON

SPICE BSIM parameters not available.

272

Hermetic Coating Test Chip 2 (HCTC ‐2) – MOSIS Run T9BM

273

CMOS Clock Test Chip 2 (CCTC ‐2) – MOSIS Run T9BM

274

  MOSIS WAFER ACCEPTANCE TESTS

RUN: T9BM VENDOR: AMIS (ON‐SEMI)
  TECHNOLOGY: SCN05 FEATURE SIZE: 0.5 microns
  Run type: SKD

INTRODUCTION: This report contains the lot average results obtained by MOSIS
from measurements of MOSIS test structures on each wafer of
this fabrication lot. SPICE parameters obtained from similar
measurements on a selected wafer are also attached.

COMMENTS: SMSCN3ME06_ON‐

TRANSISTOR PARAMETERS   W/L N‐CHANNEL P‐CHANNEL UNITS

MINIMUM 3.0/0.6
Vth 0.77 ‐0.90 volts

SHORT 20.0/0.6
Idss   465   ‐255   uA/um
Vth 0.65 ‐0.88 volts
Vpt 10.0 ‐10.0 volts

WIDE   20.0/0.6
Ids0   < 2.5     < 2.5      pA/um

LARGE 50/50
Vth 0.67 ‐0.93 volts
Vjbkd 10.4 ‐11.6 volts
Ijlk   < 50.0    < 50.0     pA
Gamma 0.48 0.56   V^0.5

K' (Uo*Cox/2)   57.8 ‐19.4 uA/V^2
Low‐field Mobility 465.34 156.19   cm^2/V*s

COMMENTS: Poly bias varies with design technology. To account for mask
  bias use the appropriate value for the parameter XL in your
  SPICE model card.
  Design Technology   XL (um) XW (um)
  ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐   ‐‐‐‐‐‐‐ ‐‐‐‐‐‐
  SCMOS_SUBM (lambda=0.30) 0.10   0.00
  SCMOS (lambda=0.35)   0.00   0.20

FOX TRANSISTORS   GATE N+ACTIVE P+ACTIVE UNITS
Vth Poly > 15.0 < ‐15.0   volts

PROCESS PARAMETERS   N+   P+   POLY   PLY2_HR POLY2   M1 M2 UNITS
Sheet Resistance 84.0   107.2 23.5   1033   41.7 0.09 0.09 ohms/sq
Contact Resistance 65.0   141.3 16.4 26.5 0.87 ohms
Gate Oxide Thickness 139   angstrom

PROCESS PARAMETERS M3 N\PLY   N_W   UNITS
Sheet Resistance   0.05 823   825   ohms/sq
Contact Resistance   0.86 ohms

COMMENTS: N\POLY is N‐well under polysilicon.

275

CAPACITANCE PARAMETERS   N+   P+   POLY   POLY2   M1   M2   M3   N_W UNITS
Area (substrate) 427 733 87   31   16 9   39   aF/um^2
Area (N+active)   2488   36   16   11 aF/um^2
Area (P+active)   2394   aF/um^2
Area (poly) 897   59   16 9 aF/um^2
Area (poly2)   54 aF/um^2
Area (metal1)   32   13 aF/um^2
Area (metal2) 32 aF/um^2
Fringe (substrate) 385 237   71   50   31 aF/um
Fringe (poly) 64   39   28 aF/um
Fringe (metal1)   51   34 aF/um
Fringe (metal2) 55 aF/um

CIRCUIT PARAMETERS UNITS
Inverters   K
Vinv 1.0   2.04 volts
Vinv 1.5   2.30 volts
Vol (100 uA) 2.0   0.12 volts
Voh (100 uA) 2.0   4.87 volts
Vinv 2.0   2.49 volts
Gain 2.0   ‐17.61
Ring Oscillator Freq.
DIV256 (31‐stg,5.0V)   98.07 MHz
D256_WIDE (31‐stg,5.0V)   154.97 MHz
Ring Oscillator Power
DIV256 (31‐stg,5.0V) 0.49 uW/MHz/gate
D256_WIDE (31‐stg,5.0V)   1.02 uW/MHz/gate

COMMENTS: SUBMICRON

T9BM SPICE BSIM3 VERSION 3.1 PARAMETERS

SPICE 3f5 Level 8, Star‐HSPICE Level 49, UTMOST Level 8

* DATE: Mar 15/10
* LOT: T9BM WAF: 2102
* Temperature_parameters=Default
.MODEL CMOSN NMOS ( LEVEL   = 49
+VERSION = 3.1 TNOM = 27   TOX   = 1.39E‐8
+XJ = 1.5E‐7   NCH   = 1.7E17   VTH0 = 0.5916823
+K1 = 0.9970315 K2 = ‐0.1275822   K3 = 32.4012543
+K3B   = ‐7.5008412   W0 = 1.064896E‐8 NLX   = 1E‐9
+DVT0W   = 0 DVT1W   = 0 DVT2W   = 0
+DVT0 = 0.7187527 DVT1 = 0.3412308 DVT2 = ‐0.5
+U0 = 449.0974306 UA = 1.085166E‐13   UB = 1.425387E‐18
+UC = ‐1.74612E‐14   VSAT = 1.669807E5   A0 = 0.5933847
+AGS   = 0.1035164 B0 = 1.668729E‐6 B1 = 5E‐6
+KETA = ‐3.86833E‐3 A1 = 2.152036E‐4 A2 = 0.3536877
+RDSW = 1.061307E3   PRWG = 0.0822544 PRWB = 1.740605E‐5
+WR = 1 WINT = 2.482453E‐7 LINT = 7.591645E‐8
+XL = 1E‐7   XW = 0 DWG   = ‐1.279102E‐8
+DWB   = 4.853492E‐8 VOFF = 0 NFACTOR = 0.9088018
+CIT   = 0 CDSC = 2.4E‐4   CDSCD   = 0
+CDSCB   = 0 ETA0 = 2.907125E‐3 ETAB = 0.9951793
+DSUB = 0.0492559 PCLM = 2.2082829 PDIBLC1 = 3.324146E‐4
+PDIBLC2 = 1.488616E‐3 PDIBLCB = ‐0.1429457   DROUT   = 8.319308E‐5
+PSCBE1 = 2.516822E9   PSCBE2 = 5E‐10 PVAG = 0

276

+DELTA   = 0.01   RSH   = 84   MOBMOD = 1
+PRT   = 0 UTE   = ‐1.5   KT1   = ‐0.11
+KT1L = 0 KT2   = 0.022 UA1   = 4.31E‐9
+UB1   = ‐7.61E‐18 UC1   = ‐5.6E‐11   AT = 3.3E4
+WL = 0 WLN   = 1 WW = 0
+WWN   = 1 WWL   = 0 LL = 0
+LLN   = 1 LW = 0 LWN   = 1
+LWL   = 0 CAPMOD = 2 XPART   = 0.5
+CGDO = 1.82E‐10   CGSO = 1.82E‐10   CGBO = 1E‐9
+CJ = 4.29253E‐4   PB = 0.8858652 MJ = 0.4353384
+CJSW = 3.622458E‐10   PBSW = 0.8 MJSW = 0.2339166
+CJSWG   = 1.64E‐10   PBSWG   = 0.8 MJSWG   = 0.2339166
+CF = 0 PVTH0   = ‐0.0504111   PRDSW   = 277.8034293
+PK2   = ‐0.0683031   WKETA   = 5.494298E‐3 LKETA   = ‐6.64089E‐3   )
*
.MODEL CMOSP PMOS ( LEVEL   = 49
+VERSION = 3.1 TNOM = 27   TOX   = 1.39E‐8
+XJ = 1.5E‐7   NCH   = 1.7E17   VTH0 = ‐0.9152268
+K1 = 0.553472   K2 = 7.871921E‐3 K3 = 0.3045116
+K3B   = 0.4207623 W0 = 5.814395E‐7 NLX   = 6.71009E‐8
+DVT0W   = 0 DVT1W   = 0 DVT2W   = 0
+DVT0 = 0.6597447 DVT1 = 0.3261802 DVT2 = ‐0.3
+U0 = 201.3603195 UA = 2.408572E‐9 UB = 1E‐21
+UC = ‐1E‐10   VSAT = 9.345943E4   A0 = 0.8172376
+AGS   = 0.097476   B0 = 5.101025E‐7 B1 = 1.701779E‐8
+KETA = ‐4.865785E‐3   A1 = 0 A2 = 0.6429159
+RDSW = 3E3 PRWG = ‐0.0280817   PRWB = ‐0.0480265
+WR = 1 WINT = 2.908993E‐7 LINT = 1.225719E‐7
+XL = 1E‐7   XW = 0 DWG   = 1.554343E‐9
+DWB   = ‐1.9281E‐8   VOFF = ‐0.0705167   NFACTOR = 1.016159
+CIT   = 0 CDSC = 2.4E‐4   CDSCD   = 0
+CDSCB   = 0 ETA0 = 6.97667E‐4   ETAB = ‐0.2
+DSUB = 1 PCLM = 2.4034299 PDIBLC1 = 0.0525071
+PDIBLC2 = 3.609644E‐3 PDIBLCB = ‐0.0206211   DROUT   = 0.2542493
+PSCBE1 = 1E8 PSCBE2 = 3.365142E‐9 PVAG = 0.0150027
+DELTA   = 0.01   RSH   = 107.2 MOBMOD = 1
+PRT   = 0 UTE   = ‐1.5   KT1   = ‐0.11
+KT1L = 0 KT2   = 0.022 UA1   = 4.31E‐9
+UB1   = ‐7.61E‐18 UC1   = ‐5.6E‐11   AT = 3.3E4
+WL = 0 WLN   = 1 WW = 0
+WWN   = 1 WWL   = 0 LL = 0
+LLN   = 1 LW = 0 LWN   = 1
+LWL   = 0 CAPMOD = 2 XPART   = 0.5
+CGDO = 2.22E‐10   CGSO = 2.22E‐10   CGBO = 1E‐9
+CJ = 7.306902E‐4 PB = 0.9419671 MJ = 0.4929212
+CJSW = 2.535867E‐10   PBSW = 0.8 MJSW = 0.3168453
+CJSWG   = 6.4E‐11 PBSWG   = 0.8 MJSWG   = 0.3168453
+CF = 0 PVTH0   = 5.98016E‐3   PRDSW   = 14.8598424
+PK2   = 3.73981E‐3   WKETA   = 1.137045E‐4 LKETA   = ‐9.773144E‐3 )
*

277

Hermetic Coating Test Chip 3 (HCTC ‐3) – MOSIS Run V15L

278

  MOSIS WAFER ELECTRICAL TESTS

RUN: V15L VENDOR: AMIS (ON‐SEMI)
  TECHNOLOGY: SCN05 FEATURE SIZE: 0.5 microns
  Run type: SHR

INTRODUCTION: This report contains the lot average results obtained by MOSIS
from measurements of MOSIS test structures on each wafer of
this fabrication lot. SPICE parameters obtained from similar
measurements on a selected wafer are also attached.

COMMENTS: SMSCN3ME06_ON‐SEMI

TRANSISTOR PARAMETERS   W/L   N‐CHANNEL P‐CHANNEL UNITS

MINIMUM 3.0/0.6
Vth   0.74   ‐0.88 volts

SHORT 20.0/0.6
Idss 472 ‐267   uA/um
Vth   0.64   ‐0.87 volts
Vpt   12.6   ‐12.4 volts

WIDE   20.0/0.6
Ids0 < 2.5      < 2.5      pA/um

LARGE 50/50
Vth   0.67   ‐0.92 volts
Vjbkd   11.0   ‐11.8 volts
Ijlk 222.8 < 50.0     pA
Gamma   0.48   0.55   V^0.5

K' (Uo*Cox/2) 59.1   ‐19.2 uA/V^2
Low‐field Mobility   475.80   154.58   cm^2/V*s

COMMENTS: Poly bias varies with design technology. To account for mask
bias use the appropriate value for the parameter XL in your
SPICE model card.
  Design Technology   XL (um) XW (um)
  ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐   ‐‐‐‐‐‐‐ ‐‐‐‐‐‐
  SCMOS_SUBM (lambda=0.30)   0.10   0.00
  SCMOS (lambda=0.35) 0.00   0.20

FOX TRANSISTORS   GATE N+ACTIVE P+ACTIVE UNITS
Vth Poly > 15.0 < ‐15.0   volts

PROCESS PARAMETERS   N+ P+ N_W_U_PLY POLY PLY2_HR POLY2 M1 UNITS
Sheet Resistance 83.3 108.1   808.5   23.1 1059   41.0   0.09 ohms/sq
Contact Resistance 59.0 141.4   16.3   26.1   ohms
Gate Oxide Thickness 139   angstrom

PROCESS PARAMETERS   M2 M3 N_W UNITS
Sheet Resistance 0.09 0.05 802 ohms/sq
Contact Resistance 0.97 0.86   ohms

279

CAPACITANCE PARAMETERS   N+   P+   POLY   POLY2   M1   M2   M3   N_W UNITS
Area (substrate) 421 713 84   28   12 8   91   aF/um^2
Area (N+active)   2486   36   16   11 aF/um^2
Area (P+active)   2401   aF/um^2
Area (poly) 909   62   15 9 aF/um^2
Area (poly2)   53 aF/um^2
Area (metal1)   29   12 aF/um^2
Area (metal2) 31 aF/um^2
Fringe (substrate)   346 259 51   33   25 aF/um
Fringe (poly) 67   37   27 aF/um
Fringe (metal1)   50   32 aF/um
Fringe (metal2) 44 aF/um
Overlap (N+active) 192 aF/um
Overlap (P+active) 227 aF/um

CIRCUIT PARAMETERS UNITS
Inverters   K
Vinv 1.0   2.00 volts
Vinv 1.5   2.25 volts
Vol   2.0   0.44 volts
Voh   2.0   4.49 volts
Vinv 2.0   2.43 volts
Gain 2.0   ‐17.88
Ring Oscillator Freq.
DIV256 (31‐stg,5.0V) 108.80 MHz
D256_WIDE (31‐stg,5.0V)   158.03 MHz
Ring Oscillator Power
DIV256 (31‐stg,5.0V) 0.51 uW/MHz/gate
D256_WIDE (31‐stg,5.0V)   1.03 uW/MHz/gate

COMMENTS: SUBMICRON

V15L SPICE BSIM3 VERSION 3.1 PARAMETERS

SPICE 3f5 Level 8, Star‐HSPICE Level 49, UTMOST Level 8

* DATE: Oct 12/11
* LOT: V15L WAF: 6103
* Temperature_parameters=Default
.MODEL CMOSN NMOS ( LEVEL   = 49
+VERSION = 3.1 TNOM = 27   TOX   = 1.39E‐8
+XJ = 1.5E‐7   NCH   = 1.7E17   VTH0 = 0.5860022
+K1 = 0.8457293 K2 = ‐0.0867997   K3 = 22.5374057
+K3B   = ‐8.5583819   W0 = 1.10556E‐8   NLX   = 1E‐9
+DVT0W   = 0 DVT1W   = 0 DVT2W   = 0
+DVT0 = 0.7878744 DVT1 = 0.4044405 DVT2 = ‐0.4998896
+U0 = 454.8330745 UA = 1E‐13 UB = 1.328915E‐18
+UC = 5.936711E‐13   VSAT = 1.990599E5   A0 = 0.6512266
+AGS   = 0.1209495 B0 = 1.666118E‐6 B1 = 5E‐6
+KETA = ‐5.477831E‐3   A1 = 1.318434E‐4 A2 = 0.3
+RDSW = 1.035028E3   PRWG = 0.1144981 PRWB = ‐6.678878E‐3
+WR = 1 WINT = 1.61606E‐7   LINT = 8.30894E‐8
+XL = 1E‐7   XW = 0 DWG   = 2.275194E‐9
+DWB   = 2.679715E‐8 VOFF = 0 NFACTOR = 0.8415838
+CIT   = 0 CDSC = 2.4E‐4   CDSCD   = 0
+CDSCB   = 0 ETA0 = 3.101025E‐3 ETAB = ‐3.251351E‐4

280

+DSUB = 0.0648038 PCLM = 2.0762866 PDIBLC1 = 4.486508E‐4
+PDIBLC2 = 1.707832E‐3 PDIBLCB = 0.1041469 DROUT   = 2.38347E‐3
+PSCBE1 = 6.26971E10   PSCBE2 = 1.263134E‐8 PVAG = 0
+DELTA   = 0.01   RSH   = 83.3   MOBMOD = 1
+PRT   = 0 UTE   = ‐1.5   KT1   = ‐0.11
+KT1L = 0 KT2   = 0.022 UA1   = 4.31E‐9
+UB1   = ‐7.61E‐18 UC1   = ‐5.6E‐11   AT = 3.3E4
+WL = 0 WLN   = 1 WW = 0
+WWN   = 1 WWL   = 0 LL = 0
+LLN   = 1 LW = 0 LWN   = 1
+LWL   = 0 CAPMOD = 2 XPART   = 0.5
+CGDO = 1.92E‐10   CGSO = 1.92E‐10   CGBO = 1E‐9
+CJ = 4.210538E‐4 PB = 0.8449659 MJ = 0.4323265
+CJSW = 3.38396E‐10 PBSW = 0.8 MJSW = 0.1985177
+CJSWG   = 1.64E‐10   PBSWG   = 0.8 MJSWG   = 0.1985177
+CF = 0 PVTH0   = 6.648606E‐3 PRDSW   = 134.0661526
+PK2   = ‐0.0750373   WKETA   = ‐6.020017E‐3   LKETA   = ‐7.935753E‐4 )
*
.MODEL CMOSP PMOS ( LEVEL   = 49
+VERSION = 3.1 TNOM = 27   TOX   = 1.39E‐8
+XJ = 1.5E‐7   NCH   = 1.7E17   VTH0 = ‐0.9152268
+K1 = 0.553472   K2 = 7.871921E‐3 K3 = 0
+K3B   = 2.4058299 W0 = 1E‐8   NLX   = 1E‐9
+DVT0W   = 0 DVT1W   = 0 DVT2W   = 0
+DVT0 = 0.5628844 DVT1 = 0.3062254 DVT2 = ‐0.3
+U0 = 201.3603195 UA = 2.408572E‐9 UB = 1E‐21
+UC = ‐1E‐10   VSAT = 7.748437E4   A0 = 0.7617795
+AGS   = 0.1245297 B0 = 7.263936E‐7 B1 = 5.265808E‐9
+KETA = ‐4.865785E‐3   A1 = 1.267344E‐3 A2 = 0.8503916
+RDSW = 3E3 PRWG = ‐0.0260866   PRWB = ‐0.0518431
+WR = 1 WINT = 1.829642E‐7 LINT = 1.271308E‐7
+XL = 1E‐7   XW = 0 DWG   = 1.098658E‐9
+DWB   = ‐5.253863E‐9   VOFF = ‐0.0608429   NFACTOR = 0.8419451
+CIT   = 0 CDSC = 2.4E‐4   CDSCD   = 0
+CDSCB   = 0 ETA0 = 3.726555E‐4 ETAB = ‐0.137281
+DSUB = 0.7967864 PCLM = 2.424885   PDIBLC1 = 0.0588933
+PDIBLC2 = 3.443668E‐3 PDIBLCB = ‐0.0295332   DROUT   = 0.2780193
+PSCBE1 = 1E8 PSCBE2 = 3.391657E‐9 PVAG = 3.4815E‐3
+DELTA   = 0.01   RSH   = 108.1 MOBMOD = 1
+PRT   = 0 UTE   = ‐1.5   KT1   = ‐0.11
+KT1L = 0 KT2   = 0.022 UA1   = 4.31E‐9
+UB1   = ‐7.61E‐18 UC1   = ‐5.6E‐11   AT = 3.3E4
+WL = 0 WLN   = 1 WW = 0
+WWN   = 1 WWL   = 0 LL = 0
+LLN   = 1 LW = 0 LWN   = 1
+LWL   = 0 CAPMOD = 2 XPART   = 0.5
+CGDO = 2.27E‐10   CGSO = 2.27E‐10   CGBO = 1E‐9
+CJ = 7.093224E‐4 PB = 0.882349   MJ = 0.4888155
+CJSW = 2.577549E‐10   PBSW = 0.8 MJSW = 0.2161724
+CJSWG   = 6.4E‐11 PBSWG   = 0.8 MJSWG   = 0.2161724
+CF = 0 PVTH0   = 5.98016E‐3   PRDSW   = 14.8598424
+PK2   = 3.73981E‐3   WKETA   = 0.0118004 LKETA   = ‐0.0167812 )
*

281

Hermetic Coating Test Chip 4 (HCTC ‐4) – MOSIS Run V17L

282

MOSIS WAFER ELECTRICAL TESTS

RUN: V17L VENDOR: AMIS (ON‐SEMI)
  TECHNOLOGY: SCN05 FEATURE SIZE: 0.5 microns
  Run type: SHR

INTRODUCTION: This report contains the lot average results obtained by MOSIS
from measurements of MOSIS test structures on each wafer of
this fabrication lot.

COMMENTS: SMSCN3ME06_ON‐SEMI

TRANSISTOR PARAMETERS   W/L   N‐CHANNEL   P‐CHANNEL UNITS

MINIMUM 3.0/0.6
Vth   0.78 ‐0.86 volts

SHORT 20.0/0.6
Idss 472   ‐256   uA/um
Vth   0.67 ‐0.84 volts
Vpt   13.1 ‐12.1 volts

WIDE   20.0/0.6
Ids0 < 2.5       < 2.5      pA/um

LARGE 20.0/20.0
Vth   0.68 ‐0.90 volts
Vjbkd   10.9 ‐11.9 volts
Ijlk < 50.0      < 50.0     pA
Gamma   0.47 0.54   V^0.5

K' (Uo*Cox/2) 58.2 ‐19.9 uA/V^2
Low‐field Mobility   465.19 159.06   cm^2/V*s

PROCESS PARAMETERS N+ P+   POLY PLY2_HR   M1   UNITS
Sheet Resistance   84.8 110.5 23.6 1031 0.09   ohms/sq
Contact Resistance   57.8 149.0 19.2   ohms
Gate Oxide Thickness   138   angstroms

CAPACITANCE PARAMETERS   N+ P+ POLY POLY2 M1 UNITS
Area (substrate) 428   713   94   aF/um^2
Area (N+active)   2494   aF/um^2
Area (P+active)   2400   aF/um^2
Area (poly)   901 60 aF/um^2

CIRCUIT PARAMETERS UNITS
Ring Oscillator Freq.
DIV256 (31‐stg,5.0V) 107.19 MHz
Ring Oscillator Power
DIV256 (31‐stg,5.0V) 0.50 uW/MHz/gate

COMMENTS: SUBMICRON

SPICE BSIM parameters not available.

283

Hermetic Coating Test Chip 5 (HCTC ‐5) – MOSIS Run V23R

284

  MOSIS WAFER ELECTRICAL TESTS

RUN: V23R VENDOR: AMIS (ON‐SEMI)
  TECHNOLOGY: SCN05 FEATURE SIZE: 0.5 microns
  Run type: SHR

INTRODUCTION: This report contains the lot average results obtained by MOSIS
from measurements of MOSIS test structures on each wafer of
this fabrication lot. SPICE parameters obtained from similar
measurements on a selected wafer are also attached.

COMMENTS: SMSCN3ME06_ON‐SEMI

TRANSISTOR PARAMETERS   W/L   N‐CHANNEL P‐CHANNEL UNITS

MINIMUM 3.0/0.6
Vth   0.78   ‐0.87 volts

SHORT 20.0/0.6
Idss 476 ‐264   uA/um
Vth   0.68   ‐0.85 volts
Vpt   11.5   ‐12.0 volts

WIDE   20.0/0.6
Ids0 < 2.5      < 2.5      pA/um

LARGE 50/50
Vth   0.69   ‐0.90 volts
Vjbkd   10.9   ‐11.9 volts
Ijlk 200.0 < 50.0     pA
Gamma   0.48   0.55   V^0.5

K' (Uo*Cox/2) 59.4   ‐19.6 uA/V^2
Low‐field Mobility   478.22   157.80   cm^2/V*s

COMMENTS: Poly bias varies with design technology. To account for mask
  bias use the appropriate value for the parameter XL in your
  SPICE model card.
  Design Technology   XL (um) XW (um)
  ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐   ‐‐‐‐‐‐‐ ‐‐‐‐‐‐
  SCMOS_SUBM (lambda=0.30)   0.10   0.00
  SCMOS (lambda=0.35) 0.00   0.20

FOX TRANSISTORS   GATE N+ACTIVE P+ACTIVE UNITS
Vth Poly > 15.0 < ‐15.0   volts

PROCESS PARAMETERS   N+ P+ N_W POLY PLY2_HR   POLY2   M1 UNITS
Sheet Resistance 83.7 107.8   805.5 22.9 1103 40.4 0.09 ohms/sq
Contact Resistance 62.7 160.4 15.6 25.7 ohms
Gate Oxide Thickness 139 angstrom

PROCESS PARAMETERS   M2 M3   N_W   UNITS
Sheet Resistance   0.09 0.05 799   ohms/sq
Contact Resistance   0.88 0.88 ohms

285

CAPACITANCE PARAMETERS   N+   P+ POLY   POLY2   M1   M2   M3 N_W   UNITS
Area (substrate) 420 713   85   26   11 7 90 aF/um^2
Area (N+active) 2482   34   16   11 aF/um^2
Area (P+active) 2402   aF/um^2
Area (poly)   918   55   15 8 aF/um^2
Area (poly2) 50 aF/um^2
Area (metal1) 29   12 aF/um^2
Area (metal2)   29 aF/um^2
Fringe (substrate) 360 247 52   32   25 aF/um
Fringe (poly)   67   36   26 aF/um
Fringe (metal1) 47   32 aF/um
Fringe (metal2)   46 aF/um
Overlap (N+active)   186 aF/um
Overlap (P+active)   232 aF/um

CIRCUIT PARAMETERS UNITS
Inverters   K
Vinv 1.0   2.04 volts
Vinv 1.5   2.29 volts
Vol   2.0   0.43 volts
Voh   2.0   4.49 volts
Vinv 2.0   2.47 volts
Gain 2.0   ‐17.68
Ring Oscillator Freq.
DIV256 (31‐stg,5.0V) 106.58 MHz
D256_WIDE (31‐stg,5.0V)   156.88 MHz
Ring Oscillator Power
DIV256 (31‐stg,5.0V) 0.50 uW/MHz/gate
D256_WIDE (31‐stg,5.0V)   1.03 uW/MHz/gate

COMMENTS: SUBMICRON

V23R SPICE BSIM3 VERSION 3.1 PARAMETERS

SPICE 3f5 Level 8, Star‐HSPICE Level 49, UTMOST Level 8

* DATE: Jul 17/12
* LOT: v23r WAF: 6101
* Temperature_parameters=Default
.MODEL CMOSN NMOS ( LEVEL   = 49
+VERSION = 3.1 TNOM = 27   TOX   = 1.39E‐8
+XJ = 1.5E‐7   NCH   = 1.7E17   VTH0 = 0.6028233
+K1 = 0.9019937 K2 = ‐0.1063649   K3 = 23.8248969
+K3B   = ‐10.5999524 W0 = 2.495E‐8   NLX   = 1.622195E‐9
+DVT0W   = 0 DVT1W   = 0 DVT2W   = 0
+DVT0 = 0.7381817 DVT1 = 0.429155   DVT2 = ‐0.4998603
+U0 = 456.8851707 UA = 1E‐13 UB = 1.337362E‐18
+UC = 1.100108E‐11   VSAT = 1.969277E5   A0 = 0.6181139
+AGS   = 0.1245191 B0 = 1.662321E‐6 B1 = 5E‐6
+KETA = ‐2.805284E‐3   A1 = 7.960277E‐5 A2 = 0.4
+RDSW = 1.108387E3   PRWG = 0.078723   PRWB = 3.1215E‐3
+WR = 1 WINT = 1.674671E‐7 LINT = 3.68945E‐8
+XL = 1E‐7   XW = 0 DWG   = 7.251138E‐10
+DWB   = 4.828285E‐8 VOFF = ‐1.143321E‐4   NFACTOR = 1.0078508
+CIT   = 0 CDSC = 2.4E‐4   CDSCD   = 0
+CDSCB   = 0 ETA0 = 1.491532E‐3 ETAB = ‐2.178885E‐4
+DSUB = 0.0692762 PCLM = 2.240135   PDIBLC1 = 1.103948E‐4

286

+PDIBLC2 = 2.05652E‐3   PDIBLCB = 0.0882174 DROUT   = 3.530731E‐3
+PSCBE1 = 2.533273E8   PSCBE2 = 2.692568E‐7 PVAG = 0
+DELTA   = 0.01   RSH   = 83.7   MOBMOD = 1
+PRT   = 0 UTE   = ‐1.5   KT1   = ‐0.11
+KT1L = 0 KT2   = 0.022 UA1   = 4.31E‐9
+UB1   = ‐7.61E‐18 UC1   = ‐5.6E‐11   AT = 3.3E4
+WL = 0 WLN   = 1 WW = 0
+WWN   = 1 WWL   = 0 LL = 0
+LLN   = 1 LW = 0 LWN   = 1
+LWL   = 0 CAPMOD = 2 XPART   = 0.5
+CGDO = 1.86E‐10   CGSO = 1.86E‐10   CGBO = 1E‐9
+CJ = 4.083504E‐4 PB = 0.8382373 MJ = 0.4270655
+CJSW = 3.342119E‐10   PBSW = 0.8 MJSW = 0.2126305
+CJSWG   = 1.64E‐10   PBSWG   = 0.8 MJSWG   = 0.2019414
+CF = 0 PVTH0   = ‐0.0117317   PRDSW   = 249.453985
+PK2   = ‐0.0895601   WKETA   = ‐2.152613E‐3   LKETA   = ‐2.772864E‐3 )
*
*
.MODEL CMOSP PMOS ( LEVEL   = 49
+VERSION = 3.1 TNOM = 27   TOX   = 1.39E‐8
+XJ = 1.5E‐7   NCH   = 1.7E17   VTH0 = ‐0.9152268
+K1 = 0.553472   K2 = 7.871921E‐3 K3 = 0
+K3B   = 1.0343517 W0 = 1E‐8   NLX   = 1E‐9
+DVT0W   = 0 DVT1W   = 0 DVT2W   = 0
+DVT0 = 0.4711614 DVT1 = 0.184542   DVT2 = ‐0.3
+U0 = 201.3603195 UA = 2.48572E‐9   UB = 1.005454E‐21
+UC = ‐1E‐10   VSAT = 9.928583E4   A0 = 0.6060342
+AGS   = 0.1349577 B0 = 9.090026E‐7 B1 = 1E‐9
+KETA = ‐4.865785E‐3   A1 = 1.245242E‐3 A2 = 0.9204559
+RDSW = 2.94E3   PRWG = ‐0.0218956   PRWB = ‐0.0585256
+WR = 1.01   WINT = 1.223084E‐7 LINT = 9.1683E‐8
+XL = 1E‐7   XW = 0 DWG   = 1.341677E‐8
+DWB   = ‐2.729018E‐8   VOFF = ‐0.0685488   NFACTOR = 0.6150316
+CIT   = 0 CDSC = 2.4E‐4   CDSCD   = 0
+CDSCB   = 0 ETA0 = 0.0155454 ETAB = ‐0.033248
+DSUB = 0.5163729 PCLM = 2.3034167 PDIBLC1 = 0.069882
+PDIBLC2 = 4.225266E‐3 PDIBLCB = ‐0.0327468   DROUT   = 0.26364
+PSCBE1 = 8.478841E9   PSCBE2 = 1.442045E‐8 PVAG = 7.040289E‐3
+DELTA   = 0.01   RSH   = 107.8 MOBMOD = 1
+PRT   = 0 UTE   = ‐1.5   KT1   = ‐0.11
+KT1L = 0 KT2   = 0.022 UA1   = 4.31E‐9
+UB1   = ‐7.61E‐18 UC1   = ‐5.6E‐11   AT = 3.3E4
+WL = 0 WLN   = 1 WW = 0
+WWN   = 1 WWL   = 0 LL = 0
+LLN   = 1 LW = 0 LWN   = 1
+LWL   = 0 CAPMOD = 2 XPART   = 0.5
+CGDO = 2.32E‐10   CGSO = 2.32E‐10   CGBO = 1E‐9
+CJ = 7.158188E‐4 PB = 0.8698145 MJ = 0.4849554
+CJSW = 2.305559E‐10   PBSW = 0.8 MJSW = 0.2637406
+CJSWG   = 6.4E‐11 PBSWG   = 0.8 MJSWG   = 0.2261452
+CF = 0 PVTH0   = 5.98016E‐3   PRDSW   = 14.8598424
+PK2   = 3.73981E‐3   WKETA   = 0.0126775 LKETA   = ‐0.021084   )
*
*

287

Hermetic Coating Test Chip 6 (HCTC ‐6) – MOSIS Run V29J

288

Hermetic Coating Test Chip 7 (HCTC ‐7) – MOSIS Run V29J

289

  MOSIS WAFER ELECTRICAL TESTS

RUN: V29J VENDOR: AMIS (ON‐SEMI)
  TECHNOLOGY: SCN05 FEATURE SIZE: 0.5 microns
  Run type: SHR

INTRODUCTION: This report contains the lot average results obtained by MOSIS
from measurements of MOSIS test structures on each wafer of
this fabrication lot. SPICE parameters obtained from similar
measurements on a selected wafer are also attached.

COMMENTS: SMSCN3ME06_ON‐SEMI

TRANSISTOR PARAMETERS   W/L N‐CHANNEL P‐CHANNEL UNITS

MINIMUM 3.0/0.6
Vth 0.78   ‐0.89 volts

SHORT 20.0/0.6
Idss   474 ‐257   uA/um
Vth 0.66   ‐0.87 volts
Vpt 12.7   ‐12.0 volts

WIDE   20.0/0.6
Ids0   < 2.5      < 2.5      pA/um

LARGE 50/50
Vth 0.67   ‐0.92 volts
Vjbkd 10.6   ‐11.9 volts
Ijlk   194.9 < 50.0     pA
Gamma 0.49   0.55   V^0.5

K' (Uo*Cox/2)   58.8   ‐19.0 uA/V^2
Low‐field Mobility 469.98   151.87   cm^2/V*s

COMMENTS: Poly bias varies with design technology. To account for mask
  bias use the appropriate value for the parameter XL in your
  SPICE model card.
  Design Technology   XL (um) XW (um)
  ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐   ‐‐‐‐‐‐‐ ‐‐‐‐‐‐
  SCMOS_SUBM (lambda=0.30)   0.10   0.00
  SCMOS (lambda=0.35) 0.00   0.20

FOX TRANSISTORS   GATE N+ACTIVE P+ACTIVE UNITS
Vth Poly > 15.0 < ‐15.0   volts

PROCESS PARAMETERS   N+ P+   N_W POLY PLY2_HR   POLY2   M1   UNITS
Sheet Resistance 81.6 105.9   819.7 23.1 1034   41.6   0.09 ohms/sq
Contact Resistance 61.1 162.8 15.9   26.8   ohms
Gate Oxide Thickness 138 angstrom

PROCESS PARAMETERS   M2 M3   N_W   UNITS
Sheet Resistance   0.09 0.05 814   ohms/sq
Contact Resistance   0.86 0.84 ohms

290

CAPACITANCE PARAMETERS   N+   P+   POLY   POLY2   M1   M2   M3   N_W UNITS
Area (substrate) 421 705 84   28   12 8   90   aF/um^2
Area (N+active)   2496   37   17   12 aF/um^2
Area (P+active)   2400   aF/um^2
Area (poly) 897   66   16 9 aF/um^2
Area (poly2)   57 aF/um^2
Area (metal1)   32   13 aF/um^2
Area (metal2) 31 aF/um^2
Fringe (substrate) 358 244   53   33   27 aF/um
Fringe (poly) 69   39   28 aF/um
Fringe (metal1)   54   33 aF/um
Fringe (metal2) 52 aF/um
Overlap (N+active) 188 aF/um
Overlap (P+active) 235 aF/um

CIRCUIT PARAMETERS UNITS
Inverters   K
Vinv 1.0   2.04 volts
Vinv 1.5   2.30 volts
Vol (100 uA) 2.0   0.45 volts
Voh (100 uA) 2.0   4.49 volts
Vinv 2.0   2.49 volts
Gain 2.0   ‐17.32
Ring Oscillator Freq.
DIV256 (31‐stg,5.0V) 104.34 MHz
D256_WIDE (31‐stg,5.0V)   158.06 MHz
Ring Oscillator Power
DIV256 (31‐stg,5.0V) 0.49 uW/MHz/gate
D256_WIDE (31‐stg,5.0V)   1.01 uW/MHz/gate

COMMENTS: SUBMICRON

V29J SPICE BSIM3 VERSION 3.1 PARAMETERS

SPICE 3f5 Level 8, Star‐HSPICE Level 49, UTMOST Level 8

* DATE: Nov 30/12
* LOT: v29j WAF: 6102
* Temperature_parameters=Default
.MODEL CMOSN NMOS ( LEVEL   = 49
+VERSION = 3.1 TNOM = 27   TOX   = 1.38E‐8
+XJ = 1.5E‐7   NCH   = 1.7E17   VTH0 = 0.5931459
+K1 = 0.879253   K2 = ‐0.0968711   K3 = 22.6428645
+K3B   = ‐9.849049 W0 = 2.636842E‐8 NLX   = 1E‐9
+DVT0W   = 0 DVT1W   = 0 DVT2W   = 0
+DVT0 = 0.6755795 DVT1 = 0.27608 DVT2 = ‐0.4974617
+U0 = 454.8293133 UA = 1E‐13 UB = 1.306857E‐18
+UC = 6.003353E‐12   VSAT = 2E5 A0 = 0.5426773
+AGS   = 0.1060338 B0 = 1.852719E‐6 B1 = 5E‐6
+KETA = ‐2.682846E‐3   A1 = 0 A2 = 0.3
+RDSW = 856.9792774 PRWG = 0.1467648 PRWB = 0.0170317
+WR = 1 WINT = 2.058331E‐7 LINT = 6.416175E‐8
+XL = 1E‐7   XW = 0 DWG   = 2.559387E‐10
+DWB   = 2.956012E‐8 VOFF = 0 NFACTOR = 0
+CIT   = 0 CDSC = 2.4E‐4   CDSCD   = 0
+CDSCB   = 0 ETA0 = 1.691727E‐3 ETAB = ‐2.990578E‐4
+DSUB = 0.0620642 PCLM = 2.1398211 PDIBLC1 = 1.636689E‐4

291

+PDIBLC2 = 1.72093E‐3   PDIBLCB = 0.0944082 DROUT   = 3.292468E‐3
+PSCBE1 = 3.840311E8   PSCBE2 = 3.949241E‐6 PVAG = 0
+DELTA   = 0.01   RSH   = 81.6   MOBMOD = 1
+PRT   = 0 UTE   = ‐1.5   KT1   = ‐0.11
+KT1L = 0 KT2   = 0.022 UA1   = 4.31E‐9
+UB1   = ‐7.61E‐18 UC1   = ‐5.6E‐11   AT = 3.3E4
+WL = 0 WLN   = 1 WW = 0
+WWN   = 1 WWL   = 0 LL = 0
+LLN   = 1 LW = 0 LWN   = 1
+LWL   = 0 CAPMOD = 2 XPART   = 0.5
+CGDO = 1.88E‐10   CGSO = 1.88E‐10   CGBO = 1E‐9
+CJ = 4.189612E‐4 PB = 0.8362037 MJ = 0.4268727
+CJSW = 3.511622E‐10   PBSW = 0.8 MJSW = 0.2006042
+CJSWG   = 1.64E‐10   PBSWG   = 0.8 MJSWG   = 0.2019414
+CF = 0 PVTH0   = 0.0862532 PRDSW   = 84.413531
+PK2   = ‐0.0885087   WKETA   = ‐0.0164054   LKETA   = 1.749206E‐3   )
*
*
.MODEL CMOSP PMOS ( LEVEL   = 49
+VERSION = 3.1 TNOM = 27   TOX   = 1.38E‐8
+XJ = 1.5E‐7   NCH   = 1.7E17   VTH0 = ‐0.9152268
+K1 = 0.553472   K2 = 7.871921E‐3 K3 = 2.8768851
+K3B   = 2.0233456 W0 = 5.780172E‐7 NLX   = 1.005775E‐9
+DVT0W   = 0 DVT1W   = 0 DVT2W   = 0
+DVT0 = 0.4714461 DVT1 = 0.1852157 DVT2 = ‐0.3
+U0 = 201.3603195 UA = 2.48572E‐9   UB = 1.005454E‐21
+UC = ‐1E‐10   VSAT = 1.051486E5   A0 = 0.7471706
+AGS   = 0.1277893 B0 = 7.349251E‐7 B1 = 2.776521E‐8
+KETA = ‐4.865785E‐3   A1 = 3.090478E‐4 A2 = 0.5651395
+RDSW = 3E3 PRWG = ‐0.0219617   PRWB = ‐0.0909377
+WR = 1.01   WINT = 2.212303E‐7 LINT = 9.977278E‐8
+XL = 1E‐7   XW = 0 DWG   = ‐4.82616E‐10
+DWB   = ‐1.585E‐8 VOFF = ‐0.0619165   NFACTOR = 0.2482253
+CIT   = 0 CDSC = 2.4E‐4   CDSCD   = 0
+CDSCB   = 0 ETA0 = 9.384854E‐3 ETAB = ‐0.2
+DSUB = 1 PCLM = 2.3408026 PDIBLC1 = 0.0767278
+PDIBLC2 = 4.024702E‐3 PDIBLCB = ‐0.0443178   DROUT   = 0.2659121
+PSCBE1 = 8E10   PSCBE2 = 8.966681E‐8 PVAG = 0.0149502
+DELTA   = 0.01   RSH   = 105.9 MOBMOD = 1
+PRT   = 0 UTE   = ‐1.5   KT1   = ‐0.11
+KT1L = 0 KT2   = 0.022 UA1   = 4.31E‐9
+UB1   = ‐7.61E‐18 UC1   = ‐5.6E‐11   AT = 3.3E4
+WL = 0 WLN   = 1 WW = 0
+WWN   = 1 WWL   = 0 LL = 0
+LLN   = 1 LW = 0 LWN   = 1
+LWL   = 0 CAPMOD = 2 XPART   = 0.5
+CGDO = 2.35E‐10   CGSO = 2.35E‐10   CGBO = 1E‐9
+CJ = 7.015391E‐4 PB = 0.8644163 MJ = 0.4849925
+CJSW = 2.448774E‐10   PBSW = 0.8 MJSW = 0.2031512
+CJSWG   = 6.4E‐11 PBSWG   = 0.8 MJSWG   = 0.2261452
+CF = 0 PVTH0   = 5.98016E‐3   PRDSW   = 14.8598424
+PK2   = 3.73981E‐3   WKETA   = 0.0140638 LKETA   = ‐0.0170643 )
*
*

Abstract (if available)

Abstract Blindness due to Age-Related Macular Degeneration and Retinitis Pigmentosa is unfortunately both widespread and largely incurable. Advances in visual prostheses that can restore functional vision in those afflicted by these diseases have evolved rapidly from new areas of research in ophthalmology and biomedical engineering. This thesis is focused on further advancing the state-of-the-art of both visual prostheses and implantable biomedical devices. ❧ A novel real-time system with a high performance head-mounted display is described that enables enhanced realistic simulation of intraocular retinal prostheses. A set of visual psychophysics experiments is presented using the visual prosthesis simulator that quantify, in several ways, the benefit of foveation afforded by an eye-pointed camera (such as an eye-tracked extraocular camera or an implantable intraocular camera) as compared with a head-pointed camera. A visual search experiment demonstrates a significant improvement in the time to locate a target on a screen when using an eye-pointed camera. A reach and grasp experiment demonstrates a 20% to 70% improvement in time to grasp an object when using an eye-pointed camera, with the improvement maximized when the percept is blurred. A navigation and mobility experiment shows a 10% faster walking speed and a 50% better ability to avoid obstacles when using an eye-pointed camera. ❧ Improvements to implantable biomedical devices are also described, including the design and testing of VLSI-integrable positive mobile ion contamination sensors and humidity sensors that can validate the hermeticity of biomedical device packages encapsulated by hermetic coatings, and can provide early warning of leaks or contamination that may jeopardize the implant. The positive mobile ion contamination sensors are shown to be sensitive to externally applied contamination. A model is proposed to describe sensitivity as a function of device geometry, and verified experimentally. Guidelines are provided on the use of spare CMOS oxide and metal layers to maximize the hermeticity of an implantable microchip. In addition, results are presented on the design and testing of small form factor, very low power, integrated CMOS clock generation circuits that are stable enough to drive commercial image sensor arrays, and therefore can be incorporated in an intraocular camera for retinal prostheses.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Novel imaging systems for intraocular retinal prostheses and wearable visual aids

$Intraocular camera for retinal prostheses: refractive and diffractive lens systems$

PDF

Intraocular camera for retinal prostheses: refractive and diffractive lens systems

PDF

PET study of retinal prosthesis functionality

PDF

Saliency based image processing to aid retinal prosthesis recipients

PDF

Manipulation of RGCs response using different stimulation strategies for retinal prosthesis

PDF

Prosthetic visual perception: retinal electrical stimulation in blind human patients

PDF

Towards a high resolution retinal implant

PDF

Oxygen therapy for the treatment of retinal ischemia

PDF

Adaptive event-driven simulation strategies for accurate and high performance retinal simulation

PDF

Prosthetic vision in blind human patients: Predicting the percepts of epiretinal stimulation

PDF

RGBD camera based wearable indoor navigation system for the visually impaired

PDF

Robot vision for the visually impaired

PDF

Investigation of the electrode-tissue interface of retinal prostheses

PDF

Stimulation strategies to improve efficiency and temporal resolution of epiretinal prostheses

PDF

Viscoelastic characterization of soft biological materials

PDF

An analytical and experimental study of evolving 3D deformation fields using vision-based approaches

PDF

Therapeutic electrical stimulation strategies for neuroregeneration and neuroprotection of retinal neurons