1
DOCTORAL THESIS
DESIGN CONTRIBUTIONS TO HIGH-END LOW-NOISE
CMOS IMAGE SENSORS
JOSÉ ÁNGEL SEGOVIA DE LA TORRE
Doc o a e P og am
Physical Sciences and Technologies
Ad iso :
P o . Ángel Rod íguez Vázquez
Tu o :
P o . Ángel Rod íguez Vázquez
Se illa, 2024
2
3
To my wi e E e o always suppo ing me
To my child en Pila , Ra ael and Fe nando
4
ACKNOWLEDGMENTS
The de elopmen o a doc o al hesis and he linked esea ch p ojec linked is a long jou ney, wi h many
di icul ies, un ui ul esul s bu wi h also qui e impo an and sa is ac o y achie emen s ha a e he esul
o a con inuous collabo a i e wo k in an adequa e en i onmen . I is e en mo e impo an , in he doc o al
hesis de eloped in he company in collabo a ion wi h he Uni e si y which is he case o his wo k.
None heless, apa om he esul s is e y impo an he knowledge and he s ong and p o essional
ela ionship de eloped along his jou ney. This hesis can be seen as an indi idual d eam ha could no
become eal wi hou he collabo a ion, he suppo and he iendship o hese people ha ha e been along
wi h me in his jou ney.
Fi s o all, I would like o exp ess my g a i ude o Ángel Rod íguez Vazquez as p o esso and ad iso o
his Thesis. He has help me no only wi h he echnical eaching, ad ice and suppo bu also he has
encou aged me o ollow and o push o achie e his goal. Addi ionally, I would like o say hanks, lo o
hanks, o my p o esso , u o , and iend Fe nando Medei o, who un o una ely passed away a he
beginning o he Thesis, and we could no con inue his jou ney oge he , I will be always g a e ul o he
ma k he p o ided on me as a p o essional and as a pe son, hanks Fe nando.
Secondly and wi h same le el o g a i ude, I wan o hank o he suppo ob ained by Ana Gonzalez and
Ra ael Romay o his Thesis, as manage s in my company Teledyne Ana ocus. They a e doing a e y good job
suppo ing and encou aging indi iduals o achie e hei own a ge s, as my case. As iends I ha e o say I am
lucky o be along wi h hem wo king and ying o add ess new ma ke s, o achie e ul ima e de ices and o
build ou s anding eam and enjoying wi h all o ha .
The hi d and no less impo an , I would like o say lo o hanks o my wi e, E a, who encou aged me o
ha e he decision o doing he Thesis when I exp essed my in e es ed on ha . Also, I would like o say lo o
hanks o my daugh e Pila , who also ook ca e o my o he wo child en Ra ael and Fe nando, she is e y
esponsible e en a he age. Thanks o all o hem, especially E a, who has suppo ed me, aking ca e o he
child en along wi h my mo he in Law, Ma i, and o he child en o he ime I ha e s olen hem, I am in deb .
Addi ionally, I wan o say hanks o many colleagues in Teledyne Ana ocus: Alex Cha le , he i s pe son
I s a ed wo king wi h in mic oelec onic design. Ra ael Dominguez hanks o he endless discussions and
he con inuous lea ning om him. Albe o Villegas who also has collabo a ed in he p ojec s ela ed wi h
his Thesis, his kindness and good job a e imp essi e. Loli Pa do who is leading he design eam and being
pa icipan in he digi al design o many p ojec s. Jesus Ruiz, An onio O iz de Galis eo, Paco Bena ides who
a e leading he echnical p ojec s wi h pe sis ence and ha d wo k. The es o he analog eam: Joao Bo elho,
Jesus Aguado, Blanca Piñe o, Nu ia Po cel, and he es . I would like o say hanks o all he good
con e sa ions and good ela ionship in Teledyne Ana ocus. The people in he cha ac e iza ion eam, who
also help in his Thesis: Gema Valles, Ma a Salamanca and Ma io Gue a. And in he es o eams in Teledyne
Ana ocus: Luis Salas, Juan Lis an, Caye ana U e a, Wal e Iandolo, Ca los Mendoza, Luis Alba, Ra a Payseo
wi h all o hem I ha e sha ed discussions and echnical exchanges, I would like o say hanks o hem as well.
And inally, I would like o exp ess my g a i ude o my pa en s and siblings, hey ha e augh me he e o ,
esilience and ha d wo k b ings ui ul esul s, and also hey ha e been suppo ing me om he momen I
le home o s a Uni e si y s udies.
5
ABSTRACT
The main objec i e o his hesis consis s o p o iding a new eadou a chi ec u e able o mee wo o he
mos demanding speci ica ions in image senso : low noise and speed. The p esen ed eadou channel
p o ides 1.3 eRMS noise images wi h a con e sion ime sho e han i e mic oseconds, wi h he a ailabili y
o exchange p ecision and speed achie ing 0.7 eRMS when ex ending he ime up o 80 mic oseconds.
Addi ionally, a pixel op imiza ion p ocedu e has been de eloped based on expe imen al da a and a model o
ex apola e he empi ical da a o he expec ed pe o mances wi h he new eadou a chi ec u e. On op o
ha , an a chi ec u e modi ica ion o he eadou channel is p o ided o wo k wi h dual con e sion gain
pixels in o de o p o ide high dynamic ange images. I has been p o ed, ha he eadou channel is alid
o la ge scale wa e de ices and one e icle de ices, wo p o o ypes a e desc ibed wi h he eadou channel
desc ibed in his Thesis. In hese p o o ypes, no only he eadou ea u es a e desc ibed, bu also o he
echniques o educe glow e ec , imp o e yield and he eadou calib a ion p ocess. Finally, a bench ma k is
p o ided o show a compa ison o he noise speed compa ison wi h o he publica ions in he s a e o he a
showing he bene i s o he p oposed a chi ec u e.
The chap e s in his Thesis a e o ganized as ollows:
• Chap e 1 a b ie snapsho abou he s a us and ends o he mic oelec onics and mic osys ems
and he impo an posi ion o he image senso s in mic oelec onic nowadays. Also, i p o ides he
e olu ion o image senso s along he las decades o he cu en si ua ion whe e mobile leads he
cu en ma ke , bu o he eme ging applica ions demands new unc ionali ies om he CIS
echnology. A e ha , he ypes o noise in image senso s a e desc ibed wi h emphasis in empo al
noise sou ces, whe e he pixel noise con ibu o s a e desc ibed in de ail.
• Chap e 2 p esen s he pixel noise and linea i y op imiza ion s udy based on a 6.5 um 4T CMOS pixel
wi h a na i e sou ce ollowe . This op imiza ion p ocedu e in ol es designing and cha ac e izing a
low-noise noise chip wi h se e al pixel a ian s whe e ansis o dimensions and o he changes a e
es ed and cha ac e ized. The op imiza ion o ope a ion ol ages has been shown o imp o e noise
and linea i y pe o mances. A semi-empi ical model has been de eloped o ex apola e da a om
he cha ac e iza ion es chip o he new eadou channel p esen ed in his Thesis. The Chap e ends
wi h he selec ion o a pixel a ian .
• Chap e 3 desc ibes he p oposed low-noise eadou channel based on wo s ages o ADC:
Inc emen al ADC and Single-Slope ADC. The Chap e also shows he unc ional and implemen a ion
desc ip ions and noise analysis o e e y ADC s age. The eadou channel can ope a e wi h in e nal
analog CDS and ex e nal digi al CDS. The Thesis discusses he wo ope a ion modes, showing he
noise pe o mances in simula ion and da a measu emen s. The analog amp gene a ion base in he
cu en s ee ing DAC is also p esen ed. Finally, he wo s ages o ADC need calib a ion o align he
anges and a oid missing codes o gaps in he con e sion. A echnique o calib a ion is ou lined, and
he me hod o implemen ing his echnique in eal de ices o online calib a ion is discussed.
• Chap e 4 p esen s a a ia ion o he eadou channel o high dynamic ange applica ions ha
employs a dual con e sion gain o 10 μm pixels wi h op imized eadou . This chap e s a s by
e ising he s a e-o - he-a ega ding high dynamic ange echniques o suppo selec ing he
me hods de eloped in his Thesis. The pixel a chi ec u e wi h dual con e sion gain is p esen ed and
he op imiza ion based on a es -chip is shown. Op imiza ion elies on a ying de ice dimensions o
op imize maximum FWC and minimum empo al noise. Finally, he ADC a chi ec u e op imiza ion
based on he p e ious eadou o dual con e sion gain is desc ibed.
• Chap e 5 p esen s wo image senso p o o ypes inco po a ing he low noise and high dynamic ange
desc ibed in his Thesis. The i s image senso is a 5.3 Megapixel de ice wi h 6.5 μm 4T CMOS pixel
deli e ing images wi h 0.7 e-RMS o empo al noise and 15000 elec ons o ull well capaci y. The
6
second image senso is a la ge-scale, 66 Megapixel de ice wi h 10um dual con e sion pixels. This
de ice can collec mo e han 100000 elec ons wi h 1 e-RMS noise achie ing 100 dB dynamic ange.
These wo de ices use di e en echniques o educe undesi able e ec s, such as he he mionic
emission om he pe iphe al ci cui y, known as he glow e ec . Addi ionally, me hods o imp o e
yield in la ge size de ices a e p esen ed.
Du ing he de elopmen and cha ac e iza ion o hose image senso p o o ypes impo an miles ones
ha e been add essed as ollows:
• Glow E ec : The glow is a limi ing phenomenon ha doesn’ allow educing he da k cu en when
cooling he image senso e en wi h nega i es empe a u es. In his hesis a p ocess ab ica ion and
a me hodology has been de eloped o minimize a e y low le els he glow con ibu ion. I is based
on he use o aluminum laye in o de o co e all he ac i e ci cui s ha may emi ligh , and
addi ionally c ea e a e ical ench co e ed wi h he same aluminum laye o block he ligh wi h is
a elling h ough he silicon o he pixel a ay. This glow e ec is qui e impo an in BSI de ices since
he mul idi ec ional emission c ea e mul ipa hs o he ligh ha each he pixel a ay.
• Yield imp o emen s: This poin is impo an specially o wa e scale la ge image senso s, whe e one
de ec in he wa e means a de ec i e de ice. A scalable epai ing ci cui y has been p oposed and
inco po a ed in he image senso o imp o e yield. Addi ionally, special DFM ules ha e been used
in he pixel a ay as using only one me al line pe me al laye and using always i is possible wo
con ac s and wo ias minimum.
• Quan um e iciency: Ve y high quan um e iciency has been achie ed using BSI de ices. This
achie emen is mainly d i en by he echnology, bu he pixel design has been done compa ible wi h
BSI echnology. I is ema kable he lack o e aloning (in e e ome y due o in e e ence wi h me al
laye s), which can be e y impo an o applica ions like spec oscopy.
• Ope a ion Modes: mul iple ope a ion modes and unc ions a e implemen ed o inc emen he image
senso lexibili y and o p o ide a ailo ed ope a ion mode o each applica ion.
Finally, a benchma k compa ison o his wo k wi h o he con ibu ions in he s a e-o - he-a has been
de eloped. This benchma k shows he image senso and eadou channel p o ide e y low noise esul s a
high-speed conside ing a e y app op ia e op ion o he low noise applica ions. In ac , he eadou noise is
p o iding e y low noise, and hen i we we e able o p o ide wi h a be e pixel in e ms o noise, he
esul an image senso would p o ide be e pe o mances wi h any change in he eadou channel.
7
TABLE OF CONTENTS
CHAPTER 1 ........................................................................................................................................................ 13
1. INTRODUCTION TO LOW NOISE IMAGE SENSORS ......................................................................................... 13
1.1 A glimpse on mic oelec onic and mic osys em ends ................................................................... 14
1.2 Snapsho s abou CIS E olu ion ....................................................................................................... 16
1.2.1 In oduc ion .............................................................................................................................. 16
1.2.2 O e iew o CIS e olu ion miles ones ..................................................................................... 21
1.3 Noise in image senso s ................................................................................................................... 25
1.3.1 Pixel noise analysis ................................................................................................................. 27
1.3.2 Sou ce ollowe noise .............................................................................................................. 29
1.3.3 Pho on sho noise .................................................................................................................... 31
1.3.4 Da k Cu en noise ................................................................................................................... 32
1.3.5 T ans e Noise ......................................................................................................................... 34
1.3.6 Rese noise .............................................................................................................................. 34
1.4 Readou a chi ec u es in image senso s ......................................................................................... 36
1.4.1 Common noise sou ce in ADCs............................................................................................... 36
1.4.2 Column ampli ica ion ............................................................................................................... 40
1.4.3 Co ela ed double sampling ..................................................................................................... 41
1.5 Thesis ou line and o ganiza ion ....................................................................................................... 42
CHAPTER 2 ........................................................................................................................................................ 45
2. PIXEL NOISE OPTIMIZATION ........................................................................................................................ 45
2.1 Tes s uc u es pixel a ian s and cha ac e iza ion esul s .............................................................. 46
2.1.1 Desc ip ion o pixel a ian s and wa e o ms ........................................................................... 46
2.1.2 Tes ehicle chip a chi ec u e .................................................................................................. 48
2.1.3 Cha ac e iza ion esul s ........................................................................................................... 50
2.2 Pixel ol age op imiza ion ................................................................................................................ 56
2.2.1 Vol age op imiza ion o da k noise ......................................................................................... 56
2.3 Pixel sou ce- ollowe noise semi-empi ical model ........................................................................... 58
2.3.1 Pixel noise The mal and 1/ Noises a e CDS ope a ion ........................................................ 58
2.3.2 Pixel noise in CMS ope a ion ................................................................................................... 63
2.4 Model ex apola ion o p oposed ADCs a chi ec u e ...................................................................... 64
CHAPTER 3 ........................................................................................................................................................ 65
3. TWO STAGES ADC FOR LOW NOISE CMOS IMAGE SENSORS ...................................................................... 65
3.1 In oduc ion ...................................................................................................................................... 66
3.1.1 Mo i a ion ................................................................................................................................. 66
3.1.2 Ou line o p oposed eadou a chi ec u e ................................................................................ 68
3.2 Fi s s age: inc emen al ADC ........................................................................................................... 70
3.2.1 Func ional desc ip ion .............................................................................................................. 70
3.2.2 Implemen a ion ........................................................................................................................ 71
3.2.3 Noise Analysis ......................................................................................................................... 71
3.3 Second s age: SS-ADC ................................................................................................................... 72
3.3.1 Func ional desc ip ion .............................................................................................................. 73
3.3.2 Implemen a ion ........................................................................................................................ 74
3.3.3 Noise Analysis ......................................................................................................................... 74
3.4 Analog amp gene a o .................................................................................................................... 75
3.5 Readou wo king in ex e nal CDS mode ......................................................................................... 76
3.5.1 Func ional desc ip ion and implemen a ion ............................................................................. 76
3.5.2 Noise Analysis ......................................................................................................................... 77
3.6 Readou channel pe o mance ........................................................................................................ 78
3.6.1 Linea i y ................................................................................................................................... 78
3.6.2 Noise ........................................................................................................................................ 80
3.7 Readou wo king in co ela ed mul isampling wi h ex e nal CDS .................................................... 81
3.8 Readou calib a ion on-chip ............................................................................................................. 84
CHAPTER 4 ........................................................................................................................................................ 85
4. TWO STAGES ADC EXTENSION FOR DUAL CONVERSION GAIN HDR .............................................................. 85
4.1 High Dynamic Range s a e-o - he-a e iew ................................................................................... 86
4.1.1 Mul iple exposu es HDR me hods ........................................................................................... 86
4.1.2 Mul iple gains HDR me hod ..................................................................................................... 89
4.1.3 Loga i hmic pixel ...................................................................................................................... 96
4.1.4 Techniques based on pe -pixel ADCs ................................................................................... 100
4.1.5 Time o sa u a ion pixel .......................................................................................................... 102
4.1.6 HDR echniques compa a i e ................................................................................................ 103
4.2 Pixel a chi ec u e ........................................................................................................................... 104
4.2.1 Pixel schema ics and wa e o ms........................................................................................... 104
4.2.2 Tes chip ................................................................................................................................ 106
4.2.3 Pixel pe o mance .................................................................................................................. 106
4.3 ADC a chi ec u e ........................................................................................................................... 110
4.3.1 Dual con e sion ADC unc ional desc ip ion ......................................................................... 110
CHAPTER 5 ...................................................................................................................................................... 113
5. LOW NOISE CISS WITH OPTIMIZED PIXELS,,DUAL GAIN AND OVERSAMPLED ADCS .................................. 113
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise ........................................................... 114
5.1.1 Gene al desc ip ion ................................................................................................................ 114
5.1.2 Senso A chi ec u e ............................................................................................................... 117
5.1.3 Ope a ion modes ................................................................................................................... 119
5.1.4 ULN5.3 Senso cha ac e iza ion ............................................................................................ 126
5.1.5 Op ical pe o mances in 14 bi s olling-shu e mode ............................................................ 131
5.1.6 Op ical pe o mances in 16 bi s olling-shu e mode ............................................................ 132
5.1.7 Pe o mances in 14 bi s mul isampling (M = 16) olling-shu e mode .................................. 133
5.1.8 Da k cu en ........................................................................................................................... 134
5.1.9 Glow e ec and mi iga ion echniques ................................................................................... 135
5.1.10 Low ligh benchma k compa ison .......................................................................................... 135
5.2 La ge scale 66 Mpixel 10 mm pixel low noise and high ................................................................ 139
5.2.1 Senso A chi ec u e ............................................................................................................... 141
9
5.2.2 Ope a ion modes ................................................................................................................... 143
5.2.3 Senso unc ions .................................................................................................................... 148
5.2.4 Senso yield imp o emen ..................................................................................................... 148
5.2.5 Op ical pe o mances measu emen s .................................................................................... 149
5.2.6 ULN66 in he ac i e cooling came a ...................................................................................... 157
5.3 Low Noise Benchma k compa ison ............................................................................................... 159
6. CONCLUSIONS, CONTRIBUTIONS AND FUTURE WORK ................................................................................ 163
Chap e 1 In oduc ion o Low Noise Image Senso s
16
P e ious pa ag aphs can be summa ized in a simple s a emen :
Images and he isual sense (ex ac ion o in o ma ion om images) a e c ucial o a i icial
sensing-in e ac ing mic osys ems
Mode n daily li e p o ides mul iple examples suppo ing he p e ious s a emen . In mode n Socie y, mos o
us ca y pe sonal elec onic appliances ha embed CMOS Image Senso (CIS) de ices. We use hem
cus oma ily o ake pho os and ideos o cap u e and e ain ou memo ies. The sec o s o mobile phones
and pe sonal elec onics demand huge amoun s o silicon wa e s being p ocessed by he semiconduc o
ound ies. Howe e , CISs a e p esen in many o he applica ions, om molecula imaging o as onomy
obse a ion, including au omo i e, medical diagnosis, b ain-machine in e acing, indus ial inspec ion lines,
su eillance, e c.
Fu he mo e, many unexpec ed applica ion scena ios will likely be opened, p omp ed by eme ging
pa adigms like he In e ne o Things, among o he s. Finally, he abili y yielded by CMOS o combine
de ec o s wi h p ocessing ci cui y enables inco po a ing image analysis ea u es in he same silicon
subs a e whe e images a e cap u ed, hus pa ing he way o he implemen a ion o CVISs (CMOS VIsion
Senso s), an e olu ion o CISs capable o comple ing ision asks on-chip [Rod i17].
The inc eased ma u i y o 3D,
wa e -s acked echnologies
le e ages CISa and CVISs
capabili ies by enabling he
dis ibu ion o non-sensing ci cui
esou ces (p ocessing, calib a ion,
con ol, memo y, e c.) and, hus,
allowing o educe he pixel pi ch
and inc ease he ill ac o
3
.
Fig. 4 a he igh inse , o Yole
De elopmen , illus a es he wide
deploymen o CISs de ices.
ILLUSTRATING CIS SECTORS
(h ps://www.yole. )
1.2 SNAPSHOTS ABOUT CIS EVOLUTION
1.2.1 In oduc ion
F om ea ly imaging o CCD
“A pic u e ells a housand wo ds.” This idiom highligh s he impo ance o images and op ical scenes as
in o ma ion ca ie s in mode n mic osys ems. Using images o con ey in o ma ion is a human p ac ice since
he ea ly ages o mankind. Wi h a bi much la ge sophis ica ion han p imi i e schema ic d awings, he i s
app oach o a came a was he came a obscu a which p inciples a e oo ed o 470 BC and he wo ks o he
Chinese philosophe Mozi. This came a ope a es on he ealms o he phenomena occu ing when he ligh
goes h ough a small hole and he image in one side is p ojec ed in he opposi e su ace and gene a e he
in e ed image [Newha82]. Al eady in he 11 h cen u y he A ab physicis Ibn Al-Hay ham w o e he book
3
Pixel pi ch and ill ac o should no equi e de ini ions in a specialized monog aph like his Thesis. Howe e , CVISs comp omise bo h ea u es as
hey include mo e non-ac i e ci cui s a he pixel le el, i.e., ci cui s no dedica ed o cap u ing pho ons, han CISs. Thus, pixel dimensions (pi ch)
and ill ac o (pe cen age o he ac i e ci cui y pe pixel) ge penalized. 3D s acking is pa icula ly ele an o CVISs, enabling be e pi ch and ill
ac o s h ough ac i e and non-ac i e ci cui y e ical dis ibu ion pe laye .
1.2 Snapsho s abou CIS E olu ion _
17
o op ics (Book o Op ics - Wikipedia). This ea ly s andou wo k in op ics s a es he i s undamen als laws
o e lexion and e ac ion along wi h he analysis and cons uc ion o i s lenses.
A he beginning o he 19 h c cen u y, Nicépho e Niépce a F ench in en o and pho og aphe (Nicépho e
Niépce - Wikipedia) ob ained he i s pho og aph by using a small came a, wi h a small hole
4
and sil e
chlo ide ilm, which became da kened when i was exposing o he ligh . A e ha , some p og ess by using
o he ma e ials and p ocess we e done up o he d y pla e, which inc eases he sensi i i y allowing he so-
called “ins an aneous snapsho ”. These p og esses lead o he in en ion o he he pho og aphic ilm came a
in 1885. The in en o o his key miles one, Geo ge Eas man called his i s came a:
KODAK
I was o e ed o sale in 1988 [Kodak19] . Kodak company de eloped he i s ue colo ilm in 1942 o s ill
pho og aphy. The oll ilm came a has been ex ensi ely used un il he end o he 20 h cen u y when he
elec onic pho og aphy eplaced hem.
Mode n physics has b ough many ad ances conce ning analyzing he dual na u e o ligh , which is bo h
an elec omagne ic wa e and a s eam o pho on pa icles ha a e he quan um o ligh . These new concep s
and heo ies we e con ibu ed by many geniuses, om Huyggens and Maxwell o Max Planck and Albe
Eins ein, among many o he cha ac e s. Maxwell’s equa ions can desc ibe elec omagne ic ileds and a e
e y use ul o s udying p opaga ion, bu objec s abso b and emi ligh in quan ized ene gy packe s
associa ed o pho on pa icles. These ad ances con e ged wi h solid-s a e elec onics du ing he 20 h
cen u y o suppo he in en ion o Cha ge Coupled De ice (CCD) de ec o s in 1969 [Boyle70] – an in en ion
awa ded he Nobel P ize in Physics in 2009. CCDs employ s uc u es simila o mode n CISs, namely
capaci o s suppo ing he ans e o cha ge packe s and deple ed MOS s uc u es as pho ode ec o s. They
expe ienced many signi ican ad ances o e he yea s [Smi h09]; howe e , hey a e in insically limi ed by
he di icul ies o embedding p ocessing ci cui y besides he de ec o s. Thus, hey ha e been p og essi ely
eplaced by CISs [Fossu97][Fossu14], al hough hey s ill ha e some niche in as onomy and scien i ic
ins umen a ion applica ions [Janes01].
F om CCDs o CISs
CISs ha e e ol ed o p og essi ely eplaced CCDs ueled by:
• The possibili y o embed mul iple unc ions oge he wi h he de ec o s. Thus, mode n CISs a e
comple e sys ems, nea ly came a-on-chip, capable o deli e ing ully digi al images ins ead o « jus »
analog aw da a. As men ioned, CVISs go a in his di ec ion by including e en image analysis on he
chip – eade s a e cau ioned ha CVISs a e no a ge ed in his Thesis.
• The in en ion o he Pinned Pho o Diode (PPD) by Nippon Elec ic Co L d esea che s in 1980. This
key in en ion by N. Te anishi, H. Shi aki and Y. Ishiha a o e came image quali y d awbacks o
p e ious CMOS-compa ible pho odiodes, hus pe mi ing o achie e acquisi ion quali y close o ha
o CCDs in e ms o noise, da k cu en (leading o images wi h no illumina ion) and o he
pe o mance ea u es[Fossu14] .
CISs deli e images when i adia ed by ligh a equencies co e ing oughly he isible and he nea
in a ed bands o he elec omagne ic spec um: [400nm < < 1000nm]. Image cap u e by CIS in ol es,
among o he p ocesses, pho on abso p ion and pho on- o-cha ge ans o ma ion; hese p ocesses happen
bo h a pho o-senso de ices embedded in o a silicon semiconduc o subs a e. The e o e his subs a e
becomes an ac i e, pho osensi i e “plane” whe e ligh ge s ocused by op ical lenses. This is he eason why
his ac i e semiconduc o plane is some imes called ocal plane.
The ocal plane also includes non-sensi i e ci cui s employed o di e en pu poses. Fig. 5, which
co esponds o a senso concei ed in his Thesis, illus a es he concep ual block diag am o a mode n CIS
4
In e es ingly, pin-hole lenses a e s ill used oday o some low-cos came a modules in some applica ions [Gomez23].
Chap e 1 In oduc ion o Low Noise Image Senso s
18
chip designed o deli e digi ally-encoded images. When implemen ed in plana echnologies, all hese blocks
a e embedded in he ocal plane, al hough only he pixel a ay. This co e pa in Fig. 5(a) consis s o a egula
a angemen o pho o-sensi i e s uc u es ha c ea e ma hema ical ep esen a ions o he incoming images
comp ised o a ma ix o ol ages “p opo ional” o he ligh ene gy abso bed a he spa ial egion occupied
by each pixel (spa ial image samples)
5
. The pixel a ay (see he illus a ion in Fig. 5(b)) comp ises a egula
a angemen o pho o-sensi i e de ices each cap u ing a spa ial sample o an incoming image. The es o
chip unc ions a e cus oma ily loca ed in he chip pe iphe y and mus be masked o educe he impac o
pa asi ic pho on-induced elec on-hole gene a ion ou side he pho osensi i e a ea.
(a) (b)
EXAMPLE OF DIGITAL CMOS IMAGE SENSOR: (A) BLOCK DIAGRAM; (B) ILLUSTRATION OF PIXEL ARRAY ARRANGEMENT
CIS chips can be illumina ed ei he a he op (F on Side Illumina ion – FSI) o a he bo om (Back Side
Illumina ion – BSI). Di e ences be ween hese wo cases a e illus a ed o a colo pixel in Fig. 6. In he case
o BSI, op ical lenses a e placed a he bo om su ace o a back hinned semiconduc o wa e (down o 4-10
m hickness) and he ligh eaches di ec ly he silicon (whe e pho osenso s a e placed) wi hou in e e ing
wi h IC laye s employed o in e connec ion and isola ion. In he case o FSI, op ical lenses a e placed a he
op su ace and ligh mus p opaga e h oughou he op chip laye s and s uc u es be o e eaching he
pho osensi i e silicon.
BSI echnologies ha e he ad an age o p o iding be e quan um e iciency (( he numbe o impinging
pho ons ha p oduce elec on-hole pai – see oo no e 1 and ill ac o , since he e a e no me als blocking
he ligh . Addi ionally o he ad an ages o BSI is he sensi i i y o UV wa e o ms can be much mo e highe
since hese ene ge ic wa e o ms a e ecombined e y close o he su ace and hen ou o he silicon egion
in FSI de ices. Chips epo ed in in his Thesis a e compa ible wi h illumina ion op ions.
5
The p opo ionali y o he esponse is a a ge in mos cases, al hough some applica ions demand a non-linea comp essi e esponse o suppo a
highe dynamic ange. Rega ding sampling, mos mode n CISs sample ac oss space (a ay plane) and h oughou ime (images composed o a
sequence o F ames). Howe e , new classes o CISs a e me ging whe e empo al is asynch onous, d i en by e en s happening in he image
sequence [Leñe18].
...
...
...
...
...
...
...
...
...
...
...
...
A/D
Readou
A/D
Readou
A/D
Readou
A/D
Readou
A/D
Readou
A/D
Readou
Pixel A ay
Ou pu s
Pixel con ol bu e s
1.2 Snapsho s abou CIS E olu ion _
19
ILLUSTRATING FSI AND BSI TECHNO OPTIONS
Despi e he applica ion, CIS pixels include sensi i e and non-sensi i e
de ices and ci cui s. Fig. 7 a he igh inse shows a concep ual oo -p in
o a mode n BSI CIS pixel whe e he ac i e, pho o-sensi i e a ea (in yellow)
amoun s ypically o ~60% o he o al pixel a ea. No e in his illus a ion
ha he BSI sensing wa e is bonded o a suppo wa e made o glass o
silicon and placed a he op su ace o he senso wa e .
ILLUSTRATING PIXEL FOOTPRINT FOR A BSI CIS
Pixel oo -p in can be op imized by eso ing o 3D-s acked echnologies. Fig. 8 illus a es his concep by
showing he spli ing o he senso o Fig. 5 in wo wa e s, namely:
• A sensing wa e ;
• A logic wa e .
In his pa icula example, 3D s acking aims o he op imum dis ibu ion o he pixel a ay, on he one hand,
and con ol-memo y-p ocessing unc ions, on he o he hand. Howe e , he same concep can be used o
op imize he oo -p in a he pixel le el, inside he pixel i sel , as i happens in CVISs [Rod i17] and he newes
class o Dynamic Vision Senso s (DVS) [Kodam23].
“Ci ius, al ius, o ius”, he Olympic mo o, applies o he e olu ion o image s ueled by CIS p ocesses,
con e gen packaging, and he e ogeneous in eg a ion echnologies. Thus, al hough i is always a bi isky o
clus e and compa imen alize echnology e olu ion, some signi ican ends can be iden i ied as desc ibed
in he bulle poin s below:
• One end is ela ed o he p e alen dec ease o he pixel pi ch, he inc ease o he ill ac o , and
he inc ease o he esolu ion (numbe o pixels) and he speed (measu ed in F ames-pe -second in
con en ional CISs). P e alen ends o CIS indus y include dec easing he pixel pi ch and inc easing
he ill ac o s. The pi ch o las -gene a ion consume senso s is below 1 m, al hough o he
applica ions, including machine ision and scien i ic imaging, use la ge pixel pi ch. Fo ins ance, his
Thesis p esen s senso s wi h a pi ch o 6.5m and a ill ac o o 78%.
• Ano he majo end is he p og essi e inco po a ion o image-dep h acquisi ion o pixels and
senso s. Di e en me hods can be used o ha pu pose, including indi ec and di ec ime-o - ligh
measu emen . Sui able pixel a chi ec u es enable e en combining he acquisi ion o in ensi y maps
Chap e 1 In oduc ion o Low Noise Image Senso s
20
(con en ional 2D images) wi h image dep hs (3D images), hus c ea ing mo e de ailed
ep esen a ions o he scenes [Bhand16].
Senso wa e Logic wa e
SPLITTING OF THE SENSOR ARCHITECTURE OF FIG.2 FOR A 3D-STACKAED TECHNOLOGY
• The las end iden i ied in his limi ed clus e ing conce ns embedding image analysis and,
e en ually, ision ask capabili ies in e e y a i icial sys em. Mee ing his a ge may equi e non-
con en ional app oaches depa ing om he con en ional s a egy o ull image digi iza ion and on
Neumann p ocessing a chi ec u es. Indeed, many mode n app oaches a e somewha inspi ed by he
ope a ion o na u al e inas [Roska01] [Ai ag20], he mos e icien imaging on -ends known o
da e. CVISs based on he beha io al ea u es o e inas ob ain ou s anding ene gy e iciencies and
da a h oughpu s, much be e han hose achie ed wi h con en ional Von-Neumann´s a chi ec u es
based on numbe -c unching [Rod i17].
Fig. 9 illus a es majo ac i i y a eas summa ized in he bulle poin s abo e.
ILLUSTRATING CMOS IMAGE AND VISION SENSOR ACTIVITY AREAS
1.2 Snapsho s abou CIS E olu ion _
21
1.2.2 O e iew o CIS e olu ion miles ones
Ea ly CISs
One i s ele an miles one o MOS image senso s da es om
1967, when Weckle epo ed an a angemen composed o
p-n junc ions and scanning swi ches in:
G. Weckle , “Ope a ion o p-n Junc ion Pho odec o s in
a Pho on Flux In eg a ion Mode.” IEEE JOURNAL OF SOLID-
STATE CIRCUITS, Vol. 2, pp. 65–73, 1967.
The inse a he igh illus a es his ea ly con ibu ion, which
employed pixels wi hou embedded ampli ica ion. Hence, i is
a Passi e Pixel Senso (PPS). The lack o in-pixel ampli ica ion
penalizes bo h image quali y, because noise in he eadou pa h is no a enua ed and speed as he e is no
gain a ailable o boos ansien beha io s.
Some o he PPS´s d awbacks we e o e come by he p oposal o he i s Ac i e Pixel Senso (APS) by
Noble:
P. J. W. Noble, “Sel -Scanned Silicon Image De ec o A ays.” IEEE TRANSACTIONS ON ELECTRON
DEVICES, Vol. 15, pp. 202–209, Ap il 1968.
As he igu e inse a he igh illus a es, APSs include ac i e
de ices and ampli y igh a he pixel, hus helping a enua e
noise impac . Mode n CIS pixels a e o he APS ype, al hough
much mo e complex han Noble's ini ial p oposal.
Ea ly CISs and indeed some mode n ones we e ead in a
p ocess ha in ol ed:
1) he sequen ial selec ion o ows,
2) he sequen ial scanning o columns wi hin each
selec ed ow.
In hese ea ly de ices, all columns we e connec ed o a common line ampli ie ha p ocessed ol age
columns sequen ially. Mode n CISs cus oma ily employ pe -column ampli ie s so ha eading is semi-
pa allel, as is he case o he CISs in his Thesis.
CIS Exposu e employs ei he o he ollowing me hods:
• Rolling shu e , in ol ing he simul aneous exposi ion o all pixels in a ow and he sequen ial
exposu e o ows;
• Global shu e , whe e all pixels a e exposed simul aneously [Nakam06].
Global shu e p ecludes a i ac s caused by objec s mo ing ac oss he scene du ing exposu e. CISs concei ed
in his Thesis employ bo h exposu e ypes and, indeed, di e en global shu e ypes o educe e o s.
O e iew o come majo CIS e olu ion miles ones
Fig. 10 summa izes some majo miles ones obse ed du ing he e olu ion o CISs since Noble´s APS p oposal.
The iden i ica ion o one o ano he miles one is a ma e o opinion, and he selec ion o miles ones in his
igu e does no a emp o suppo a heo y bu o illus a e a end unde a subjec i e iew.
Al hough Fig. 10 does no explici ly highligh he in en ion o he PPD, his in en ion by NEC esea che s
and he adop ion o he so-called 4-T APSs, based on PPDs, was ins umen al o CIS de elopmen . 4T-APSs
sepa a e he nodes whe e cha ges a e accumula ed, on he one hand, and whe e hey a e ans o med in o
ol ages, on he o he hand. Bo h nodes a e elec ically connec ed by a swi ching ansis o ( ans e ga e).
This sepa a ion c ea es some challenges, bu 4T-APSs a e widely used oday. These pixels and some a ian s
a e a he co e o he CISs in his Thesis.
Chap e 1 In oduc ion o Low Noise Image Senso s
22
REPRESENTATIVE CIS EVOLUTION MILESTONES: A SUBJECTIVE VIEW
Pe iod 1968 → 2000
As Fig. 11 illus a es, signi ican miles ones du ing his pe iod included majo achie emen s ha ha e al eady
been men ioned, namely:
• The in en ion o he pinned pho odiode in 1980, and he p oposal o he so-called 4T-APS.
• The embedding o p ocessing ci cui y oge he wi h he pho o-de ec o s and he e olu ion o he
concep o came a-on-chip.
SOME MAJOR MILESTONES IN THE PERIOD 1968 → 2000
1.2 Snapsho s abou CIS E olu ion _
23
PPD was ins umen al in achie ing image quali y. The so-called 4-T pixel, illus a ed a he bo om o Fig.
11, based on PPDs, is cus oma ily adop ed o mode n CISs as i enables quali y le els ha we e p e iously
a ainable only wi h CCDs. Once he image quali y ba ie was su moun ed, CISs we e p ac ically unbea able
owing o hei ad an ages o combine de ec ing and p ocessing ci cui y in o a single de ice, hus pa ing he
way o ue sensing-p ocessing sys em-on-chip implemen a ion. Todays, CCDs a e ba ely used o some
specialized, niche applica ions pa icula ly in as onomy.
Pe iod 2000 → 2007
Besides many o he concu en ac i i ies du ing his pe iod, sys em in eg a o s g adually ansi ioned om
CCD-based came as o CIS-based ones d i en by SWaP and cos mo i a ions. Semi-pa allel eading and
encoding o he in o ma ion, h ough pe -channel eadou pa hs and ADCs we e also ma u ing du ing his
pe iod. In e es ingly enough, many came a manu ac u e s we e eluc an o adop hese solu ions owing o
he isk o p oducing image a i ac s (pa e n noise) caused by misma ches among pa allel ampli ie s and ADC
ins ances. These issues a e he e and canno be igno ed. Howe e , hey can be con on ed by adop ing
digi ally-assis ed analog design me hods, on-chip calib a ion, and e o co ec ion. Again, he po en ial o
CISs o inco po a e p ocessing ci cui y is ins umen al o hese solu ions.
SOME MAJOR MILESTONES IN THE PERIOD 2000 → 2007
Pe iod 2008 → > 2020
Doub s, i any, ega ding he wide accep ance o CISs o imaging we e emo ed du ing his pe iod, whe e
wo signi ican miles ones a e highligh ed, namely:
• The gene alized adop ion o BSI op ions ollowing imp o emen and ma u i y o wa e hinning
me hods;
• The ansi ion owa ds e ically-in eg a ed, 3D-s acked echnologies o imp o ed pixel layou and
enla ged came a on-chip unc ionali y.
As highligh ed in he igu e, he i s comme cial BSI CISs and he i s comme cial BSI-s acked CISs we e
eleased du ing his pe iod. The as majo i y o u u e CISs will likely adop hese s a egies. Indeed, CISs
employ bo h o hem in his Thesis.
Concu en ly o CIS p og ess, CVISs also expe ienced signi ican p og ess du ing his pe iod – see Fig. 14.
Pa icula ly:
Chap e 1 In oduc ion o Low Noise Image Senso s
24
• The i s comme cial ully p og ammable ision sys em-on-chip was eleased by ANAFOCUS company
be o e i s acquisi ion by Teledyne.
• SONY and OMNIVISION eleased he i s comme cial DVSs.
This Thesis does no deal wi h CVIS. Howe e , e e ing o hese miles ones is ele an because o he as
expec ed impac o his echnology in he o hcoming yea .
SOME MAJOR MILESTONES IN THE PERIOD 2008 → >2020
SOME MAJOR CVISS MILESTONES IN THE PERIOD 2008 → >2020
1.3 Noise in image senso s _
25
1.3 NOISE IN IMAGE SENSORS
A gene al, b oad de ini ion o ci cui noise would ead as ollows:
Any al e a ion o he nominal alues o a ci cui 's ol ages, cu en s, cha ges o magne ic luxes.
O cou se, such a b oad de ini ion includes e o s caused by pa asi ic componen s, non-linea i ies, leakages,
and he like ha a e no echnically noise, al hough in he con ex o CISs, some imes we e e o hem as
noise as hey p oduce a i ac s in he cap u ed images. Fo ins ance, pe -column ADCs may ope a e a
di e en speeds due o non-uni o m eac i e pa asi ic, and hese di e en ope a ion paces may esul in
image quali y deg ada ion.
De ailed co e age o image quali y me ics and noise ypes can be ound elsewhe e [EMVA21] [K iss15]
[Janes07] [Nakam06]. Roughly speaking, noise sou ces can be classi ied in wo la ge clus e s depending o
he phenomena causing hem:
• Fixed pa e n noise;
• Tempo al noise.
Fixed Pa e n Noise
Fixed Pa e n Noise (FPN) FPN e e s o he spa ial a ia ion pixel o pixel o e he a ay which mani es
clea ly as changes in he pixel ou pu s unde uni o m illumina ion condi ions. These spa ial a ia ions can be
di e en om chip o chip, bu a e he same o all images g abbing om he same chip.
E o s esul in di e en ans e unc ions o
esponsi i ies o each pixel and di e en
ansmi ances o each channel (see Fig. 15) and
ha e a global componen equal o all pixels and
eadou channels and a andom componen due o
his misma ch o de ice beha io . The las one
gene a es FPN on he sensed image so ha he
esul ing image p esen s spa ial a ia ions unde
uni o m illumina ion. This spa ial noise emains
supe imposed on he pic u e in a ideo sequence.
Among o he easons hese e o s a e due o:
• de ice misma ches,
• non-uni o m ou ing,
• non-uni o m loading,
• misma ch in he op ical s ack,
• e c.
ILLUSTRATING PIXELS AND READOUT
CHANNELS NON-UNIFORMITY
These and o he simila phenomena a e usually cha ac e ized bo h unde illumina ion and in da k
condi ions and a e desc ibed by wo main pa ame e s:
• DSNU: Da k Signal Non-Uni o mi y. The pixel- o-pixel a ia ion in da k condi ions emo ing he
empo al componen , which is usually achie ed a e aging ce ain numbe o images o emo e he
empo al noise.
• PRNU: Pho on Response Non-Uni o mi y. I is he pixel- o-pixel a ia ion in p esence o ligh , and i is
measu ed as he gain a ia ion o he pho o esponse om pixel o pixel.
Chap e 1 In oduc ion o Low Noise Image Senso s
32
Then, he pho on sho noise esponds o his kind o p ocess, when in a e age a numbe λ o pho ons
a i e o he pixel du ing in e al [0, ], i ollows a Poisson dis ibu ion, whe e he p obabili y o ecei ing n
pho ons in he in e al [0, ] is gi en by
𝑝𝑛(𝑡)=(𝜆∙𝑡)𝑛
𝑛! ∙𝑒−𝜆∙𝑡
(1.2)
An s a is ical a iable ollowing a Poisson law o p obabili y is cha ac e ized by he ac ha he a iance
o he s a is ical p ocess is equal o he a e age o said p ocess. Then, in an image senso which is in eg a ing
a pho on lux du ing he ime Tin wi h an a e age numbe o pho ons N he a iance is gi en by
𝑉𝑎𝑟[𝑠ℎ𝑜𝑡 𝑁𝑜𝑖𝑠𝑒 𝑁]=𝑁
(1.3)
The a iance e sus he a e age ou pu signal is called he Pho on T ans e Cu e (PTC) [Janes07]. F om
he PTC he con e sion gain, CG, om he collec ed elec ons o he image senso ou pu can be ob ained
measu ing he slope o his cu e in he linea pa . Addi ionally, he poin he PTC cu e s a s decaying is
conside ed he sa u a ion capaci y [EMVA21], he sa u a ion can come om he pho odiode, he pixel sense
node o he eadou channel, ypically inpu ange o
eadou channel is adjus ed o he pixel ou pu ange
o maximize dynamic ange. Fig. 24 shows one
example o PTC o one image senso de eloped in he
scope o his hesis. The cu e shows he sho noise
domina es in he linea pa o he cu e as expec ed,
and he CG measu ed wi h his image senso ope a ing
in his mode is CG= 4.27 DN/e-, whe e DN s ands o
digi al numbe since he ou pu is deli e ed in digi al
wo ds o 16 bi s in his example.
PHOTON TRANSFER CURVE
1.3.4 Da k Cu en noise
Da k cu en is he pa ame e used o quan i y he amoun o cha ge gene a ed and accumula ed in he pixel
pho odiode in he absence o ligh . This phenomenon is in insic o he semiconduc o de ice, any de ec in
he semiconduc o s uc u e can c ea e an ene ge ic s ep in he band gap o he semiconduc o , ha allows
a he mal gene a ion o ecombina ion o he ee mino i y ca ie s. This e ec has been modelled in he
Shockley-Read-Hall (SRH) equa ion [Shock52]:
𝑈𝑆𝐻𝑅=𝜎𝑝𝜎𝑛𝑣𝑡ℎ𝑁𝑇(𝑛𝑝−𝑛𝑖2)2
𝜎𝑛(𝑛+𝑛𝑖𝑒𝐸𝑡−𝐸𝑖
𝑘𝑇 )+𝜎𝑛(𝑝+𝑝𝑖𝑒−𝐸𝑡−𝐸𝑖
𝑘𝑇 )
(1.4)
whe e USHR is he ne ca ie gene a ion/ ecombina ion, NT is he concen a ion o de ec a ene gy le el E ,
Ei is he in insic Fe mi le el, 𝑣𝑡ℎ he he mal eloci y o he ca ie s, 𝜎𝑛 and 𝜎𝑝 a e he elec on and hole
c oss sec ion, and n and p a e he numbe o ee elec ons and holes.
The equa ion (1.4) shows ha a ew main condi ions d i e he da k cu en in he pixel. Fi s , a he mal
de-equilib ium mus occu o gene a e a cu en . In da k condi ions, a e he pho odiode ese , he deple ed
a ea i a diode is a ypica case o non-equilib ium, because we ha e n=p=0, since he pho odiode is deple ed.
Nex , he cu en will be linea wi h he de ec densi y NT. The ene gy le el o de ec E also de e mine he
de ec ac i i y: he close o he in insic Fe mi ene gy le el Ei, he highe he p obabili y o ecombina ion
1.3 Noise in image senso s _
33
be ween elec ons om he conduc ion band and holes om he alance band. Finally, he Fe mi le el Ei will
de e mine he n and p densi y and ix he illing p obabili y o hese de ec s.
Then, he da k cu en gene a ion is dependen on he p ocess, he highe numbe o de ec s he highe
da k cu en is gene a ed [Ca e14]. In ac , he pinned pho odiode (PPD) in oduced an impo an
imp o emen in he da k cu en hanks o he educ ion o de ec s on he pho odiode su ace [Te an14] ,
becoming a p ima y echnology elemen in CMOS image senso . New s a egies a e de eloped in o de o
educe he numbe o de ec s su ounding he pho odiode, as educing he shallow enches isola ion egion
a ound he pho odiode [Te an14] .
SRH model has been used o iden i y he main oo cause o he dominan con ibu o s o he da k
cu en o he momen du ing he las decades. This has success ully been used o educe me allic
con amina ion in he p ocess h ough imp o ed cleanliness and ge e ing. I has elimina ed he gene a ion
o he in e aces by passi a ion and by p ocess and obus design [Pa k10][Theuw03]. These imp o emen s
pe mi educing he da k cu en and change he na u e o he main con ibu o s o he empe a u e
dependence, and consequence he doubling coe icien a ies om ypical alues o 11⁰C o 6 ⁰C o example.
Fig. 25 p o ides a common c oss-sec ion o pixel pho odiode wi h he possible loca ions whe e he di e en
mechanism and oo causes o da k cu en gene a ion appea :
3) TX edge: di usion om in e ace (Eg ); lucky d i om in e ace (Eg /2); in e ace s ess (Eg );
4) FD injec ion: blooming o wa d bias;
5) O he junc ion: o wa d bias o ca ie injec ion;
6) STI in e ace: di usion om in e ace (Eg ); in e ace s ess (Eg );
7) Pinning laye : di usion om in e ace (Eg ); in e ace s ess (Eg ); doping-based SRH (Eg );
8) Deep deple ion: SRH om con amina ion (Eg /2);
9) Ligh ly doped egion: di usion om weak SRH (Eg /2 o Eg);
10) Back in e ace: di usion om weak in e ace SRH (Eg );
11) Con ac subs a e cu en (Eg)
DARK CURRENT MECHANISMS AND ROOT CAUSES [MCGRA17]
Chap e 1 In oduc ion o Low Noise Image Senso s
34
The da k cu en is he pa ame e limi ing he in eg a ion ime o any image senso , since he da k cu en
has an associa ed noise which ollows same s a is ical beha io as sho noise, he da k cu en noise s an da
de ia ion is squa e oo o he da k cu en a e age:
𝜎𝑑𝑎𝑟𝑘 𝑐𝑢𝑟𝑟𝑒𝑛𝑡=√𝜇𝑑𝑎𝑟𝑘 𝑐𝑢𝑟𝑟𝑒𝑛𝑡=√𝐼𝑑𝑎𝑟𝑘 𝑐𝑢𝑟𝑟𝑒𝑛𝑡∙𝑡𝑖𝑛𝑡
(1.5)
Then, o long in eg a ion ime low da k cu en a e equi ed along wi h colling sys ems in o de o educe
he noise in oduced by his phenomenon.
1.3.5 T ans e Noise
T ans e noise migh be o en a limi ing ac o in he pe o mance o low noise image senso s when he a ge
ead noise is e y low, bellow 1 e- RMS. In o de o op imize and educe he cha ge ans e noise,
unde s anding he sou ces and how hey can be educed is needed o he design o low noise image senso s.
T ans e noise has been analyzed ex ensi ely in CCDs [Janes01] [Omu a80][Omu a80]. These wo ks
ob ained a model o he cha ge noise gene a ed by su ace apping and de apping while signal cha ge is
mo ed om he pho oga e pixel o he ou pu ampli ie . This model is alid o CCDs wi h a la ge numbe o
cha ge ans e s ages and a la ge numbe o aps [Fossu03]. A mo e p ecise model is de i ed o he case
o 4T CMOS image senso [Fowle07].
In a ypical 4T CMOS image senso pixel, he e a e wo main causes o ans e noise, he i s is he
incomple e cha ge ans e be ween he pinned pho odiode and he loa ing sense node, and he second is
cha ge apping in he in e ace pho odiode and ans e ga e, Si- SiO2 in e ace. Incomple e cha ge
ans e esul s in a ia ions in he he mionic emission and quan um mechanical unneling om a po en ial
ba ie o a po en ial pocke be ween he pinned pho odiode and he loa ing di usion node, i has been
also analyzed wi h he lag, since i e e s o he incomple e cha ge ans e ence om he pho odiode o he
sense node.
In o de o es ima e he con ibu ion om incomple e cha ge ans e sou ce in 4T CMOS pixels, he
a e age ime equi ed o emp y he cha ge om small po en ial ba ie o small po en ial pocke in he
pho odiode is analyzed esul ing ha his ime is less han 1 nanosecond conside ing he he mionic emission
is he dominan sou ce o elec on anspo o a ypical p ocess. This ime is much as e han ypical cha ge
ans e ha should be la ge han hund eds o nanoseconds. The e o e, incomple e cha ge ans e is no a
signi ican sou ce o ans e noise, as i was in CCDs.
Cha ge apping and de apping a he Si-SiO2 in e ace is he dominan sou ce o cha ge ans e noise in
4T CMOS image senso s. In addi ion, while he ans e ga e ol age is high elec ons om he loa ing
di usion can in e ac wi h he su ace aps as well in he ans e channel. Bo h mechanisms enable cha ge
apping and o de apping du ing he cha ge ans e . A la ge posi i e ol age can be applied o he ans e
ga e in o de o c ea e ol age di e ence and accele a e he di usion o he accumula ed cha ges o he
sense node. Bu his posi i e ol age can a ac he elec ons o he Si-SiO2 in e ace allowing some elec ons
o be apped du ing he ans e p ocess. Unlike he noise in oduced by he incomple e cha ge ans e , by
lag, he a iance o his luc ua ion is no easy o measu e and p ac ical op imiza ion is needed wi h ega ds
he high ol age o he ans e signal in o de o op imize he ans e noise ( see sec ion 562.2.1 ).
1.3.6 Rese noise
E e y ime a ol age is sampled in a capaci o , no only he nominal ol age le el in he capaci o bu also he
noise coming om his ol age is in eg a ed in he capaci o . KT/C noise s ands o he noise s o ed in a
capaci o when i is cha ged h ough a esis o . The he mal noise appea s in any conduc i e ma e ial when
he absolu e empe a u e is highe han 0°K, as desc ibed p e iously.
When a capaci o is cha ged by a esis o he noise calcula ion can be done using an ideal model o esis o
in se ies wi h he equi alen noise ol age sou ces.
1.3 Noise in image senso s _
35
RC NOISE MODEL
The o al powe o Vo( ) is gi en by he in eg a ed spec al densi y o powe mul iply by squa ed module
o ans e unc ion seen by Vo( ) signal as shown in equa ion equa ion (1.6) ende ing in a o al ou pu
powe equal o kT/C.
𝑃𝑜𝑢𝑡=∫ 𝑆𝑉(𝑓)∙|𝐻(𝑓)|2𝑑𝑓
∞
0=∫ 4𝑘𝑇𝑅∙1
(2𝜋𝑅𝐶)2+1𝑑𝑓
∞
0
(1.6)
𝑃𝑜𝑢𝑡=𝑘𝑇
𝐶
(1.7)
Same esul is eached using he equi alen noise bandwid h, Bn, as he band wid h o an ideal low pass il e
which p oduces he same in eg a ed noise powe as he ac ual il e o sys em ans e unc ion. In he case
o one pole sys em, he equi alen noise bandwid h is equal o π/2 imes he pole equency.
𝑃𝑜𝑢𝑡=𝑣𝑛2
∙𝐵𝑛=4𝐾𝑇𝑅∙ 1
2𝜋𝑅𝐶 ∙ 𝜋
2=𝑘𝑇
𝐶
(1.8)
When MOS swi ch sample a da a in a capaci o he s o ed ol age alue, he swi ch goes om a low alue
esis o , Ron, o a e y high alue, Ro , ozen he ol age le el in he capaci o . The ol age noise sampled in
he capaci o is equal o he calcula ed in equa ion (1.7), being he ol age noise equal o √𝑘𝑇
𝐶.
SWITCH CAPACITORS NOISE MODEL
I is he noise s o ed in he sense node e e y ime he ese ansis o is ac i a ed o se he ese ol age
in he loa ing sense node. The ese noise can be educed using so ese , ha consis s o ope a ing he
ese ansis o in sub h eshold egion [Te an16]. To pe o m a p ope ese o he loa ing di usion node
he so ese ope a ion is p eceded by a ha d ese ope a ion whe e he ese ansis o is wo king in ohmic
egion o se he ese ol age independen ly on he ini ial alue p io o he ese ope a ion, his ope a ion
is known as ha d-so ese as depic ed in Fig. 28.
Chap e 1 In oduc ion o Low Noise Image Senso s
36
RESET OPERATION (A) HARD RESET (B) SOFT RESET (C) HARD-SOFT RESET
In 4T-APS CISs wo king in olling shu e ope a es ollowing he nex sequence:
• i s he ese o he loa ing di usion, secondly ese le el is ead;
• hen he cha ge is ans e ed o he loa ing di usion; and
• inally he signal le el is ead.
To ge id o he ese noise he co ela ed double sampling ope a ion is pe o med sub ac ing he signal
le el om he ese le el, emo ing o ally he ese noise since he cha ge in eg a ion is done o e he
unique sampled ese (see sec ion 1.4.3 ). Howe e , 3T CMOS image senso s used in la ge pixels wi h la ge
ull well capaci y a e subjec ed o he ese noise and analysis and op imiza ion is needed. The ese noise
becomes ou o he scope o his Thesis as i ocuses low-noise 4T-APS CISs.
1.4 READOUT ARCHITECTURES IN IMAGE SENSORS
The pixel signal should be adap ed and digi alized h ough he eadou channel. The eadou channel could
be composed by only one ADC con e ing he pixel ese le el and signal le el o pe o m he co ela ed
double sampling in digi al domain, which is named Digi al Double Sampling (DDS). Ano he op ion, is using a
CDS analog s age which can be sampling he pixel ou pu and sub ac ion he ese and signal le el in he
analog domain p io o he digi aliza ion. Addi ionally, a p og ammable gain ampli ie (PGA) can be inse ed
o p o ide analog gain o he pixel ou pu signal.
In his sec ion, we will conside he main de ice noise sou ces and he con ibu ion in he wo main
p ocess in he pixel ou pu signal condi ioning and con e sion: sampling p ocess and compa a o s. A e ha ,
column ampli ica ion is p esen ed, and inally, he co ela ed double sampling ope a ion employed in image
senso s is p esen ed. The ADC a chi ec u e de eloped in his Thesis is desc ibed in de ail in Chap e 3.
1.4.1 Common noise sou ce in ADCs
1.4.1.1. NOISE IN SAMPLING CAPACITORS
Swi ched capaci o s ci cui s a e commonly used in image senso s ADCs in o de o sampling he da a coming
om he pixels and digi alize he alue a e CDS ope a ion. I is impo an o ecognize he noise s o ed in
he capaci o s when a ol age is sampled in i . When he e is only a MOS swi ch sampling he signal, he noise
is gi en by equa ion (1.8) desc ibed in Sec ion 1.3.6 . Howe e when an ac i e elemen like an Ope a ional
T ansconduc ance Ampli ie (OTA) is p esen in he signal sampling, he noise con ibu ions and he
equi alen noise bandwid h change, esul ing in a di e en o mula o his sampled noise.
Fo exampling, in he case o he single edge common sou ce ampli ie showed in Fig. 29 he MOS
swi ches a e designed wi h low enough equi alen esis ance, in o de o ensu e ha he sys em bandwid h
is gi en by he pole coming om he ampli ie and wi h noise spec al densi y is much lowe han he said
ampli ie . In his case, when he MOS swi ch passes o “OFF” s a e, he signal is ozen in he capaci o . The
noise ol age pe uni equency a he ou pu is gi en by equa ion (1.9). To be e e ed a he inpu o he
OTA he noise is di ided by he gain ende ing in he esul gi en in equa ion (1.10). When his OTA is used
Readou Rese Readou Rese Readou Rese
SEL
RST
SEL
RST
SEL
RST
(a) (b) (c)
1.4 Readou a chi ec u es in image senso s _
37
in a swi ched capaci o ci cui s he inpu e e ed noise should be mul iplied by he equi alen bandwid h
noise when he swi ch and he capaci o a e connec ed o he OTA, in his case he uni gain bandwid h (GB)
is 𝑔𝑚𝑎𝑚𝑝𝑛
𝐶1 ad/s i he pole in oduced by he swi ch esis ance is a away om he dominan pole, and i
has been sized o ul ill his condi ion. The equi alen noise bandwid h Bn=GB·π/2= GB·π/2, being he o al
in eg a ed noise in he capaci o gi en by he exp ession (1.12). The o al noise is a alue named alpha equal
o 𝛾(1+ 𝑔𝑚𝑏𝑝
𝑔𝑚𝑎𝑚𝑝𝑛), which is dependen on
he echnology h ough he pa ame e 𝛾
and on he design by he ac o 𝑔𝑚𝑏𝑝
𝑔𝑚𝑎𝑚𝑝𝑛.
SC AMPLIFIER EMBODIMENT
𝑣𝑛,𝑜𝑢𝑡
2
=4𝐾𝑇𝛾(𝑔𝑚𝑎𝑚𝑝𝑛+𝑔𝑚𝑏𝑝)
(𝑔𝑑𝑠𝑎𝑚𝑝𝑛+𝑔𝑑𝑠𝑏𝑝)2
(1.9)
𝑣𝑛,𝑖𝑛
2
=𝑣𝑛,𝑜𝑢𝑡
2
𝐺𝑂𝑇𝐴2=4𝐾𝑇𝛾 (𝑔𝑚𝑎𝑚𝑝𝑛+𝑔𝑚𝑏𝑝)
(𝑔𝑑𝑠𝑎𝑚𝑝𝑛+𝑔𝑑𝑠𝑏𝑝)2∙(𝑔𝑑𝑠𝑎𝑚𝑝𝑛+𝑔𝑑𝑠𝑏𝑝)2
𝑔𝑚𝑎𝑚𝑝𝑛2
𝑣𝑛,𝑖𝑛
2
=4𝐾𝑇𝛾 1
𝑔𝑚𝑎𝑚𝑝𝑛(1+ 𝑔𝑚𝑏𝑝
𝑔𝑚𝑎𝑚𝑝𝑛)
(1.10)
𝑉𝑛,𝑜𝑢𝑡,𝑡𝑜𝑡
2
=4𝐾𝑇𝛾 1
𝑔𝑚𝑎𝑚𝑝𝑛(1+ 𝑔𝑚𝑏𝑝
𝑔𝑚𝑎𝑚𝑝𝑛)∙𝑔𝑚𝑎𝑚𝑝𝑛
2𝜋∙𝐶1∙𝜋
2
(1.11)
𝑉𝑛,𝑜𝑢𝑡,𝑡𝑜𝑡
2
=𝛾(1+ 𝑔𝑚𝑏𝑝
𝑔𝑚𝑎𝑚𝑝𝑛)𝐾𝑇
𝐶1
(1.12)
In summa y, he powe noise in sampling capaci o s is a ac o o he kT/C noise. The la ge numbe o
ansis o s composing he OTA o he sys em, he ac o o noise can be la ge . Addi ionally, gene ally
speaking, in o de o op imize noise he ampli ica ion ansis o s should maximize he ansconduc ance,
while he load ansis o s should minimize he ansconduc ance by inc easing he o e d i e ol age (VGS -
VT) as much as possible.
1.4.1.2. NOISE IN COMPARATORS
Many analog o digi al a chi ec u es use compa a o s o digi alized he analog wo d. In pa icula , single slope
ADC is a well-known a chi ec u e whe e he noise o compa a o s has an impo an con ibu ion o he o al
noise. A de ailed noise analysis o compa a o is desc ibed in [Sepke09]. Common equency domain noise
analysis assumes ha he ampli ie is in s eady s a e, bu compa a o s do no necessa ily each s eady s a e
and add noise du ing hei ansien s. The e o e, he usual assump ion o wide-sense-s a iona y noise sou ce
is no always alid. The nons a iona y noise analysis o a i s o de ansconduc ance ampli ie when d i en
by an inpu amp (Fig. 30), used in single-slope ADC, is desc ibed in his sec ion.
Chap e 1 In oduc ion o Low Noise Image Senso s
38
INPUT RAMP IN FIRST ORDER TRANSCONDUCTANCE AMPLIFIER COMPARATOR
The s ep amp inpu and he impulse esponse o he ansconduc ance ampli ie a e gi en by:
𝑣𝐼𝐷=(𝑑𝑣𝑥
𝑑𝑡)∙𝑡∙𝑢(𝑡)
(1.13)
ℎ(𝑡)=𝐺𝑚
𝐶𝐿∙𝑒−𝑡/𝜏𝑜∙𝑢(𝑡)
(1.14)
whe e he las one is coming om he
equi alen ci cui shown in Fig. 31.
LINEAR SMALL SIGNAL
TRANSCONDUCTANCE AMPLIFIER EQUIVALENT
CIRCUIT
The esponse o he ampli ie o he s ep amp inpu is:
𝑣𝑂𝐷(𝑡)=𝐴0𝑀𝑠𝑙𝑜𝑝𝑒[𝑡−𝜏0(1−𝑒−𝑡 𝜏0
⁄)]𝑢(𝑡)
(1.15)
whe e 𝐴0=𝐺𝑚𝑅0 is he dc gain, 𝜏0=𝑅0𝐶𝐿 is he ou pu ime cons an and 𝑀𝑠𝑙𝑜𝑝𝑒=𝑑𝑣𝑥/𝑑𝑡 is he inpu
amp slope. The e o e, he slope o he ou pu o he ampli ie is:
𝑑𝑣𝑂𝐷
𝑑𝑡 (𝑡)=𝐴0𝑀𝑠𝑙𝑜𝑝𝑒[1−𝑒−𝑡 𝜏0
⁄]𝑢(𝑡)
(1.16)
I we conside ha 𝑡≪𝜏0, hen he slope simpli ies o:
𝑑𝑣𝑂𝐷
𝑑𝑡 (𝑡)=𝐴0𝑀𝑠𝑙𝑜𝑝𝑒 𝑡
𝜏0𝑢(𝑡)=𝐺𝑚
𝐶𝐿×𝑀𝑠𝑙𝑜𝑝𝑒×𝑡=𝐺𝐵×𝑀𝑠𝑙𝑜𝑝𝑒×𝑡
(1.17)
The impulse esponse om he noise cu en sou ce o he ou pu ol age is
ℎ𝑛(𝑡)=1
𝐶𝐿𝑒−𝑡 𝜏0
⁄𝑢(𝑡)
(1.18)
being 𝑆𝑥0 he whi e noise PSD o he noise p ocess and he impulse esponse ℎ𝑛(𝑓) in equency, om he
noise sou ce o he ou pu . The ou pu a iance simpli ies o
𝜎𝑦2(𝑡)=𝑆𝑥0∫|ℎ𝑛(𝑓)|2𝑑𝑓
+∞
−∞
(1.19)
Vx( )
Gm
+
-
Vx
Vin CL
1.4 Readou a chi ec u es in image senso s _
39
Modeling he noise cu en PSD o he ansconduc ance ampli ie as 𝑆𝑥0=4𝑘𝑇𝐺𝑛 as he he mal noise
om an equi alen noise conduc ance 𝐺𝑛, and subs i u ing his alue in (1.19), he noise a he ou pu o he
ampli ie is:
𝑣𝑜𝑛
2
(𝑡)=(𝐺𝑛
𝐺𝑚)𝑘𝑇
𝐶𝐿(𝐺𝑚𝑅𝑜)[1−𝑒−2𝑡 𝜏0
⁄]𝑢(𝑡)
(1.20)
I we conside ha 𝑡≪𝜏02
⁄, hen he noise simpli ies o:
𝑣𝑜𝑛
2
(𝑡)=2𝑘𝑇𝐺𝑛
𝐶𝐿2𝑡
(1.21)
The ji e a he ou pu o he ampli ie can be exp essed as he ou pu noise di ided by he squa e o he
ou pu slope a he momen o he ansi ion, 𝑣𝑂𝐷=𝑣𝑀𝑜, hen
𝜎𝑡𝑖
2(𝑡)=𝑣𝑜𝑛
2
(𝑡)|𝑑𝑣𝑂𝐷
𝑑𝑡 (𝑡)|−2𝑣𝑂𝐷=𝑣𝑀𝑜
(1.22)
By using o mula o he noise and slope a ou pu o ampli ie , we ge he a iance o he ji e :
𝜎𝑡𝑖
2(𝑡)=2𝑘𝑇(𝐺𝑛
𝐺𝑚)(1
𝐺𝑚×𝑡×𝑀𝑠𝑙𝑜𝑝𝑒2)
(1.23)
The esponse o he ampli ie o he s ep amp inpu can be app oxima ed in second o de as:
𝑣𝑂𝐷(𝑡)≈𝐴0𝑀[𝑡−𝜏0(1−(1−𝑡 𝜏0
⁄+(𝑡 𝜏0
⁄ )2
2))]𝑢(𝑡)
(1.24)
𝑣𝑂𝐷(𝑡)≈𝐴0𝑀𝑠𝑙𝑜𝑝𝑒[𝑡2
2𝜏0]𝑢(𝑡)
(1.25)
We a e in e es ed o he ji e a a pa icula poin o he ansien esponse. Le ’s call 𝑉𝑄 he h eshold
ol age o he ga e d i en by he ampli ie and 𝑡𝑄 he ime when he ampli ie eaches such h eshold 𝑉𝑄.
𝑉𝑄=𝑣𝑂𝐷(𝑡𝑄)≈𝐴0𝑀𝑠𝑙𝑜𝑝𝑒[𝑡𝑄2
2𝜏0]
(1.26)
𝑡𝑄=√𝑉𝑄×2
𝐺𝐵×𝑀𝑠𝑙𝑜𝑝𝑒=√𝑉𝑄×2×𝐶𝐿
𝐺𝑚×𝑀𝑠𝑙𝑜𝑝𝑒
(1.27)
Then he a iance o he ji e a poin (𝑡𝑄,𝑉𝑄) is:
𝜎𝑡𝑖
2(𝑡𝑄)=√2𝑘𝑇(𝐺𝑛
𝐺𝑚)( 1
(𝑉𝑄)12
⁄×(𝐺𝑚)12
⁄×(𝐶𝐿)12
⁄×(𝑀𝑠𝑙𝑜𝑝𝑒)32
⁄)
(1.28)
F om he exp ession o he s anda d de ia ion o ji e 𝜎𝑡𝑖(𝑡𝑄), i can be concluded ha :
Chap e 1 In oduc ion o Low Noise Image Senso s
40
• 𝑀𝑠𝑙𝑜𝑝𝑒 has a bigge impac on ji e ha any o he pa ame e s because o he powe 3/2.
• 𝐺𝑚 and 𝐶𝐿 ha e he same impac on educing he ji e , bu inc easing 𝐶𝐿 also inc eases he delay
𝑡𝑄.
• 𝐺𝑛
𝐺𝑚 can be minimized by using an ampli ie wi h he minimum numbe o ansis o s con ibu ing o
he noise. The e o e, he ampli ying ansis o should ha e i s 𝑔𝑚 maximized while he 𝑔𝑚 o he
load/cu en sou ce should be minimized.
• 𝑉𝑄 should be also maximized.
This analysis p o ides wi h a i s es ima ion o he noise coming om he compa a o s. In o de o ob ain
a p ecise alue ansien noise simula ions a e needed conside ing all he noise con ibu ions in he
con e sion: he noise in he analog amp and he noise in he compa a o .
1.4.2 Column ampli ica ion
The column ampli ie is a commonly used and impo an block in he eadou channel. I is he i s block
a e he pixel ou pu in he eadou chain, i is loca ed be ween he pixel and he analog- o- digi al
con e sion. The e o e, by applying analog gain a he e y beginning o he eadou chain, all he noise
in oduced by he blocks connec ed behind he column ampli ie is a enua ed by he analog gain [Schan00].
Addi ionally, he column le el ampli ie can be used o con ol and limi he bandwid h educing he he mal
noise in oduced by he pixel sou ce ollowe and pixel cu en sou ce [K yms03].
Fig. 32 shows wo examples o column ampli ie block based on swi ch capaci o ampli ie . The gain is
gi en by he a io be ween capaci o s C1/ C2, i ende s in wo gains:
• High Gain: GHG= C1/C2HG, whe e C2HG is lowe alue han C1. In ac , C1 and C2 used o be composed o
uni a y elemen s, ha a e combined o achie e he a ge gain.
• Low Gain: GLG= C1/C2LG, in his case C2LG is selec ed o achie e he uni a y gain o o adap he pixel
ou pu ange o he maximum ADC inpu ange.
(a) (b)
COLUMN GAIN AMPLIFIER (A) PROGRAMMABLE GAIN AMPLIFIER (B) DUAL GAIN AMPLIFIER
The ampli ie is ese in e e y eadou ac i a ing he au o-ze o swi ch. I has an impo an e ec o
educing he low equency noise con ibu ion o he ampli ie ci cui y[Enz96].
Howe e , using high analog gain has he d awback o limi ing he maximum signal coming om he pixel
which is ansla ed in o a educ ion o he maximum numbe o elec ons ha can be inally con e ed, and
hen a educ ion o he dynamic ange. Fo his eason, using he combina ion o low gain and high gain alue
1.4 Readou a chi ec u es in image senso s _
41
o he pixel ou pu is used o minimize noise (high gain) and maximize he sa u a ion capaci y (low gain), 0a.
I means wo con e sions a e needed, which equi es ex ending he con e sion ime, and hen educing he
ame a e. To o e come his issue, wo con e sions o high and low gain can be done in pa allel, 0b,
conse ing he con e sion ime bu equi ing la ge a ea and powe consump ion.
Adap a i e column gain ampli ie can be used o educe he con e sion ime using only one
p og ammable gain ampli ie [Sakak05]. This solu ion uses an addi ional compa a o o check i he pixel
ou pu o e pass ce ain h eshold le el o he signal o selec he op imum gain and apply his gain o he
pixel signal.
A digi al pos p ocessing is needed o combine he low gain and high gain componen s o ob ain he ou pu
wo d wi h op imum noise and dynamic ange.
1.4.3 Co ela ed double sampling
The Co ela ed Double Sampling (CDS) ope a ion is a common echnique used in CMOS image senso o
educe spa ial (FPN) and empo al noise coming om he pixel. The CDS ope a ion is depic ed in Fig. 33, i
consis s o sampling he pixel ou pu wice, he i s sampling is done a e ese ing he loa ing di usion
(FD) wi hou any cha ge coming om he pho odiode, i is called ese le el (VRST), he seco nd sampling is
done a e ans e ing all he cha ge om he pho odiode o he loa ing di usion, i is called signal le el
(VSIG). The ime be ween he wo samples is call CDS ime, To. A e ha , he signal le el is sub ac ed om
he ese le el educing he noise con ibu ions as ollows:
• Spa ial Noise (FPN): The misma ch a ia ion in pixel ansis o s like he h eshold ol age in he sou ce
ollowe ansis o gene a es ol age a ia ion in he o de o cen s o mV o s anda d de ia ion a
pixel ou pu . Wi h he CDS ope a ion, his ol age misma ch is cancelled a i s o de elimina ing he
dependence om pixel ansis o pa ame e s.
• Tempo al Noise: The CDS
ope a ion is simila o he
au oze o ope a ion, hen
he e is a educ ion o he low
equency noise componen s
ha is mo e e icien when To
is sho e . A de ailed analysis is
done in sec ion 2.3.1 .
PIXEL CDS OPERATION
WAVEFORMS
Fig. 34 illus a es he eadou column CDS ci cui y. The CDS ope a ion can be done ei he in he digi al
domain, Fig. 34a, o in he analog domain, Fig. 34b. When he CDS ope a ion is done in he digi al domain,
he ese le el is sampled in one capaci o , CR, he signal le el is sampled in ano he capaci o , CS, he wo
analog le els a e con e ed and he sub ac ion is done in he digi al domain, also called Digi al Double
Sampling (DDS). The e a e mul iples a chi ec u es o pe o m his ope a ion [Kawah18][Kawah18], i can be
done using wo ADCs in pa allel o he same ADC in a sequen ial mode, i is he case o he well know single
slope ADC wi h up/down coun e [Wei20]. When he CDS ope a ion is done in he analog doming an speci ic
block can p ecede he ADC, his block is a swi ched-capaci o ci cui in cha ge o doing he sub ac ion in he
analog domain. Fo he example p o ided in Fig. 34b he ese le el, VRST, is sampled in C1 and he e e ence
ol age is sampled in he C2 capaci o as SHR sampling phase is ac i e. A e ha , SHS phase becomes ac i e
and he signal le el, VSIG, is now sampled in C1 ans e ing cha ge om C1 o C2 ende ing he ou pu o CDS
block as Vo=V e -C1/C2(VSIG-VRST). The analog CDS can be implemen ed as well embedding he analog
Chap e 2 Pixel Noise Op imiza ion
48
PIXEL OPERATION WAVEFORMS
2.1.2 Tes ehicle chip a chi ec u e
A chip es ehicle has been used o measu e he pe o mance o he di e en pixel a ian s. This es chip
has been designed o ea u e low-noise eadou , namely, i has a eadou noise o 112 V ms ha allows
ob aining low noise pixels measu emen s. Addi ionally, i allows mul isampling es ing when one ow is
selec ed.
The es chip includes 8 a ian s o e e y ype o pixel 4T/5T and one o wo supplies, namely:
• 4T ype s uc u e,
• 4T ype s uc u e 2 supplies,
• 5T ype s uc u e,
• 5T ype s uc u e 2 supplies.
Pixel a ay dis ibu ion in he es chip
Fig. 39 shows he pixel a ay dis ibu ion and he suba ays de ini ion.
• The e a e 4 main columns o sub a ays o :
• 4T pixel one supply,
• 4T pixel wo supplies,
• 5T pixel one supply and
• 5T pixel wo supplies.
• E e y column consis s o 8 suba ays.
• Fu he mo e, he e a e wo addi ional columns on he le and on he igh ha a e simila o he 4T
pixel one supply bu wi h he me al co e ing in o de o check he op ical black co ec ions.
2.1 Tes s uc u es pixel a ian s and cha ac e iza ion esul s _
49
Fig. 40 shows an image ead ou om he senso chip. No e ha he image includes di e en egions,
each one coming om a co esponding suba ay.
PIXEL ARRAY DISTRIBUTION
REAL IMAGE FROM THE TEST CHIP
Chap e 2 Pixel Noise Op imiza ion
50
2.1.3 Cha ac e iza ion esul s
All sub-a ays desc ibed in p e ious sec ions ha e been cha ac e ized ega ding noise pe o mance o he
di e en supply combina ions. As a p elimina y s ep, he eadou channel noise con ibu ed by he ADC has
been measu ed o an inpu ull scale o 1.5 V. En ies labelled Channel Noise in he Tables below include his
me ic. This channel noise powe mus be emo ed om he o al noise o ob ain he pixel noise con ibu ion.
One ele an conside a ion ega ding da k empo al noise in image senso s conce n he quan i a i e
me ic employed. Fig. 41 shows a gene ic da k noise dis ibu ion om a ypical low-noise image senso
de ice. Depending o me ic used o p o ide he inal alue: median, mean o Roo Mean Squa ed (RMS), he
inal alue can change a lo . In his Thesis, all he epo ed alues a e gi en in RMS since i is a mo e ealis ic
pa ame e o quali y al hough despi e p o iding mo e pessimis ic es ima es as Fig. 41 illus a es.
READOUT NOISE DISTRIBUTION AND DARK NOISE DEFINITIONS
2.1.3.1. 4T PIXEL WITH 1METAL_1EXT OPTION
Table 2 p o ides he pe o mances o he 4T pixel a ian wi h one supply and connec ed o ex e nal ol age
supply. Acco ding o he measu emen esul s:
he pixel e sion 2, wi h a empo al noise o 1.0 e− is he bes pe o ming ega ding noise.
I is wo h men ioning ha his e sion has he la ges sou ce ollowe ga e: W=0.8 m/L=0.8 m.
As a coun e pa , his pixel e sion shows signi ican non-linea i y a he low-pa o in eg a ed signal
egion o he esponsi i y cu e − see Fig. 42 (a), whe e images a e ob ained wi h cons an illumina ion and
sweeping exposu e ime. Indeed, his e sion has smalle esponsi i y han he o he wi hin ha egion and
can hence p o ide wo se images unde hese condi ions despi e ha ing lowe empo al noise. The Fig. 42 (b)
shows he da k noise dis ibu ion ha co esponds wi h he pixel noise his og am [Woo99] o e e y pixel
a ay.
Table 2. 4T PIXEL PERFORMANCES FOR THE 1METAL_1EXT OPTION
0 1 2 3 4 5 6 7 Uni s
Mic olens
1 1 1 1 1 1 1 1 -
H-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
V-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
FSR 1.43 1.38 1.37 1.46 1.40 1.25 1.43 1.29 V
DR 79.6 79.4 82.2 79.8 77.1 75.9 79.8 79.0 dB
SNRmax 40.64 40.61 40.88 41.02 40.99 40.36 40.75 40.28 dB
CG 125.72 121.68 109.17 116.95 116.43 121.09 121.59 122.31 uV/e-
FWC 11396 11307 12573 12489 12042 10312 11782 10542 e-
PRNU 0.16 0.16 0.29 0.15 0.16 0.18 0.16 0.18 %
DSNU 0.56 0.53 0.84 0.51 0.50 0.58 0.51 0.57 e-
150 147 107 149 195 199 147 144 uV ms
1.2 1.2 1.0 1.3 1.7 1.6 1.2 1.2 e-
118.23 118.23 118.23 118.23 118.23 118.23 118.23 118.23 uV ms
0.94 0.97 1.08 1.01 1.02 0.98 0.97 0.97 e-
Linea i y E o 0.88 0.92 2.15 0.79 0.36 0.58 0.46 0.69
±%FS
Tempo al noise (da k)
Channel Noise
2.1 Tes s uc u es pixel a ian s and cha ac e iza ion esul s _
51
(a) (b)
4T PIXEL WITH 1 SUPPLY (A) RESPONSIVITY AND (B) DARK NOISE DISTRIBUTION [FOWLE15][FOWLE15]
Righ ul compa ison among pixel e sions mus ake his non-linea i y in o accoun . To ha pu pose, he
esponsi i y is linea ized, and he noise om pixels wi h la ge non-linea i y is e-calcula ed using he gain
applied o linea ize he low-exposu e egion. Fig. 43 shows he ou come o linea iza ion. The linea iza ion
p ocedu e in ol es applying a hi d o de polynomial i ing o ob ain he app oxima ed cu e and hen
linea izing using a piece-wise me hod o apply he espec i e gain. mended noise alues a e ob ained a e
mul iplica ion by he gain ob ained by di iding he slope o he linea ized esponse by he slope o he ac ual
esponse a low exposu e.
PIXEL RESPONSE LINEARIZATION
Table 3 shows he ecalcula ed noise me ics a e linea iza ion. In his able, only sui able pixel e sions
wi h low noise and good linea i y a e included. Acco ding wi h hese esul s 2 e sion migh be disca ded a
ha poin . Howe e , as will be seen in pos e io sec ion u he alida ions may be needed o inal decision.
Table 3. 4T PIXEL NOISE PERFORMANCES AFTER LINEARIZATION
0 1 2 3 4 5 6 7 Uni s
CG 139.32 130.91 129.35 132.75 130.67 uV/e-
234.81 226.96 226.74 225.59 230.42 uV ms
1.69 1.73 1.75 1.70 1.76 e-
217.23 212.95 213.38 211.62 216.72 uV ms
1.56 1.63 1.65 1.59 1.66 e-
250.69 238.17 238.41 237.62 242.17 uV ms
1.80 1.82 1.84 1.79 1.85 e-
Tempo al noise (da k)
MEAN
Tempo al noise (da k)
MEDIAN
Tempo al noise (da k)
RMS
Chap e 2 Pixel Noise Op imiza ion
52
2.1.3.2. 4T PIXEL WITH 2METAL_1EXT OPTION
Table 4 p o ides pixel pe o mances o he 4T pixel a ian wi h wo supplies and connec ed o an
ex e nal ol age supply- All wo-supplies pixel implemen a ion showed highe comp ession han 1-supply
ones a low exposu e, as Fig. 45 illus a es. Hence, his enla ged non-linea i y esul s in highe noise a e
linea iza ion as highligh ed by he da a in Table 5.
Table 4. 4T PIXEL PERFORMANCES FOR THE 2METAS_1EXT OPTION
(a) (b)
4T PIXEL WITH 2 SUPPLIES (A) RESPONSIVITY AND (B) DARK NOISE DISTRIBUTION
Table 5. 4T NOISE PERFORMANCE OF THE VALID PIXEL OPTIONS AFTER LINEARIZATION
0 1 2 3 4 5 6 7 Uni s
Mic olens
1 1 1 1 1 1 1 1 -
H-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
V-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
FSR 1.46 1.42 1.22 1.45 1.39 1.24 1.46 1.33 V
DR 81.3 81.3 80.7 80.9 79.1 78.2 81.5 80.7 dB
SNRmax 40.95 41.03 40.46 41.27 41.42 40.95 41.07 40.59 dB
CG 116.36 113.80 103.64 107.56 107.10 110.12 112.57 114.34 uV/e-
FWC 12556 12474 11792 13468 12977 11264 13014 11599 e-
PRNU 0.31 0.26 0.37 0.21 0.23 0.26 0.26 0.33 %
DSNU 0.85 0.73 1.02 0.65 0.67 0.78 0.75 0.97 e-
126 122 113 130 153 153 123 123 uV ms
1.1 1.1 1.1 1.2 1.4 1.4 1.1 1.1 e-
118.23 118.23 118.23 118.23 118.23 118.23 118.23 118.23 uV ms
1.02 1.04 1.14 1.10 1.10 1.07 1.05 1.03 e-
Linea i y E o 1.55 1.31 2.28 1.46 0.71 1.05 0.96 1.10
±%FS
Tempo al noise (da k)
Channel Noise
0 1 2 3 4 5 6 7 Uni s
CG 141.69 132.76 141.36 124.46 125.19 127.34 134.68 136.27 uV/e-
336.24 281.89 962.30 238.19 287.45 294.86 292.51 326.88 uV ms
2.37 2.12 6.81 1.91 2.30 2.32 2.17 2.40 e-
316.45 269.61 855.85 227.05 278.42 286.51 278.04 308.68 uV ms
2.23 2.03 6.05 1.82 2.22 2.25 2.06 2.27 e-
355.24 293.24 1065.56 247.72 295.89 303.28 305.76 343.34 uV ms
2.51 2.21 7.54 1.99 2.36 2.38 2.27 2.52 e-
Tempo al noise (da k)
MEAN
Tempo al noise (da k)
MEDIAN
Tempo al noise (da k)
RMS
2.1 Tes s uc u es pixel a ian s and cha ac e iza ion esul s _
53
4T PIXEL WITH 2-SUPPLIES VS 1-SUPPLY LINEARITY AT LOW INTEGRATED LIGHT LEVEL
2.1.3.3. 4T PIXEL WITH 1METAL_1BUFFER OPTION
Table 6 shows he pixel pe o mances o he one supply pixel e sion connec ed o in e nal bu e . These
op ions a e showing good linea i y a low esponse o e sion 0, 1, 3, 6 and 7 as showed in Fig. 46.
Then, noise pe o mances do no signi ican ly change a e linea iza ion.
Table 6. 4T PIXEL PERFORMANCES FOR THE 1METAS_1BUFFER OPTION
RESPONSIVITY AND DARK NOISE DISTRIBUTION FOR THE 4T PIXEL WITH ONE SUPPLY CONNECTED TO BUFFER
0 1 2 3 4 5 6 7 Uni s
Mic olens
1 1 1 1 1 1 1 1 -
H-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
V-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
FSR 1.37 1.31 1.28 1.39 1.33 1.23 1.41 1.22 V
DR 78.2 78.0 81.8 78.4 75.2 74.3 78.7 77.3 dB
SNRmax 40.66 40.65 40.69 40.96 40.87 40.51 40.88 40.20 dB
CG 124.00 120.33 108.12 116.04 114.83 117.04 120.10 121.07 uV/e-
FWC 11063 10913 11794 11983 11612 10542 11732 10061 e-
PRNU 0.15 0.14 0.30 0.12 0.12 0.14 0.12 0.17 %
DSNU 0.51 0.48 0.79 0.43 0.55 0.72 0.43 0.58 e-
169.39 166.24 103.47 167.76 232.91 239.18 163.70 165.53 uV ms
1.37 1.38 0.96 1.45 2.03 2.04 1.36 1.37 e-
153.90 155.83 101.97 155.69 226.64 233.09 151.73 155.14 uV ms
1.24 1.30 0.94 1.34 1.97 1.99 1.26 1.28 e-
182.33 175.24 105.85 177.44 239.27 245.33 173.42 174.43 uV ms
1.47 1.46 0.98 1.53 2.08 2.10 1.44 1.44 e-
127.74 127.74 127.74 127.74 127.74 127.74 127.74 127.74 uV ms
1.03 1.06 1.18 1.10 1.11 1.09 1.06 1.06 e-
Linea i y E o 0.59 0.65 1.34 0.56 0.24 0.30 0.41 0.43
±%FS
Tempo al noise (da k)
MEAN
Channel Noise
Tempo al noise (da k)
MEDIAN
Tempo al noise (da k)
RMS
Chap e 2 Pixel Noise Op imiza ion
54
2.1.3.4. 4T PIXEL WITH 2METAL_1BUFFER_1EXT OPTION
Table 7 p o ides he pixel pe o mances om he op ion wi h wo supplies connec ing he d ain o ese
ansis o o he in e nal bu e and he d ain o sou ce ollowe o ex e nal supply. Fig. 47 shows he
esponsi i y and da k noise dis ibu ion o he pixel a ian wi h wo supplies connec ed o in e nal bu e
and ex e nal supply. The in e nal ol age and he ex e nal ol age a e bo h se o 3.3 V in hese
measu emen s.
Table 7. 4T PIXEL PERFORMANCES FOR THE 2METALS_1BUFFER_1EXT OPTION
RESPONSIVITY AND DARK NOISE DISTRIBUTION TO THE 4T PIXEL WITH TWO SUPPLIES CONNECTED TO BUFFER AND
EXTERNAL VOLTAGE SUPPLY
2.1.3.5. 5T PIXEL WITH ANTIBLOOMING TRANSISTOR AND ONLY ONE SUPPLY
Table 8 shows pixel pe o mances o 5T pixel e sion wi h an iblooming ansis o and only one supply. This
e sion p esen s highe noise han 4T pixel and simila alues ega ding he o he pe o mance me ics.
Responsi i y and da k noise dis ibu ion a e showed in Fig. 48, whe e he e e ence e sion ( 0) shows ea ly
sa u a ion due o he p esence o an iblooming, o he es o a ian s he sa u a ion appea ea lie han 4T
pixel e sions. Simila pe o mances a e compiled in Table 9 and Fig. 49, co esponding o he 5T pixel wi h
wo supplies. Based on hese cha ac e iza ion esul s 5T pixels a e disca ded o u u e analysis in his chap e
owing o hei combined noise and non-linea i y pe o mances.
0 1 2 3 4 5 6 7 Uni s
Mic olens
1 1 1 1 1 1 1 1 -
H-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
V-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
FSR 1.42 1.37 1.33 1.43 1.38 1.24 1.46 1.28 V
DR 78.7 78.2 80.3 78.5 75.3 74.1 79.1 77.9 dB
SNRmax 41.17 41.19 41.43 41.53 41.53 40.98 41.38 40.78 dB
CG 113.15 109.39 96.56 104.87 102.48 105.58 108.79 110.68 uV/e-
FWC 12534 12489 13798 13635 13435 11701 13382 11556 e-
PRNU 0.13 0.10 0.15 0.09 0.10 0.11 0.11 0.15 %
DSNU 0.50 0.52 0.57 0.39 0.40 0.43 0.43 0.67 e-
165 168 129 170 237 243 162 163 uV ms
1.5 1.5 1.3 1.6 2.3 2.3 1.5 1.5 e-
150.93 156.38 123.34 157.89 229.94 236.49 150.30 151.68 uV ms
1.33 1.43 1.28 1.51 2.24 2.24 1.38 1.37 e-
176.40 176.93 134.39 179.00 243.60 249.57 170.76 171.76 uV ms
1.56 1.62 1.39 1.71 2.38 2.36 1.57 1.55 e-
127.74 127.74 127.74 127.74 127.74 127.74 127.74 127.74 uV ms
1.13 1.17 1.32 1.22 1.25 1.21 1.17 1.15 e-
Linea i y E o 0.77 0.70 1.33 0.54 0.26 0.26 0.40 0.46
±%FS
Tempo al noise (da k)
MEAN
Channel Noise
Tempo al noise (da k)
MEDIAN
Tempo al noise (da k)
RMS
2.1 Tes s uc u es pixel a ian s and cha ac e iza ion esul s _
55
Table 8. 5T PIXEL PERFORMANCES FOR THE 1METAL_1EXT OPTION
RESPONSIVITY AND) DARK NOISE DISTRIBUTION FOR THE 5T PIXEL WITH ONE SUPPLY CONNECTED TO EXTERNAL VOLTAGE
SUPPLY
Table 9. 5T PIXEL PERFORMANCES FOR THE 2METALS_1EXT OPTION
0 1 2 3 4 5 6 7 Uni s
Mic olens
1 1 1 1 1 1 1 1 -
H-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
V-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
FSR 0.65 1.00 0.85 0.98 0.99 1.05 1.45 1.01 V
DR 69.7 73.6 75.5 73.5 71.6 71.6 76.6 73.8 dB
SNRmax 38.30 39.94 39.72 39.98 39.90 40.19 41.06 40.00 dB
CG 125.70 114.28 118.17 116.02 111.61 113.77 116.20 110.57 uV/e-
FWC 5177 8783 7151 8452 8878 9224 12440 9091 e-
PRNU 0.11 0.07 0.22 0.08 0.07 0.07 0.05 0.07 %
DSNU 0.53 0.40 0.81 0.39 0.39 0.38 0.36 0.40 e-
213 210 142 208 261 275 213 206 uV ms
1.7 1.8 1.2 1.8 2.3 2.4 1.8 1.9 e-
120.49 120.49 120.49 120.49 120.49 120.49 120.49 120.49 uV ms
0.96 1.05 1.02 1.04 1.08 1.06 1.04 1.09 e-
Linea i y E o 0.22 0.42 1.54 0.40 0.36 0.51 0.82 0.31
±%FS
Tempo al noise (da k)
Channel Noise
0 1 2 3 4 5 6 7 Uni s
Mic olens
1 1 1 1 1 1 1 1 -
H-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
V-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
FSR 0,83 1,15 0,97 1,12 1,13 1,13 1,48 1,08 V
DR 72,4 75,1 77,0 75,1 74,8 75,1 77,6 74,8 dB
SNRmax 40,17 41,12 40,92 41,33 41,38 41,28 41,86 40,91 dB
CG 113,26 102,58 106,38 107,59 106,09 108,86 106,13 109,12 uV ms/e-
FWC 7333 11172 9155 10374 10627 10390 13966 9919 e-
PRNU 0,12 0,10 0,18 0,09 0,08 0,09 0,09 0,11 %
DSNU 0,50 0,39 0,69 0,39 0,40 0,40 0,38 0,43 e-
199 201 138 195 206 198 195 197 uV ms
1,8 2,0 1,3 1,8 1,9 1,8 1,8 1,8 e-
128,15 128,15 128,15 128,15 128,15 128,15 128,15 128,15 uV ms
1,13 1,25 1,20 1,19 1,21 1,18 1,21 1,17 e-
Linea i y E o 0,40 0,40 1,44 0,51 0,55 0,32 0,93 0,34
±%FS
Tempo al noise (da k)
Channel Noise
Chap e 2 Pixel Noise Op imiza ion
56
RESPONSIVITY AND DARK NOISE DISTRIBUTION FOR THE 5T PIXEL WITH TWO SUPPLIES AND EXTERNAL VOLTAGE SUPPLY
2.2 PIXEL VOLTAGE OPTIMIZATION
2.2.1 Vol age op imiza ion o da k noise
Table 10 illus a es he empi ical dependence o he pixel da k noise wi h he high le el o he ans e ga e
ol age, VGON.
Table 10. DARK NOISE VS 𝑽𝑮𝑶𝑵 FOR THREE TYPES OF PIXELS.
Pixel ype
SF W/L
[μm]
CG
[mV/e]
VGON
[V]
Da k Noise
[eRMS]
Pixel 4T 0
0.42/0.65
112
3.3
1.84
3.2
1.49
3.1
1.47
Pixel 4T 1
0.42/0.8
108
3.3
1.96
3.2
1.48
3.1
1.46
Pixel 4T 2
0.8/0.8
95
3.3
1.79
3.2
1.47
3.1
1.46
Table 11 shows he pixel pe o mances when pixel ol age adjus men s a e made wi h a wo old pu pose:
• o imp o e linea i y;
• o imp o e noise.
Fig. 50 shows he esponsi i y and da k noise dis ibu ion. As al eady explained in he p e ious sec ion,
da k noise pixel measu emen s a e co ec ed by conside ing ha he es chip channel eadou noise is
empo al noise and aking in o accoun i s measu ed alue o he which amoun s o:
118 V ms
The pixel noise is ob ained by sub ac ing his channel noise powe om he measu ed empo al noise powe .
Following his p ocedu e:
he o al empo al noise noise 0 has been imp o ed up o 153.9 V ms (Table 11) when he high
le el o he ans e is se o 3.1 V .
2.2 Pixel ol age op imiza ion _
57
Table 11. 4T PIXEL PERFORMANCES WITH TWO SUPPLIES, RESET SUPPLY 3.0 V, SF SUPPLY 3.3V AND VGON 3.1 V
(a) (b)
4T PIXEL WITH RESET SUPPLY AT 3.0V, SOURCE FOLLOWER SUPPLY AT 3.3V AND VGON=3.1V: (A) RESPONSIVITY AND
(B) DARK NOISE DISTRIBUTION
Table 12 collec s he o e all noise con ibu ions o he di e en mul i-supplies 4T pixel op ions and shows
he pixel noise a e sub ac ing he measu ed channel powe noise om he es chip eadou .
Table 12. 4T PIXEL NOISE
0
1
2
3
4
5
6
7
Uni s
Tempo al Noise
153.91
151.11
134.15
153.60
248.86
257.57
150.57
151.28
V ms
Channel Noise
118.64
118.64
118.64
118.64
118.64
118.64
118.64
118.64
V ms
Pixel Noise
98.05
93.59
62.61
97.54
218.75
228.61
92.71
93.85
V ms
CG
112.07
108.39
95.63
103.58
102.77
104.43
107.97
108.89
V ms
Pixel Noise
0.87
0.86
0.65
0.94
2.13
2.19
0.86
0.86
e-
0 1 2 3 4 5 6 7 Uni s
Mic olens
1 1 1 1 1 1 1 1 -
H-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
V-Pixel size 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 um
FSR 1.48 1.42 1.45 1.37 1.23 1.10 1.39 1.36 V
DR 79.7 79.5 80.7 79.0 73.9 72.6 79.3 79.1 dB
SNRmax 41.38 41.24 41.90 41.25 41.11 40.74 41.14 41.09 dB
CG 112.07 108.39 95.63 103.58 102.77 104.43 107.97 108.89 uV/e-
FWC 13215 13103 15181 13215 11967 10540 12878 12531 e-
PRNU 0.04 0.03 0.05 0.02 0.03 0.04 0.04 0.04 %
DSNU 0.33 0.29 0.32 0.29 0.29 0.29 0.29 0.32 e-
153.91 151.11 134.15 153.60 248.86 257.57 150.57 151.28 uV ms
1.37 1.39 1.40 1.48 2.42 2.47 1.39 1.39 e-
141.13 140.84 126.81 143.64 244.63 254.07 139.72 140.99 uV ms
1.26 1.30 1.33 1.39 2.38 2.43 1.29 1.29 e-
164.52 158.59 139.36 161.28 253.86 262.43 158.61 159.01 uV ms
1.47 1.46 1.46 1.56 2.47 2.51 1.47 1.46 e-
118.64 118.64 118.64 118.64 118.64 118.64 118.64 118.64 uV ms
1.06 1.09 1.24 1.15 1.15 1.14 1.10 1.09 e-
Linea i y E o (EMVA) 0.15 0.17 0.27 0.11 0.16 0.17 0.14 0.20
±%FS
Linea i y E o (Full Scale) 0.29 0.30 0.46 0.19 0.23 0.27 0.27 0.34
±%FS
Tempo al noise (da k)
MEAN
Channel Noise
Tempo al noise (da k)
MEDIAN
Tempo al noise (da k)
RMS
Chap e 2 Pixel Noise Op imiza ion
64
𝒗𝒏,𝒕𝒉,𝑪𝑴𝑺
𝟐
≈𝑴 𝑺𝑵,𝒕𝒉𝝎𝒄
𝟐
(2.45)
The inpu e e ed he mal noise powe o he CMS ope a ion is gi en by:
𝒗𝒏,𝒕𝒉,𝑪𝑴𝑺𝒊𝒏𝒑𝒖𝒕
𝟐
)𝒗𝒏,𝒕𝒉,𝑪𝑴𝑺
𝟐
/𝑴𝟐
(2.46)
Hence, he he mal noise powe is M- imes smalle han he alue co esponding o a simple CDS ope a ion.
𝒗𝒏,𝒕𝒉,𝑪𝑴𝑺𝒊𝒏𝒑𝒖𝒕
𝟐
≈ 𝑺𝑵,𝒕𝒉𝝎𝒄
𝟐∙𝑴
(2.47)
Low- equency noise
Al hough CMS e ec i ely a enua es he impac o he mal noise, i is no so e ec i e o he low equency
noise because he highe he numbe o samples M, he longe he ime in e al be ween ese and signal:
(M·To+TG). Hence, he low equency noise con ibu ion inc eases wi h M and he deco ela ion o low
equency noise con ibu ions o ese and signal inc eases as well. This aises a ade-o conce ning M
selec ion which is also cons ained by he equi emen no o educe eadou speed ine icien ly.
The pixel es chip can be ope a ed wi h non-des uc i e eadou o mul isampling echnique. I has been
used o measu e he e iciency o mul isampling echnique o he pixel a ian s. Fig. 61 shows he noise
measu emen when a ying he M pa ame e om 1 o 28, he dash line is showing he heo ical da k noise
i he noise con ibu ion would be all he mal noise, and hen i would be di ided by M. These expe imen al
da a e eal ha :
Mos o he pixel a ian s in he es chip ha e and op imum alue o M=8.
PIXEL VARIANTS WITH CMS FOR DIFFERENT M PARAMETER
2.4 MODEL EXTRAPOLATION FOR PROPOSED ADCS ARCHITECTURE
This sec ion is in blank because i is con aining con iden ial in o ma ion.
65
CHAPTER 3
3. TWO STAGES ADC FOR LOW NOISE CMOS
IMAGE SENSORS
ABSTRACT
This chap e desc ibes he a chi ec u e, op imiza ion and he pe o mances o a wo-s age ADC concei ed
o low noise and high Dynamic Range (DR) applica ions. The p edecesso o he p esen ed wo s age ADC
has been published in [Gonza15] whe e he i s s age consis s o a sigma-del a ADC [Chae10] and he second
s age is he well-known single slope ADC [Yoshi06][Toyam11].
The a chi ec u e p esen ed in his chap e op imizes he a ea and noise since he sample-and-hold is
emo ed om he eadou chain, he Co ela ed Double Sampling (CDS) ope a ion is embedded in o he i s
s age o he ADC and he sigma-del a a chi ec u e op imizes he ade-o be ween capaci o s size and inpu
e e enced noise. The powe consump ion o he comple e eadou chain including all sub-blocks is 610 µW
o con e he pixel signal wi h a noise o 72 uV ms in a ow ime o 5 mic oseconds wi h 14 bi s esolu ion
in a s anda d 0.18 µm CMOS echnology. The ADC a chi ec u e is ully lexible and can inc ease he ADC
esolu ion ex ending he ow ime up o 15 and 16 bi s, educing he noise when inc easing esolu ion.
In he li e a u e [Kwon18], se e al Figu es o Me i (FoM) a e de ined o compa e column ADC
pe o mances and cos . This ADC a chi ec u e p esen s a FoM alue o 0.006 V ms· J, acco ding o he
ollowing exp ession:
𝐹𝑜𝑀=𝑁𝑠∙𝑃𝑜𝑤𝑒𝑟
2𝑁𝑏𝑖𝑡𝑠∙𝑓𝑠
(3.1)
This FoM alue is much smalle han hose achie ed wi h a ypical single-sope ADC, such as he CIS in [Lim10]
whose FoM is 0.041 V ms· J.
This chap e is o ganized as ollows. Sec ion 3.1 will p esen he mo i a ion o his eadou a chi ec u e
based in wo s ages ADC. Sec ion 3.2 desc ibes he i s s age o he ADC and he Sec ion 3.3 desc ibes he
second s age o he ADC, and he Sec ion 3.4 he analog amp gene a o used in he second s age o he ADC.
The eadou can wo k in non-des uc i e mode, deli e ing he da a o ese le el and signal le el, i is
desc ibed in Sec ion 3.5. Sec ion 3.6 shows he eadou channel pe o mances. The eadou wo king in
mul isampling is desc ibed in Sec ion 3.7, and inally. he desc ip ion o he needed calib a ion and co ec ion
is gi en in Sec ion 3.8.
Chap e 3 Two s ages ADC o low noise CMOS Image Senso s
66
3.1 INTRODUCTION
3.1.1 Mo i a ion
In many applica ions, as he speed equi emen s inc ease, he numbe o eadou channels
(each one concep ually composed o one CDS ampli ie plus an ADC) needs o be inc eased as
well. Theo e ically, he achie able ame a e o a CIS linea ly inc eases wi h he numbe o
eadou elemen s wo king in pa allel. In ac , hese can be seen as a single eadou channel wi h
e ec i e speed boos ed by he numbe o eplicas.
In p ac ice, his pa allelism can be inc eased as much as needed, al hough a e he ADC he
da a should be se ialized and send o he ecei e , whe e i is aced ano he ac ual limi a ion;
he da a h oughpu , which is de ined by he o al numbe o easible da a po s by he allowable
da a a e h ow e e y po . Today many CIS employ a eadou elemen pe pixel column, so ha
such eadou elemen is in cha ge o p ocessing he signal coming om pixels in a single column.
The adeo be ween accu acy and speed o his solu ion can be add essed in wo manne s: on
he one hand, high-speed CIS clea ly bene i om he enhanced o e all bandwid h; on he o he ,
in high esolu ion applica ion, he inc ease in pa allelism wi h he consequen elaxa ion o he
indi idual ADC bandwid h allows imp o ing he accu acy o each uni elemen .
The abili y o ade powe , noise and speed is in insic o all ADC opologies bu i becomes
pa icula ly e iden in he so-called in eg a ing ADC opologies. In pa icula , he single- amp
ADC has been ound e y app op ia e o CIS, wi h column-le el eadou pa allelism, because
he con en in ci cui y pe columns is small, hus ende ing i compa ible wi h e y low pixel
pi ch [Nie20] [Sukew13]. In hese ADCs he numbe o clock cycles equi ed o digi ize wi h B-bi
esolu ion is p opo ional o 2B. Fo example, in case o he well-known and ex ensi ely used
single-slope ADC he 10-bi esolu ion would equi e 1024 clock cycles o ob ain a single digi ized
da a, while 4096 clock cycles a e equi ed o code 12 bi s, e c., which illus a es he heo e ical
accu acy-speed adeo : sol ing one mo e bi amoun s o doubling he con e sion ime in case
o single-slope amp ADC.
Howe e when inc easing he esolu ion and educing consequen ly he quan iza ion noise,
o he noise sou ces should scale down in o de o ge ac ual bene i om he inc emen o
esolu ion. Among hese, sampling noise mus also be educed. In Swi ched Capaci o (SC)
ci cui s, he na u al way o achie e such a educ ion consis s o inc easing he alue o he
sampling capaci o . Howe e , sampling noise RMS alue scales wi h squa e oo o he
capaci ance, which soon equi es huge capaci o s o be complian wi h he inc emen o ADC
esolu ion. This leads o e y ine icien pe -column implemen a ions, bo h in powe
consump ion and a ea occupa ion, especially o e y low pixel pi ch.
Using o e sampled ADCs has been p oposed In his con ex [Kim12]. They employ
o e sampling o lowe he sampling noise o gi en capaci ance. Mo eo e , noise- eedback
echniques can be applied o educe he quan iza ion noise o low- esolu ion quan ize s. This is
he case o example o he adi ional i s -o de sigma-del a ADC [Pelg 13], which may use a
single-bi quan ize . No e ha his opology is e y simple in e ms o analog con en because i
equi es jus one SC in eg a o besides he compa a o used in he single- amp ADC. The
o e sampling sigma-del a ADC a chi ec u e, depic ed in Fig. 62, has been used in ensi ely in
audio, communica ions and ins umen a ion applica ions [Manga12] [Medei99].
This a chi ec u e akes ad an age o wo basic p ope ies o educe he quan iza ion powe
noise inside he signal band and inc ease he in-band ADC equi alen esolu ion, namely:
3.1 In oduc ion _
67
OVERSAMPLING ƩΔ ADC ARCHITECTURE AND BLOCK DIAGRAM
1. O e sampling: i is based on sampling he signal a highe equency han i s Nyquis
equency. The a io be ween he sampling equency and he Nyquis equency de ines
he o e sampling a io, OSR= s/(2 b), whe e b is he bandwid h equency o he signal.
The o e sampling implies wo main ad an ages:
• Since s /2 is highe han b, he componen s in he ange [ b, s - b] a e no olded
in o he signal band a e he sampling p ocess. As consequence he demands o
he an i-aliasing il e a he inpu o he ADC a e much mo e elaxed in he
ansi ion om he pass band o he ejec ion band.
• In he quan iza ion o an o e sampled signal, only a pa o he quan iza ion e o
is in he band o he signal, and hen powe in he band is:
𝑃𝑄=∫ 𝑆𝐸(𝑓)𝑑𝑓=
𝑓𝑏
−𝑓𝑏∫𝛥2
12𝑓𝑠𝑑𝑓=
𝑓𝑏
−𝑓𝑏𝛥2
12𝑂𝑆𝑅
(3.2)
Then he highe he o e sampling a io he lowe is he quan iza ion noise.
2. Noise shaping: wi h ce ain p ocessing he quan iza ion noise can be pushed o highe
equencies ou o he signal band, and hen i can be educed using low-pass il e ing in
he digi al domain.
ANTI-ALIASING IN OVERSAMPLING CONVERTERS
Howe e , in image senso s eadou channel he sigma-del a a chi ec u e canno be applied
di ec ly since he pixel signal is changing om one pixel o he nex pixel o be eadou , and
addi ionally he pixel eadou ope a ion needs o pe o m he CDS ope a ion in o de o cancel
he misma ch be ween pixel sou ce ollowe h eshold ol age a ia ion ou .
In ac , Inc emen al ADCs (IADC), see [Chen15], a e sui able o image senso eadou s. IADCs
a e a special subclass o sigma-del a ADCs, whe e he ope a ion is pe iodically ese using a ini e
memo y. The IADCs a e es a ed in e e y con e sion. They o e simila ad an ages as high
accu acy, s abili y and low quan iza ion noise allowing he possibili y o mul iplexing he signals
coming om di e en pixels.
Like he sigma-del a modula ion, IADC´s ope a ion is obus in he p esence o ci cui non
ideali ies and in e e ences, whose e ec s a e in essence a enua ed by he numbe o imes
he inpu signal is sampled o ge a single con e sion. We call o e sampling a io OSR o such a
igu e. A s aigh - o wa d analysis [Rio06] shows ha he sampling noise o an o e sampled
case equals kT/(OSR·Cs), whe e Cs s ands o he sampling capaci o alue. No e ha OSR and
he capaci ance alues a e exchangeable in e ms o noise.
b s- b
xs( ) S/H
s
Ʃ
xs(n)
H(z)
DAC
B b
OSR
Y(n) Y (n)
Yd(n)
BN
Decima ion
Digi al il e Down sampling
Modula o ƩΔ
an i-aliasing il e
-
b s
s/2- s s- b
An i-aliasing
il e
Chap e 3 Two s ages ADC o low noise CMOS Image Senso s
68
Fo ully exploi ing he bene i s o he educed sampling noise (b ough h ough a high M
ins ead o a la ge capaci o ), i is clea ha noise sou ces be o e he ADC mus be minimized,
which is no always achie ed. Fo example, in [Bisia17] a CDS ampli ie p ecedes he sigma-del a
ADC, o in [Gonza15] whose ADC is composed by a o e sampled ADC and i is p eceded by wo
Sample-and-Hold (SH) ci cui s, one o sample he ese le el and ano he o sample he signal
le el as shown in Fig. 64. As highligh ed h ough he noise me ics included in he igu e, he
noise con ibu ion om he p e ious s ages may mask he bene i s o he sigma-del a opology,
unless huge capaci o a e used in he CDS block on end.
ADC ARCHITECTURE COMPOSE OF 1ST STAGE SIGMA-DELTA PRECEDED BY SAMPLE-AND-HOLD
A possible solu ion consis s in emo ing he analog CDS block, eplacing i by a digi al
sub ac ion sec ion 1.4.3 This consis o pe o ming indi idual con e sions o he pixel ese and
signal le els, by using he pe -column ADC and, once in he digi al domain, sub ac ing hem
om each o he o ge he ideo signal [Chen18]. In combina ion wi h o e sampled sigma-del a
echniques, his kind o digi al CDS b ings wo clea ad an ages:
• on he one hand, sampling noise is de e mined by he p oduc M·Cs a he han by he
sampling capaci ance i sel ;
• on he o he , wha e e noise sou ce p eceding he ADC is also o e sampled, wi h he
subsequen a enua ion by M.
As a d awback, in digi al CDS ope a ion he con e sion noise is doubled in powe bu signi ican
a ea and powe can be sa ed i he analog sampling and o he noises a e a enua ed, because
small capaci o s a e usable. Besides ha , he digi al CDS has wo ob ious penal ies:
• on he one hand, i doubles he con e sion ime (because wo con e sions pe pixel a e
ac ually needed);
• on he o he , i equi es digi al da a s o age and sub ac ion o doubling he h ough-pu
in case he digi al CDS is decided o be done in ex e nal ecei e .
3.1.2 Ou line o p oposed eadou a chi ec u e
The new echnique p esen ed in his Chap e embeds analog CDS in he IADC ope a ion wi hou
equi ing a p eceding CDS block. The inc emen al ADC is implemen ed in he i s s age o he
=
3.1 In oduc ion _
69
eadou channel, as shown in Fig. 65, whe e he ele an noise me ics a e included o
compa ison o Fig.3.
This i s s age inco po a es he CDS unc ion di ec ly along wi h he o e sampling unc ion
inhe i o he IADC a chi ec u e. The i s s age ADC gene a es he Mos Signi ican Bi s (MSB)
o he digi ized ou pu s and p oduces an analog esidue which is con e ed by he second s age,
he Single-Slope ADC (SS-ADC). Finally, he SS-ADC gene a es he Leas Signi ican Bi s (LSB) o
he digi ized pixel wo d. The pixel ou pu s digi ized e sion including CDS ideo a e ob ained
wi h no ex a digi al memo y o pos -p ocessing needed, i educes he o e all powe
consump ion and equi ed a ea.
TWO STAGE ADCS INCREMENTAL AND SINGLE-SLOPE WITHOUT NEITHER S&H NOR CDS BLOCKS
This a chi ec u e op imizes he con e sion ime because i wo ks in he so-called o e lap
mode, as Fig. 66 illus a es. The i s ADC ope a es a he same ime he pixel is ou pu ing he
ese and signal pixel le els, a e ha he analog esidue coming om he i s s age ADC is
sampled by he second s age ADC and e en ually con e ed in o he digi al domain. The
con e sion o he N- h da a in he second ADC is pe o med in pa allel a he same ime he da a
N+1 is con e ed by he i s s age ADC. This pa allel ope a ion educes he ow ime o he
con e sion ime needed by he i s ADC and he pixel ope a ion. The digi al ou pu wo d is
composed by he combina ion o he wo digi al wo ds coming om he wo ADCs as shown in:
𝐷𝑜𝑢𝑡=𝐷1∙2𝑁2+𝐷2
(3.3)
whe e he ou pu o he i s ADC is scaled by he numbe o bi s o he second s age (𝑁2)
powe ed by 2.
TWO STAGE ADCS OVERLAP OPERATION
da a column line
Pixel SF
sel
P og.
Cu en
Sou ce
+
P og.
O se
1s s age ADC
2nd s age
ADC
7,6,5,4 10,9
MSBs LSBs
CDS
Analog
Bu e
Residue
61.44uV ms 25.8uV ms
Readou channel noise = 75.4uV ms
Rese se ling Signal se ling
Residue ou
SEL
RST
TRF
4.74usec
1s s age ADC
Residue CONV2nd s age ADC
4.74usec
4.74usec
On-Chip
SIGNAL in eg a ionRESET in eg a ion
4.74usec
Ou pu Wo d=D1·2N2+D2
Rese se ling Signal se ling
Residue ou SIGNAL in eg a ionRESET in eg a ion
Chap e 3 Two s ages ADC o low noise CMOS Image Senso s
70
3.2 FIRST STAGE: INCREMENTAL ADC
This wo k uses he i s -o de sigma-del a ADC opology shown in Fig. 67. I is basically composed o an
in eg a o ha each phase cycle accumula es he di e ence be ween he inpu signal and he eed-back
signal, and a compa a o ha compa es he in eg a o ou pu wi h a ol age e e ence V /2.
FIRST-ORDER INCREMENTAL ADC
In a con en ional IADC, in case he compa a o ou pu is high (1), a ol age V is ed back. O he wise, i
he compa a o ou pu is low (0), g ound ol age is ed back. In he he ein p oposed a chi ec u e, he
eedback signal is modi ied as explained below in acco dance wi h he pixel le el being p ocessed, in o de
o pe o m he CDS and a oid any sa u a ions.
Con a y o gene al pu pose IADCs, a he beginning o each con e sion, bo h he in eg a o and he digi al
accumula o a e ese . This ese is needed in o de o cope wi h he high- equency componen s associa ed
o he pixel- o-pixel ansi ions.
A e he ese , du ing each con e sion, he ou pu o he compa a o (0 o 1) is accumula ed digi ally,
and he ou pu code is o med o he sum o he consecu i e compa a o decisions. Fo example, gi en a
numbe 4 o consecu i e compa isons, possible digi al ou pu s a e:
• 0000 -> digi al code 0
• 1000 o 0100 o 0010 o 0001 -> digi al code 1
• 1100 o 0110 o 0011 -> digi al code 2
• 1110 o 0111 -> digi al code 3
• 1111 -> digi al code 4
which allows esol ing o 5 digi al le els. I ollows ha he numbe o digi al le els ob ained wi h a numbe
o consecu i e compa isons, nC, is nC +1, so ha he equi alen esolu ion equals o N1=log2(nC +1), whe e N1
s ands o he numbe o bi s deli e ed by he i s s age o he ADC.
3.2.1 Func ional desc ip ion
The i s s age implemen a ion consis s o employing signal in e sion o pe o m he CDS ope a ion as he
sigma-del a p ocessing is being ca ied ou . So, he inal code a he coun e ep esen s he sub ac ion
be ween he wo pixel le els. In pa icula , i s he pixel ese le el will be accumula ed (by he in eg a o ) a
numbe o imes and hen he signal le el will be accumula ed he same numbe o imes, bu wi h in e ed
pola i y so ha ac ual mul iple sub ac ion ake place. Along his p ocess, he accumula ed compa a o
decisions will o m a digi al ep esen a ion o he di e ence.
In o de o keep he app op ia e eedback so ha he in eg a o ou pu emains bounded by V , du ing
he p ocessing o he ese le el he DAC ou pu is V when he compa a o is high (0 o he wise). Howe e ,
g
DAC
−
Coun e
In eg a o
Adde
Inpu
Rese
Ou pu
Compa a o
V /2
V
Y
3.2 Fi s s age: inc emen al ADC _
71
while he nega ed signal le el is being con e ed, he DAC ou pu becomes 0 o a high compa a o ou pu
and V o a low one.
Fig. 68 shows he implemen a ion o he i s s age o he ADC ( he digi al coun e is no shown). I consis s
o a SC in eg a o and a la ched compa a o . Clock phases equi ed o ope a ion a e also included a he
bo om.
C1
C2
V
−
+
+
−
Vcomp
pha
AND
phb
ph0
ph2
OTA NOT(ph2)
ph0
ph1
ph2
YTo coun e
ph1
F om pixel
pha
phb
nC In eg a ions o he ese le el nC In eg a ions o he signal le el
AND
XOR
phc
phc
Vcm
Vze o
Vze o
FIRST STAGE IMPLEMENTATION
The es o sec ion is in blank because i is con aining con iden ial in o ma ion.
3.2.2 Implemen a ion
This sec ion is in blank because i is con aining con iden ial in o ma ion.
3.2.3 Noise Analysis
Pa o his sec ion is in blank because i is con aining con iden ial in o ma ion.
Table 1 shows he inpu e e enced noise compu a ion o di e en bi esolu ion in i s s age o he
ADC, and he noise con ibu ion assuming pixel con e sion gain (CG) equal o 100uV/e-.
Table 1. NOISE POWER ANALYSIS VS NUMBER OF BITS
Bi s in 1s
s age ADC
Powe Noise
(V ms2)
Vol age noise
(V ms)
Noise con ibu ion
(CG=100V/e-) in e-
4
4.46E-9
66.8E-6
0.67
5
2.17E-9
46.6E-6
0.47
6
1.07E-9
32.7E-6
0.33
Chap e 3 Two s ages ADC o low noise CMOS Image Senso s
72
3.3 SECOND STAGE: SS-ADC
Fig. 69 shows he concep o he single slope ADC in he second AD s age. Da a coming om he i s s age
a e sampled and hen compa ed wi h an analog amp e e ence. Ini ially, he analog amp and a digi al
coun e s a s he coun simul aneously, a e a p og ammable delay. When he analog amp e e ence
c osses he analog inpu le el, he compa a o oggles, and he digi al coun is sampled in he digi al egis e .
The p og ammable delay allows inse ing digi al o se be ween he analog amp and he digi al coun , and
hence in he inal digi al wo d.
Fig. 70 (a) shows he schema ic design o he second s age ADC. Typical clock phase and wa e o ms a e
shown in Fig. 70 (b).
SECOND STAGE ADC ARCHITECTURE.
(A) SECOND STAGE ADC SCHEMATIC; (B) ASSOCIATED WAVEFORMS
(a)
(b)
3.3 Second s age: SS-ADC _
73
3.3.1 Func ional desc ip ion
The e a e wo main ope a ing phases in he SS-ADC: he sampling phase and he compa ison phase.
• Du ing he sampling phase, he compa a o is ope a ing in close-loop, so p node is se o he inpu
common mode ol age (Vcmi ), and he n node is connec ed o he ou pu o he ope a ional ampli ie
in a uni a y gain con igu a ion se ing he common mode ol age plus he ope a ional ampli ie o se :
Vcmi +VOTAo se . A he end o he sampling phase, capaci o CSA_ADC2 s o es he esidue om i s s age
o he ADC, minus he ampli ie o se . This sampling ope a ion is including au o-ze o ope a ion
imp o ing he ampli ie ope a ion o se dependence and he low noise con ibu ion [Enz96].
• Du ing he compa ison phase, he compa a o is ope a ing in open-loop, so CSA_ADC2 capaci o is
keeping he s o ed cha ge du ing he sampling phase. I can be esol ed ha he ol age in n and p
nodes du ing compa ison phase a e as ollows:
𝑣𝑛=𝑣𝑐𝑚𝑖+𝑣𝑜𝑓𝑓
1+1 𝐴𝑂
⁄+𝛼(𝑣𝑟𝑎𝑚𝑝𝑛−𝑣𝑑𝑎𝑡𝑎)
𝑣𝑝=𝑣𝑟𝑎𝑚𝑝𝑝
(3.4)
whe e Vcmi is he ol age a he p node when sampling (compa a o common-mode ol age), o is he
compa a o o se ol age, CP is he pa asi ic inpu capaci ance a he inpu nodes o he compa a o ,
amp_p, amp_n a e he posi i e and nega i e amp ol ages and da a is he sampled ol age s o ed
du ing he phi_sa phase; is he scale ac o gi en by nex exp ession:
𝛼= 𝐶𝑆𝐴𝐴𝐷𝐶2
𝐶𝑃+𝐶𝑆𝐴𝐴𝐷𝐶2
(3.5)
Assuming an ideal compa a o , he di e en ial ol age a he inpu o he compa a o can be
app oxima ed as:
∆𝑉=𝑣𝑝−𝑣𝑛=𝑣𝑑𝑎𝑡𝑎−𝑣𝑑𝑎𝑡𝑎−(𝑣𝑟𝑎𝑚𝑝𝑛−𝑣𝑟𝑎𝑚𝑝𝑝)
(3.6)
The sampled ol age ( da a) is always posi i e and s ands o he in o ma ion whe eas he di e ence ol age
( amp_p - amp_n) is linea ly dec easing. The e o e, he ini ial ou pu o he compa a o would be a high ol age
and his would change i s decision o a low ol age when he di e ence amp ol age would be such ha
∆ ≤0. The signal can be codi ied by using a digi al coun e ha ini ializes when he analog amp s a
dec easing. The bigge he numbe o coun ed cycles, he bigge he inpu signal is.
Assume ha he exp essions o amp_p and amp_n a e:
𝑣𝑟𝑎𝑚𝑝𝑝=𝑣𝑐𝑚𝑖2
𝑣𝑟𝑎𝑚𝑝𝑛=𝑣𝑚𝑖𝑛+𝑚·𝑡 𝑣𝑟𝑎𝑚𝑝𝑝∈[𝑣𝑚𝑖𝑛,𝑣𝑚𝑖𝑛+𝐹𝑆𝐴𝐷𝐶2]
(3.7)
whe e max co esponds o he lowes ol age p og ammed in he analog amp e e ence, 𝑣𝑐𝑚𝑖2 is he
ol age e e ence gene a ed a 𝑣𝑝 node and he 𝐹𝑆𝐴𝐷𝐶2 is he inpu ull scale ange o he second s age o
he ADC. Then, he exp essions o n and p (du ing compa ison) a e:
𝑣𝑛=𝑣𝑐𝑚𝑖+𝑣𝑜𝑓𝑓
1+1 𝐴𝑂
⁄+𝛼(𝑣𝑚𝑖𝑛+𝑚·𝑡−𝑣𝑑𝑎𝑡𝑎)
𝑣𝑝=𝑣𝑐𝑚𝑖2
(3.8)
I can be deduced om p e ious equa ion, ha i cmi and cmi2 a e no equal, p is shi ed om cmi (du ing
sampling) o cmi2 (in compa ison), and i in oduces and analog o se equal o cmi − cmi2. When n eaches
he cmi2 alue, he compa a o will oggle. I mus be ensu ed ha he compa a o wo ks p ope ly o his
inpu ol age and in any case he 𝑣𝑝,𝑣𝑝 ol age do no o e pass he supply and g ound ail ol ages.
Fig. 71 shows he ansien e olu ion o he p and n ol ages depending o he inpu signal ( da a),
assuming an ideal compa a o (A0→∞, o =0). The maximum and minimum alues o n will be gi en by:
𝑣𝑛,𝑚𝑎𝑥=𝑣𝑐𝑚𝑖+ 𝛼(𝑣𝑚𝑖𝑛+𝐹𝑆−𝑣𝑑𝑎𝑡𝑎)
𝑣𝑛,𝑚𝑖𝑛=𝑣𝑐𝑚𝑖+𝛼(𝑣𝑚𝑖𝑛−𝑣𝑑𝑎𝑡𝑎)
(3.9)
Chap e 3 Two s ages ADC o low noise CMOS Image Senso s
80
3.6.2 Noise
Table 5 and Table 6 show he noise simula ion esul s and he noise measu emen s espec i ely, o he
in e nal CDS eadou mode. The noise simula ion and calcula ion is e y simila o he measu emen s
alida ing he me hodology o calcula e and op imize he noise which is e y impo an in low noise
applica ions. Conside ing a con e sion gain o 100uV/e he eadou con ibu ion o he o al noise would be
0.72e in 14bi s, 0.48 in 15 bi s and 0.33 when using 16bi s.
Table 7 shows he simula ed noise calcula ion o he comple e eadou noise when i ope a es in ex e nal
CDS. In his case, he noise con ibu ion om he i s s age is lowe since he numbe o in eg a ions is he
hal due o he CDS ope a ion is done ex e nally a he digi al domain. Howe e , he inal noise calcula ion
should conside he da a a e CDS, and hen i esul s in he double o powe noise o one componen ei he
i is coming om he ese le el o he signal le el.
Table 5. NOISE POWER ANALYSIS FOR COMPLETE READOUT IN INTERNAL CDS MODE
Readou
mode
#I e
ADC1
Noise
ADC2
Noise
Ramp
Noise
Quan iza ion
Noise
Readou
Noise
Readou
Noise
[bi s]
[V ms]
[V ms]
[V ms]
[V ms]
[V ms]
CG=100 V/e
[e]
14
16
66.80
17.84
25.75
26.43
78.37
0.78
15
32
46.60
12.49
12.88
13.21
51.65
0.52
16
64
32.70
6.25
6.44
6.61
34.55
0.35
Table 6. NOISE MEASURE IN CHARACTERIZATION
Readou
mode
#I e
Readou
Noise
Readou Noise
[bi s]
[V ms]
CG=100V/e [e]
14
16
72.32
0.72
15
32
48.06
0.48
16
64
32.73
0.33
Table 7. NOISE POWER ANALYSIS FOR COMPLETE READOUT IN EXTERNAL CDS MODE
Readou
mode
[bi s]
#I
ADC1
Noise
[VRms]
ADC2
Noise
[VRms]
Ramp
Noise
[V ms]
Quan.
Noise
[V ms]
Readou
Noise
One sample
[V ms]
Readou
Noise
A e CDS
[V ms]
Readou
Noise
CG=100V/e
[e]
14
16
49.30
17.84
25.75
26.43
64.11
90.66
0.91
15
32
33.91
12.49
12.88
13.21
40.58
57.38
0.57
16
64
23.66
6.25
6.44
6.61
26.16
36.99
0.37
3.6 Readou channel pe o mance _
81
The eadou ex e nal CDS mode has been measu ed wo king a 14 bi s, he esul is gi en in Table 8. In
case, o eques ing highe p ecision in his mode mul isampling mode can be used, his mode is desc ibed in
nex sec ion.
Table 8. NOISE MEASURE IN CHARACTERIZATION FOR EXTERNAL CDS MODE
Readou
mode
#I e
Readou Noise
one con e sion
Readou Noise
a e CDS
To al Readou
Noise
[bi s]
[V ms]
[V ms]
CG=100V/e [e]
14
16
70.5
81.4uV ms
0.81
3.7 READOUT WORKING IN CORRELATED MULTISAMPLING WITH EXTERNAL
CDS
The ex e nal eadou CDS mode (Sec ion 3.5) is also he basis o he mul isampling mode wi h ex e nal CDS,
his mode is also known as Non Des uc i e Readou (NDR). In con as o he no mal eadou , when he NDR
mode is pe o med, ei he he ese le el o he signal le el is con e ed and digi alized om he pixel in each
con e sion ime. The pixel le els a e con e ed wi h espec o on-chip e e ence. Such a e e ence le el is
sampled by au oma ically selec ion he ope a ion mode, as desc ibed in Sec ion 3.5.1 , whe e he e e ence
is 𝑉𝑧𝑒𝑟𝑜.
Basic noise es ima ion
The ope a ion does no o e w i e he sense node o he pixel, hence ei he he ese le el o he signal le el
can be ead (wi h espec he same e e ence) a p ede ined numbe o imes, hus p o iding he mul iple
NDR (mNDR), also known as co ela ed mul isampling echnique. Fig. 79 depic s he eadou channel wo king
in co ela ed mul isampling mode, whe e a e done he se ling o he ese signal i s s age o he ADC
s a s con e ing he i s ime, a e ha he esidue is con e ing by he second s age o he ADC and in
pa allel he i s s age con e s he second sample o he ese , and his ope a ion is done up o he M
con e sions o he ese . A e ha , he cha ge is ans e ed om he pho odiode o he loa ing di usion
and he signal le el s a s doing he se ling. Simila o he ese le el, he i s s age o he ADC con e s he
i s sample, and he second sample is con e ed in pa allel wi h he con e sion o he esidue by he second
s age o he ADC, up o he M con e sions o he signal.
This a chi ec u e only needs M imes he con e sion ime. This ea u e ep esen s a g ea ad an age wi h
espec o he ypical implemen a ions ha need 2∙M imes he con e sion ime. I is possible due o he
inhe en pa allelism in oduced by he wo s ages ADC.
Ex e nal p ocessing can be used in o de o lowe a pa o he eadou know by a e aging he samples o
he ese and he signal be o e being sub ac ed. In pa icula all wide-band noise con ibu ions gene a ed in
he eadou pa h including he in-pixel sou ce ollowe ansis o a e educed by √𝑀 in ol age (RMS) o by
M in powe noise. Then, a e he CDS sub ac ion (signal minus ese ), he ese kTC-noise is emo ed and
he inal noise con ibu ion coming om whi e noise con ibu o s can be es ima ed as:
𝑁𝑚𝑁𝐷𝑅=√2
𝑀𝑁𝑟𝑜
(3.12)
whe e 𝑁𝑟𝑜 s ands o he ol age noise in oduce by he noise o all whi e noise con ibu o s in he eadou
pa h when con e ing one componen , ei he he ese o he signal, he ac o √2 comes om he double
eadou equi ed o pe o m he digi al double sampling o ese noise emo al. This exp ession highligh s
Chap e 3 Two s ages ADC o low noise CMOS Image Senso s
82
he impac o he numbe o con e sions (M) done pe e e y pixel le el. Fo example, wi h a eadou noise
o 𝑁𝑟𝑜=1e-, wi h a mul isampling o M=1 he o al noise would be 1.41e-, whe eas wi h M=2, he noise is 1e-
, and wi h M=4 i is 0.7e-.
Low- equency noise impac
The abo e es ima ion o he noise in mul isampling mode does no ake in o accoun he e ec o low-
equency noise componen s in he eadou pixel pa h. Flicke noise (1/ ) and Random Teleg aph Signal noise
(RTS) a e gene a ed signi ican ly in he pixel sou ce ollowe since ypical he ansis o size is small in o de
o achie e high con e sion gain . The pixel sou ce ollowe low equency noise sou ces a e sampled wice
and sub ac ed om each o he , educing he o al con ibu ion. Howe e , i depends on he elapsed ime
be ween he wo samples, he low equency con ibu ion will be la ge as soon as he ime be ween he wo
samples is la ge . I wo ks in opposi e di ec ion wi h he mul isampling echnique, he la ge mul isampling
ac o M he longe he ime be ween he ese sampling and he signal sampling. I ende s in he esul ha
inc easing he M ac o does no educe he o al noise since he whi e noise con ibu ion educ ion by M
imes a e aging is compensa ed by he inc emen o he low equency componen s due o he la ge ime
be ween he co ela ed double sampling ope a ion.
READOUT IN CORRELATED MULTISAMPLING, CMS, MODE
The ADC noise pe o mance has been measu ed using a es ow o se he inpu o he ADC. This es ow
is consis ing o a ow o es pixels. This es pixel is based on a sou ce ollowe ansis o equal o he ac i e
pixel sou ce ollowe whose inpu and ou pu s has been sho ci cui be ween all es pixels. The inpu o he
es pixel is connec ed o a e e ence and he ou pu is connec ed o e e y column ADC. The e o e, his
measu emen is no including he pixel noise, since all he pixel noise con ibu ion is a e aged by all he es
pixel sou ce ollowe s.
Table 9 shows he noise measu emen s o e e y componen conside ing a mul isampling o M=1 o M=16.
I can be obse ed, he e ec i eness o mul isampling is educed along he mul isampling ac o M, due o
he low equency noise componen s a e no so e ec i ely educed when inc easing M.
3.7 Readou wo king in co ela ed mul isampling wi h ex e nal CDS _
83
Table 9. NOISE MEASURE IN CHARACTERIZATION FOR CORRELATED MULTISAMPLING WITH EXTERNAL CDS MODE
Readou mode
M
Readou Noise one
con e sion
Readou Noise a e
CDS
To al Readou Noise
[bi s]
[V ms]
[V ms]
CG=100V/e [e]
14 CMS
1
73.60
89.04
0.89
14 CMS
2
69.03
64.80
0.65
14 CMS
3
59.88
51.88
0.52
14 CMS
4
54.39
47.20
0.47
14 CMS
5
47.00
43.47
0.43
14 CMS
6
46.69
41.15
0.41
14 CMS
7
45.47
39.52
0.40
14 CMS
8
41.89
37.14
0.37
Chap e 3 Two s ages ADC o low noise CMOS Image Senso s
84
3.8 READOUT CALIBRATION ON-CHIP
This sec ion is in blank because i is con aining con iden ial in o ma ion.
85
CHAPTER 4
4. TWO STAGES ADC EXTENSION FOR DUAL
CONVERSION GAIN HDR
ABSTRACT
This chap e desc ibes he ex ension o he wo s ages ADC desc ibed in he p e ious chap e in o de o
digi alize he wo componen s coming om he dual con e sion gain pixel a chi ec u e and ope a ion. Dual
con e sion ope a ion is a well-known echnique implemen ed o inc ease he dynamic ange o e he
limi a ion coming om ei he he inpu ull scale o he ADC o he maximum ol age swing in he sense node.
The dual con e sion ope a ion, desc ibed in Sec ion 4.1.2.2, i p o ides wo pixel da a:
• he low gain componen and
• he high gain componen .
The ADC a chi ec u e p esen ed is an ex ension he wo s a ed ADC o op imize he e iciency o he
con e sion based on he assump ion o he low gain componen does no need low noise con e sion, since
he low gain componen is con aining he noise con ibu ion coming om he sho noise. The op imiza ion
consis s o educing he a ea, he powe consump ion and ime needed o con e s he high gain componen
wi h low ADC noise and high p ecision, and he low gain componen wi h enough noise le el o no a ec o
he o al signal o noise a io (SNR).
This chap e is o ganized as ollows. Sec ion 4.1 p esen s he s a e-o - he-a e iew o high dynamic
ange echniques and he eason o dual con e sion gain selec ion. Sec ion 4.2 is p esen ing he dual
con e sion gain pixel and he pixel pe o mances o se e al a ian s in a es chip. Finally, he Sec ion 4.3
desc ibes he ADC a chi ec u e used o con e he wo componen s om he dual gain pixel, and he
op imiza ion done o pe o m low noise and high Full Well Capaci y (FWC) wi h high e iciency in e ms o
powe , a ea and iming.
Chap e 4 Two s ages ADC ex ension o dual con e sion gain HDR
86
4.1 HIGH DYNAMIC RANGE STATE-OF-THE-ART REVIEW
The na i e Dynamic Range (DR) o s anda d CMOS Image Senso s (CIS), i.e. he ange cap u ed o s ill images
and assuming ha no DR enhancemen echnique is employed, is a ound 60-65 dB, al hough i is dependen
on he pixel size and he maximum cha ge he senso can accumula e [Teled23]. Howe e , he luminance
dynamic ange o he na u al wo ld is much la ge ; indeed, he human eye can cap u e his dynamic ange
because i can adjus i s esponse o a wide ange o ligh condi ions. Na u al scenes o en span a DR o 90
dB and o e . High Dynamic Range (HDR) scenes a e commonplace in ou doo applica ions. Fo ins ance, in
au omo i e, la ge dynamic ange is undamen al o ecognize a ic signals, pedes ians, and oad signals in
ex eme condi ions o luminescence, like when going ac oss a unnel o nigh a ic ligh s. Indoo
applica ions may also in ol e HDR scenes; o ins ance, in scien i ic applica ions, i is p e y s anda d o ha e
e y b igh a eas coexis ing in he same scene wi h dimly illumina ed a eas. Simila si ua ions a e
encoun e ed in machine ision applica ions, among many o he s.
The dynamic ange e e s o he illuminance in e al ha senso s can adequa ely cap u e. Limi a ions
impac ing his in e al on he low side and high side a e, espec i ely:
• on he low side, he minimum illuminance is limi ed by he da k cu en noise,
• on he high side, he maximum illuminance is limi ed by he amoun o cha ge he pixel can in eg a e
and ead ou .
Rigo ously speaking, enhancing DR equi es ac ing on bo h sides o he ange. Howe e , some comp ession
echniques ocus only on he high side, hus masking iny con as in he iny illumina ed egions.
Se e al HDR echnologies ha e been p oposed o enhance CIS capabili ies o cap u e he highligh s and
shadows simul aneously and he eby inc ease hei ope a ing illumina ion ange. This sec ion summa izes a
quali a i e o e iew o di e en HDR echniques a ending o he ollowing h ee main pe o mance
pa ame e s:
1) Da k condi ions pe o mances, da k noise.
2) DR, quan i ied as he a io be ween he maximum and minimum de ec able illuminances
3) SNR, Signal o Noise Ra io: quan i ied as he a io be ween signal and he noise along he comple e
illuminance ange.
Depending on he applica ion, ex ending he dynamic a ending in one, wo, o he h ee cen al
pe o mances lis ed abo e migh be necessa y. Fo example, DR i sel is essen ial o isual applica ions since
i allows us o ecognize scenes wi h high con as . Howe e , bo h DR and da k noise le els a e ele an in
scien i ic applica ions like mic oscopy because low illuminance condi ions a e applied when wo king wi h
biological samples, o example. Fu he mo e, in applica ions like g ound as onomy, whe e he same scene
migh include dimly illumina ed pixels co esponding o As os su ounded by high-illuminance egions, he
h ee pe o mances men ioned abo e a e wo h add essing o de ec and isualize he ele an objec s and
dis inguish hem om he su ounding backg ound.
4.1.1 Mul iple exposu es HDR me hods
A di ec app oach o DR enhancemen consis s o me ging mul iple exposu es cap u ed wi h s anda d (low
dynamic ange) image senso s in o a single linea HDR image, which has a much la ge dynamic ange han a
single exposu e image. Fig. 80 illus a es his concep .
The i s image, Fig. 80 (a), co esponds o a sho exposu e ime, and he second, Fig. 80 (b), shows he
same scene bu wi h a longe exposu e ime. Hence, dimly illumina ed egions a e be e de ined in he la e
case a he cos o sa u a ing he highly illumina ed a eas. On he con a y, he o me image be e
ep esen s he highly illumina ed a eas because he sho e exposu e p ecludes sa u a ion bu hides de ails
4.1 High Dynamic Range s a e-o - he-a e iew _
87
o he low illumina ed egions. Fig. 80(c) shows he ou come o combining hese wo images o ob ain a HDR
image.
Once he HDR image is me ged, he numbe o bi s needed o ep esen he da a mus inc ease as,
oughly speaking N ~ DR/6, whe e N is he numbe o bi s and DR is he dynamic ange in dBs. Howe e , his
o mula co esponds o a linea encoding ule ha would o ce using many bi s in he eadou channel, which
is imp ac ical due o a ea occupa ion and powe consump ion. Hence, nonlinea encoding echniques a e
ad isable in his scena io. Exp essly, a nonlinea ans o ma ion p ocess called one mapping pe mi s he
con e sion o composed images back o he o iginal image o ma and enables he display o hese composed
images using he usual 8-bi displays.
TONE MAPPING deals wi h educing he la ge con as om he eal scene o he displayable ange
while p ese ing he image de ails and colo appea ance, keeping he impo an in o ma ion o
app ecia e he o iginal scene con en .
(a) (b)
(c)
HIGH DYNAMIC RANGE THROUGH MULTIPLE EXPOSURES. (A) SHOT EXPOSURE CAPTURE; (B) LONG EXPOSURE CAPTURE ;
(C) TONE MAPPED COMBINED IMAGE.
Se e al one mapping algo i hms ha e been de eloped in he las decades
[Debed07][Me e07][Robe 99]. They a e based on he esponse o he image senso along he exposu e
ange o combine se e al exposu e images, enhance he de ails, and comp ess he de ailed in o ma ion.
Some con ibu ions ha e also add essed inco po a ing one mapping di ec ly a he pixel le el based on
nonlinea ligh - o- ime con e sion [Va ga14] wi h single exposu e and one mapping based on p e ious
ame his og am e alua ion.
Chap e 4 Two s ages ADC ex ension o dual con e sion gain HDR
88
Fig. 81 ep esen s he in e esponse unc ion o wo algo i hms:
• The one a he le was epo ed by Debe ec in [Debed07] ;
• The one a he igh was epo ed by Robe son in [Robe 99].
These g aphs show ha he maximum in ensi y is ep esen ed by he maximum displayable code, 255, and
apidly decay o comp ess he high le els o in ensi y, un il i eaches almos linea beha iou o low le els
o in ensi y.
(a) (b)
INVERSE CAMERA RESPONSE FUNCTION FROM :(A) DEBEVEC AND (B) ROBERTSON ALGORITHMS
In [Rod i11] an algo i hm is p esen ed based in a his og am c ea ed du ing exposu e ime o de e mine
he shape o he one-mapping cu e which is applied o c ea e he inal image as shown in Fig. 82. This
algo i hm is e y sui able o ha dwa e implemen a ion in Focal Plane P ocesso s (FPP) since i only equi es
a compa a o , educed local memo y and small logic pe pixel. The his og am adap s o he ligh condi ions
in only one ame. The e o e i is e y use ul o HDR eal- ime came as unde mode a e speed changing
demands on he scene. E e y pixel is codi ied wi h 4-bi s His og am da a plus 7-bi s o he one mapped
image.
HISTOGRAM AND TONE MAPPED IMAGE
4.1 High Dynamic Range s a e-o - he-a e iew _
89
Howe e HDR echniques based on mul iple exposu es a e no well sui ed o applica ions whe e scene
objec s mo e be ween sho s, since images wi h di e en exposu es mus be s o ed and aligned. Addi ional
d awbacks appea when illumina ion condi ions change be ween he di e en exposu es.
The main pe o mances o his HDR echnique a e gi en in Table 1.
Table 1. MAIN PERFORMANCES IN MULTIPLE EXPOSURE HDR TECHNIQUE
Da k Noise
Dynamic ange, DR
Signal o Noise a io, SNR
Da k Noise pe o mances is
gi en by image senso used.
The inc eamen in he dynamic
ange is gi en by he a io
be ween exposu e ime(Tin ):
∆𝐷𝑅=𝑇𝑖𝑛𝑡𝑙𝑜𝑛𝑔
𝑇𝑖𝑛𝑡𝑠ℎ𝑜𝑟𝑡
The SNR is no ealy enhanced wi h
his echnique. The maximum SNR
will come om he maximum
pho ons ha can be accumula e in
he pixel, ei he in he long o sho
exposu e
Fo example, conside ing an image senso de ice wi h 2 e-RMS o da k noise and 10 Ke- FWC (Full Well
Capaci y), he na i e dynamic ange would be, DR = 73.98 dB. I we use he mul i exposu e echnique o
ex end he dynamic ange, wi h a exposu e in eg a ion a io o 8 he e ec i e dynamic ange would be DR =
92 dB, in case inc easing o a io 16 he dynamic ange is inc eased in 6dB, DR = 98 dB. Howe e , no only
he signal bu also he sho noise is ampli ied by he a io o exposu e imes, and i has an e ec in he SNR
cu e in he in lec ion poin when he a io change is applied he noise is ampli ied as well and he SNR d ops
ab up ly. Fo his easons, mul i exposu e wi h numbe o exposu es highe han wo is equen ly used o
ex end he dynamic ange wi h high a ios o exposu es. Again, i adds e en mo e complexi y o he p oblem
o he mo ing objec s in he scenes du ing sequen ial in eg a ions.
SNR CURVE FOR THE MULTIEXPOSURE HDR METHOD, RATIOS 8 AND 16
4.1.2 Mul iple gains HDR me hod
As an al e na i e o he mul i-exposu e me hod, se e al au ho s ha e explo ed DR ex ension by applying
o he me hods, such as o ins ance:
• mul iple gains echniques
• loga i hmic comp ession a pixel le el,
• mul i-sampling pixel-le el ADCs,
Chap e 4 Two s ages ADC ex ension o dual con e sion gain HDR
96
Fig. 91 shows he SNR o wo cases o Spli pho odiode case, one wi h a a io o sensi i i y o 8 and o he
wi h a a io he sensi i i y o 16. When he a io is highe , he gain o eco e he da a is highe and he sho
noise om he small pho odiode becomes highe in oducing highe d op in he SNR cu e.
SNR CURVE FOR SPLIT PHOTODIODE PIXEL
4.1.3 Loga i hmic pixel
Fig. 92 shows a loga i hmic pixel a chi ec u e combining he 3T pixel wi h a s anda d column eadou ci cui .
The ansis o M1 wo ks is connec ed in a diode con igu a ion and hence wo ks as a esis o . Thus, i con e s
he pho ogene a ed cu en in o a ol age pd. Assuming ope a ion in he weak in e sion egime whe e he
cu en i s an exponen ial unc ion o he ol age, his esis o has a loga i hmic cha ac e is ic [Enz06].
Hence, he ol age pd is p opo ional o he loga i hm o he d ain cu en , and he dynamic ange ge s
comp essed in he ol age ou pu . Indeed, his ol age mus emain low enough o gua an ee he M1 keeps
wo king in he weak in e sion egime, limi ing i s excu sion. Owing o he nonlinea dependence, digi al
p ocessing is needed o linea ize he pixel ou pu . Fu he mo e, he esis o cha ac e is ics depend on locally
a ying echnological pa ame e s; his ea u e means ha senso s designed acco ding o his p inciple may
equi e dedica ed echniques o co ec he bidimensional Fixed Pa e n Noise (FPN).
Ano he way o p oducing a loga i hmically comp essed ol age is by using pho o-diodes in he
pho o ol aic egime an using he open-ci cui ol age o encode ligh in ensi y [Fe na24]. Ei he case, This
me hod can be used in olling shu e mode. Du ing each ow eadou he pixel ol age is ead h ough he
LOGARITHMIC PIXEL
4.1 High Dynamic Range s a e-o - he-a e iew _
97
Loga i hmic comp ession enla ges he numbe o pho ons he pixel can abso bed be o e sa u a ion,
achie ing alues la ge han 120 dB. One d awback o his echnique is high ead noise due o 3T
a chi ec u e. The ixed pa e n noise is also much highe han s anda d 3T o 4T CIS due o sub- h eshold
ansis o ope a ion M1, gene a ing non-linea a ia ions om pixel o pixel, limi ing his a chi ec u e o
monoch ome applica ions.
Table 6 summa izes he p ima y pe o mance o his HDR echnique
Table 6. MAIN PERFORMANCES IN LOGARITHMIC PIXEL HDR TECHNIQUE
Da k Noise
Dynamic ange, DR
Signal o Noise a io, SNR
Da k Noise pe o mances is poo
since i is simila o 3T pixel
including ese noise (kTC) as
majo con ibu o .
The inc eamen in he dynamic
ange is gi en by maximum
pho ocu en can be in eg a ed
be o e sa u a ing and he
minimum pho ocu en equi alen
o ead noise, 𝑣𝑛. K s ands o
M1 sub h eshold e iciency.
∆𝐷𝑅=𝑖𝑚𝑎𝑥
𝑖𝑚𝑖𝑙 =𝑒(𝑣𝑑𝑑−𝑣𝑡ℎ)𝐾/𝑣𝑡ℎ
𝑒(𝑣𝑑𝑑−𝑣𝑛)𝐾/𝑣𝑡ℎ
The SNR maximum will come om
he maximum pho ocu en ha can
be abso ed be o e eaching
sa u a ion, imax.The shape o he
SNR is gi en by he g aph bellow, i
shows is almos plain o highe
alues.
Fig. 93 illus a es he SNR o he loga i hmic pixel whe e i is showed ha o he highe pa o he SNR
cu e is almos la , since he majo pa o he econ e ed pho ons a e d ained o he supply h ough he
M1 ansis o in sub h eshold s a e.
SNR CURVE FOR LOGARITHMIC PIXEL
4.1.3.1. HDR BY ADJUSTING WELL CAPACITY
The well-capaci y adjus men echnique consis s o adjus ing he equi alen capaci y o he loa ing di usion
node h ough a la e al o e low cha ge pa h ha d ains cha ge in excess in ha node, hus p ecluding ol age
sa u a ion [Yang99b] [Decke98]. This la e al o e low cha ge pa h can be implemen ed by using he ese
ansis o and adequa ely con olling he ol age a he ese ansis o ga e e minal – node RST in Fig. 94 .
The ol age applied a his ga e e minal de e mines he well-capaci y. Du ing he in eg a ion, he well-
Chap e 4 Two s ages ADC ex ension o dual con e sion gain HDR
98
capaci y is changed using his con ol ol age o d ain he excess pho ogene a ed cha ge ia his o e low
ga e, clipping he collec ed cha ge o ha le el. A e some ime, he well capaci y is inc eased, he ol age
is mo ed down, and he pho odiode can now in eg a e mo e cha ge in o he memo y node. M2 ga e ol age
es ablishes a po en ial ba ie o he elec on low. When pho o-cha ge a e accumula ing on he sense node,
he cha ge le el ises and i he cha ge le el exceeds he ba ie le el, he ex a cha ge lows o he d ain.
The dynamic ange is inc eased by mo ing down he ese ga e ol age o e he in eg a ion pe iod, as shown
in Fig. 95. Fo low illumina ion, he in eg a ed cha ge is unal e ed by he ba ie when dec easing VR, and
hen he pixel collec s all he pho o cha ges in he sense node. Fo high illumina ion, pho o-cha ges go in o
he sense node and exceed he ba ie going o he M2 d ain du ing he ime in e als de ined by he ime
ins an s in ol ed in he algo i hm, namely: [ 0 , 1, ]; [ 1 , 2]; e c.
PIXEL SCHEMATICS
The inal in eg a ed cha ge is educed be ween he dashed and solid line. As illumina ion inc eases, a
g ea e p opo ion o he pho ocu en is d ained o he supply. Depending on he de ined s eps o VRST and
iming, any a bi a y comp ession cha ac e is ic can be achie ed, gene a ing he pixel ol age showed in Fig.
96.
GATE VOLTAGE WAVEFORM AND CHARGE INTEGRATION CURVES FOR LOW AND HIGH ILLUMINATION
The well adjus ing HDR echnique ex ends he dynamic ange o he pixel c ea ing a comp ession non-
linea pixel esponse, consis ing in a piece-wise cu e due o he mechanism o d aining cha ge due o he
mo ing ba ie c ea es h ough he la e al o e low ansis o achie ing alues la ge han 100 dB. One p o
4.1 High Dynamic Range s a e-o - he-a e iew _
99
o his echnique is he s aigh o wa d implemen a ion and compa ibili y wi h classical pixel con ol and
esponse. Coun e s include high ead noise since he cha ge is in eg a ed in he sense node. Besides, FPN
may be la ge han ypical 4T ansis o since a ia ions in la e al o e low ansis o may change he
comp ession cu e om pixel o pixel.
Fig. 96 shows he main pe o mances o his HDR echnique. Fig. 97 shows he SNR he well adjus men
echnique, whe e he SNR d ops in he in lexion poin s as consequence o d aining he cha ge.
PIXEL OUTPUT RESPONSE
Table 7. MAIN PERFORMANCES IN WELL ADJUSTING HDR TECHNIQUE
Da k Noise
Dynamica ange, DR
Signal o Noise a io, SNR
Da k Noise pe o mances is poo
since i is simila o 3T pixel
including ese noise (kTC) as
majo con ibu o .
The inc eamen in he dynamic
ange is gi en by maximum
pho ocu en can be in eg a ed
when applying he well adjus men
and he maximum pho ocu en in
linea s anda d mode.
∆𝐷𝑅=𝐼𝐿max(𝑤𝑎)
𝐼𝐿max(𝑙𝑖𝑛)
The SNR maximum will come om
he maximum pho ocu en ha can
be abso ed in linea mode. As he
cha ge is d ained when exceeds he
ba ie he SNR d ops in he
in lexion poin s.
SNR CURVE FOR WELL ADJUSTMENT TECHNIQUE
Chap e 4 Two s ages ADC ex ension o dual con e sion gain HDR
100
4.1.4 Techniques based on pe -pixel ADCs
4.1.4.1. MULTIPLE SAMPLED PIXEL LEVEL ADC
This a chi ec u e combines a pho odiode wi h a Sample and Hold (S&H) plus ADC ha con e s e e y sample
du ing he in eg a ion. As example, a single-slope ADC a pixel and a simple digi al eadou ci cui a column
le el, as showed in Fig. 98(a). The signal amp is he e e ence o he ADC which is sha ed be ween pixels.
The senso can be eadou se e al imes du ing each in eg a ion pe iod i he ame a e is low enough. This
allows he ue co ela ed double sampled and he ex ension o dynamic ange [To su16] [Yang99a].
Fig. 98(b) shows an example o he senso ’s esponse wi h mul iple sampling and exponen ial inc easing
exposu e imes: T, 2T, 4T …2KT. Each sample pixel ou pu is digi ized o m bi s, and hen hese alues a e
combined o gene a e m+k numbe wi h highe esolu ion, and hen ex ension o dynamic ange.
(a) (b)
MULTIPLE SAMPLED PIXEL LEVEL ADC PIXEL
This a chi ec u e equi es a signi ican amoun o memo y and digi al signal p ocessing o p oduce HDR
image da a. In addi ion, i should s o e all he ames cap u ed du ing an in eg a ion pe iod and pe o m pixel
by pixel p ocessing o minimize noise and maximize DR. P os o his a chi ec u e include:
• he possibili y o inc ease dynamic ange o e 100 dB,
• he peaks in he SNR is small and
• he linea i y is good.
Coun e s include high cos , la ge pixel equi ed and high powe consump ion.
Table 8 summa izes main ea u es o his echnique. Fig. 99 shows he SNR o mul iple sampled pixel
echnique, whe e he SNR d ops e e y ime he exposu e ime is mul iplied by w ( wo).
Table 8. MAIN PERFORMANCES IN MULTIPLE SAMPLED PIXEL HDR TECHNIQUE
Da k Noise
Dynamica ange, DR
Signal o Noise a io, SNR
Da k Noise pe o mances may
be poo since ese noise (kTC)
is included.
The inc eamen in he dynamic
ange is gi en by a io be ween
he maximum exposu e ime and
he minimum (s anda d) one.
∆𝐷𝑅=2𝐾
The SNR maximum will come om
he maximum pho ocu en ha can
be abso ed in e e y con e sion.
4.1 High Dynamic Range s a e-o - he-a e iew _
101
SNR CURVE FOR MULTIPLE SAMPLED PIXEL LEVEL ADC
4.1.4.2. PIXEL LEVEL SIGMA DELTA ADC
The sigma del a ADC pe pixel wi h esidue eadou h ough sou ce ollowe is a HDR a chi ec u e ha
combines a sigma del a ADC wi h a sou ce ollowe ou pu wi hin he pixel and a mixed mode analog and
digi al column eadou ci cui . I is a case o mul iple sampled pixel le el, whe e he pixel is con e ed a he
same ime i is sampled. The column eadou ci cui y p ocesses he sou ce ollowe analog signal ou pu and
he digi al ou pu o he digi aliza ion and eadou o he one bi pixel memo y.
Fig. 100 shows he schema ic o he sigma del a ADC pixel wi h esidue eadou in pixel. The pixel cha ge
is in eg a ed om he pho odiode on o C b using a Capaci i e T ans-Impedance Ampli ie (CTIA). Then he
ou pu in eg a ed ol age su pass he h eshold max he compa a o oggles and i is la ched by he 1 bi
memo y wi h he clock clk ising edge. Addi ionally, when he memo y ou pu is 1, cha ge is dumped in o C b
om Cdump o cing he ou pu o CTIA o all. Clk is a global signal o all pixels, s o ing one bi e e y ime his
ope a ion occu s. The pixels a e ead ou mul iple imes du ing each in eg a ion pe iod. Du ing each eadou ,
excep he las one, only he digi al memo y is ead ou . In he inal eadou , he digi al ou pu and analog
ou pu om sou ce ollowe a e ead ou . The numbe o ones ha a e eadou ou du ing one in eg a ion
ime is equi alen o FWC·Nones cha ges ha was collec ed by he pho odiode. The ol age a he ou pu o
he SF is he esidue ha is less han he FWC o he pixel. This a chi ec u e equi es digi al memo y pe pixel
o hold all o he bina y digi al pixel ou pu s du ing one in eg a ion pe iod, and i also equi es digi al signal
p ocessing o da a o c ea e he inal image da a and co ec o se and gain.
SIGMA DELTA ADC WITH RESIDUE READOUT IN PIXEL
Chap e 4 Two s ages ADC ex ension o dual con e sion gain HDR
102
This implemen a ion allows cap u ing da a in a global shu e mode. A he beginning o each ame all o
he cha ge in eg a o s in he a ay a e ese in pa allel by ac i a ing RST1 swi ch. E e y clock pe iod one
bina y image is coming om he a ay, and a e N-1 bina y images he inal bina y image along wi h he
analog esidue is eadou .
P os o his a chi ec u e a e high dynamic ange highe han 100 dB, excellen linea i y, la ge pho ocu en
can be in eg a ed, and SNR inc eases mono onically as pho ocu en inc eases. The disad an ages o his
a chi ec u e include: la ge pixel size, low ill ac o (al hough i can be elaxed using 3D s acking echnology
[Kagaw19]) and poo low ligh pe o mance due o he lack o ue CDS and high-powe consump ion. Main
pe o mances and SNR cu e a e simila o he mul iple samples pixel since same HDR echnique is used.
4.1.5 Time o sa u a ion pixel
The ime o sa u a ion pixel wi h esidue eadou is a combina ion o a ime o sa u a ion pixel wi h wo
sou ce ollowe s:
• one o ime o sa u a ion digi aliza ion ;
• ano he o he esidue eadou digi alized wi h an analog column eadou ci cui .
The column ci cui y is p ocessing he da a om he wo sou ce ollowe s, and hen digi ized and combined
in o he inal high dynamic ange pixel da a. Fig. 101 shows he schema ic o ime o sa u a ion pixel wi h
esidue eadou . The pixel in eg a es di ec ly he cha ge in o he loa ing di usion node, when he ol age
goes bellow a h eshold min he compa a o oggles and he ol age on he amp signal amp is s o ed on a
sampling capaci o . Then he sa u a ion ime and ol age on he pho odiode a e eadou , and he wo alues
should be combined digi ally o c ea e he HDR ou pu da a.
The ime o sa u a ion da a can be collec ed in global shu e mode, bu he esidue eadou should be
done in olling shu e mode. The amp signal is ising om he beginning o he exposu e ime and eaches
he maximum alue a he end o he exposu e ime.
TIME TO SATURATION PIXEL WITH RESIDUE READOUT
The dynamic ange is inc eased by he a io be ween he in eg a ion ime, Tin , and he unce ain y in he
sa u a ion ime, σsa . σsa is he s anda d de ia ion o he ime in oduce by he noise in oduced by he
compa a o , he amp gene a o and he es o he eadou ci cui y.
One p o o his a chi ec u e is he e y high dynamic ange, g ea e han 150 dB. Coun e s include:
• high powe dissipa ion,
• la ge pixel and low ill- ac o ,
• he SNR alls a high illumina ion le els, and
• no good low ligh pe o mances.
4.1 High Dynamic Range s a e-o - he-a e iew _
103
Table 9 o e iews main ea u es o his HDR echnique
Table 9. MAIN PERFORMANCES IN TIME TO SATURATION PIXEL HDR TECHNIQUE
Da k Noise
Dynamica ange, DR
Signal o Noise a io, SNR
Da k Noise pe o mances may
be poo since ese noise (kTC)
is included.
The inc eamen in he dynamic
ange is gi en by a io be ween
he in eg a ion ime and he
unce ain y in he sa u a ion ime.
∆𝐷𝑅=𝑇𝑖𝑛𝑡
𝜎𝑠𝑎𝑡
The SNR maximum will come om
he maximum FWC o he
pho odiode, conside ing de ined by
min ol age. A e ha , he SNR
s a s slowly decaying due o he
unce ain y in he sa u a ion ime
Fig. 102 shows he SNR o ime o sa u a ion pixel echnique. In his echnique, he e a e wo
di e en ia ed pa s: one whe e he da a is coming om he sou ce ollowe M3, and he second whe e he
da a is coming om he sa u a ion compa a o ci cui y. The i s pa he SNR is equi alen o s anda d 3T
pixel. In he second pa , he SNR does no inc ease since he numbe o elec ons in he pixel is always he
same and he ela ed sho noise, howe e he SNR s a s decaying due o he unce ain y in he sa u a ion
ime ha , when i is ansla ed in elec ons, he noise inc ease wi h he ligh in ensi y.
SNR CURVE FOR TIME TO SATURATION PIXEL
4.1.6 HDR echniques compa a i e
Table 10 shows he summa y o all HDR echniques commen ed in his sec ion, whe e i is shown he
inc emen o dynamic ange along wi h he main limi a ion in he signal o noise a io.
Table 10. HDR TECHNIQUES COMPARATIVE TABLE
HDR echnique
∆𝑫𝑹
SNR
Commen s
Mul iple exposu e
𝑇𝑖𝑛𝑡𝑙𝑜𝑛𝑔
𝑇𝑖𝑛𝑡𝑠ℎ𝑜𝑟𝑡
Limi ed by pixel
FWC
Complexi y when objec s a e
mo ing in he scene
Mul iple gain
Dual gain column ampli ie
𝜎𝑑𝑎𝑟𝑘_𝑙𝑜𝑤𝑔𝑎𝑖𝑛
𝜎𝑑𝑎𝑟𝑘_ℎ𝑖𝑔ℎ𝑔𝑎𝑖𝑛
Limi ed by pixel
FWC
Dynamic ange inc eased by
educing he da k noise wi h high
gain
Chap e 4 Two s ages ADC ex ension o dual con e sion gain HDR
104
Mul iple gain
Dual con e sion gain
𝐶𝐺𝑙𝑜𝑤𝑔𝑎𝑖𝑛
𝐶𝐺ℎ𝑖𝑔ℎ𝑔𝑎𝑖𝑛
Limi ed by pixel
pho odiode FWC
Dynamic ange inc eased by
educing he da k noise wi h high
gain and inc easing FWC limi ed by
ol age ange in he sense node
Mul iple gain
Dual con e sion gain wi h
LOFIC
𝐹𝑊𝐶𝐶𝑆𝑙𝑜𝑤_𝑔𝑎𝑖𝑛
𝐹𝑊𝐶ℎ𝑖𝑔ℎ_𝑔𝑎𝑖𝑛
Limi ed by ange
in sense node in
low gain.
Dynamic ange is no limi ing by he
pho odiode, i is limi ed by ange in
sense node in low con e sion gain
Mul iple gain
Spli pho odiode
𝑠𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦𝑙𝑎𝑟𝑔𝑒
𝑠𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦𝑠𝑚𝑎𝑙𝑙
Limi ed by pixel
FWC om la ge
pho odiode
Dynamic ange is inc eased by he
a io be ween small and la ge
pho odiode
Loga i hmic pixel
𝑖𝑚𝑎𝑥
𝑖𝑚𝑖𝑙 =𝑒(𝑣𝑑𝑑−𝑣𝑡ℎ)𝐾/𝑣𝑡ℎ
𝑒(𝑣𝑑𝑑−𝑣𝑛)𝐾/𝑣𝑡ℎ
Limi ed by
h eshold ol age
in diode ansis o
Dynamic ange is limi ed by he
h eshold ol age in diode ansis o .
Non linea unc ion in he esponse.
Well capaci y adjus men
𝐼𝐿max(𝑤𝑎)
𝐼𝐿max(𝑙𝑖𝑛)
Limi ed by pixel
FWC
Full well is changed du ing
in eg a ion ime .
Mul iple sampled pixel
le el ADC
2𝐾
Limi ed by pixel
FWC
Mul iple sampled pixel allows o
ha e se e al in eg a ions alues.
Pixel le el sigma del a
ADC
2𝐾
Limi ed by pixel
FWC
Mul iple sampled pixel allows o
ha e se e al in eg a ions alues.
Time o sa u a ion pixel
𝑇𝑖𝑛𝑡
𝜎𝑠𝑎𝑡
Limi ed by pixel
FWC
The ime o sa u a ion is con e ed
along wi h he nominal ou pu .
Based on his compa a i e analysis, he dual con e sion gain echnique was selec ed o implemen a ion
because:
• he a ge dynamic ange is be ween 95 dB and 100 dB and o his le el he e is no limi a ion in he
pho odiode, conside ing he pixel pi ch 10um.
• he dual con e sion gain allows achie ing high le el o SNR which is also impo an o scien i ic and
as onomy applica ions.
4.2 PIXEL ARCHITECTURE
As said abo e, he pixel pho odiode will no ep esen an obs acle o achie e 100 Ke- FWC and a dynamic
ange o 100 dB when he eadou channel is p esen ing 1 elec on noise. A es chip has been designed o
alida e he design me hods, es and op imize he se e al la o s o he pixel and hen ob ain he op imum
sizzing o he de ices inside he pixel along wi h he op imiza ion o he p ocess in o de o minimize he
noise and ex end he maximum capaci y o e he 100 Ke-.
4.2.1 Pixel schema ics and wa e o ms
Fig. 103 shows he pixel and Fig. 104 shows he co esponding wa e o ms. Ope a ion is acco ding o he
sequence lis ed in he numbe ed poin s below.
1) Fi s , he signals RST, TX and GAIN a e ac i a ed o emo e all he cha ge om he pho odiode
(pho odiode ese ) and o se he ese le el in A and B node, i is in CFD and CS capaci o s.
2) A e ha , he exposu e s a s wi h he alling edge o he TX signal. F om ha momen on, all he
pho ons ha a e collec ed in he pho odiode a e s o ed as elec on cha ges.
4.2 Pixel a chi ec u e _
105
3) Jus be o e he eadou ope a ion, he RST and GAIN signals a e ac i a ed in o de o sample he
ese ol age in CFD and CS capaci o s.
DUAL CONVERSION GAIN PIXEL IN THIS WORK
DUAL CONVERSION PIXEL WAVEFORMS
4) The eadou ope a ion s a s ac i a ing he SEL ansis o ha connec he pixel ou pu o he
column eadou . The eadou ope a ion is a ollows:
a. The ese s o ed in he node A+B, CFD + CS. This is he ese le el ela ed o he low con e sion
gain (LGC RST)
b. A e ha , he RST signal goes down and he ese le el s o ed in he loa ing di usion is
ead. This le el o ese is ela ed o he high con e sion gain (HCG RST).
c. Once he wo le els o ese ha e been ead, he ans e pulse (TX) is applied and he
pho oelec on cha ge is s o ed in he CFD. Now, he signal le el ela ed o he high gain is
s o ed (HCG SIG). When applying co ela ed double sampled echnique, he ese noise
sampled in he loa ing di usion is sub ac ed and he conce ned noise is o ally elimina ed.
In his way, he pixel signal eadou o high gain is also low noise as a ypical 4T app oach. IN
GAIN
RST
TG
SEL
RST_LG RST_HG SIG_HG SIG_LG
Baja iluminación
Al a iluminación
ReadOu
Exposu e
RST
GAIN
CS
CFD
112
113
CHAPTER 5
5. LOW NOISE CISS WITH OPTIMIZED
PIXELS,,DUAL GAIN AND OVERSAMPLED ADCS
ABSTRACT
This chap e desc ibes wo CMOS image senso chips ha demons a e he a chi ec u es and me hods
desc ibed in p e ious Thesis´s chap e s. Besides, hese chips ha e eached indus ial ma u i y up o TRL 9
ollowing se e al p o o ype ab ica ion e-spins.
The i s o hese chips embeds wo p ima y con ibu ions o he Thesis, namely
• A low noise pixel;
• A wo-s age o e sampled ADC.
Pixels and ADCs i he a chi ec u es and me hods desc ibed in Chap e 2 and Chap e 3, espec i ely. As
explained in he o me o hese chap e s, he pixel includes wo supplies o enhanced noise esponse.
Associa ed ol age d i e s o hese ol ages a e compa ible wi h he p ocedu es desc ibed in ha chap e
ega ding linea i y and noise op imiza ion. O cou se, he design o he comple e senso chip aises p oblems
ha go a beyond he each o hese p e ious chap e ma e ials. Indeed, se e al challenges ha e been aced
du ing senso implemen a ion, like he glow e ec ha limi s he da k cu en a low empe a u es and he
ow empo al noise ha should emain much lowe han he bidimensional empo al noise o a oid being
isible, among o he s desc ibed in his chap e .
The second o hese chips is a la ge- o ma low-noise de ice ha employs he dual con e sion gain pixel
op imiza ion p esen ed in Chap e 4. This image senso inco po a ed he ADC desc ibed in Chap e 3 plus he
modi ica ion desc ibed in Chap e 4 o ea u e a high dynamic ange while keeping a ea and powe demands
con ained. Implemen a ion o his senso chip aced se e al challenges ela ed o yield and wa e -scale
in eg a ion.
This chap e is o ganized as ollows. Sec ion 5.1 desc ibes he senso speci ica ions and main ea u es.
Sec ion 5.1.2 shows he senso a chi ec u e wi h all he main building blocks ha allows o ob ain he da a
om he pixel o he ou pu image in e ace. Sec ion 5.1.3 desc ibes he senso ope a ion modes whe e he
in e nal and ex e nal co ela ed double sampled is desc ibed along wi h mul isampling ope a ion, along wi h
addi ional unc ionali ies like windowing. The senso pe o mances and cha ac e iza ion a e desc ibed in
sec ions om 5.1.4 o 5.1.8 . The glow e ec and how o op imize image senso agains his e ec is
desc ibed in sec ion 5.1.9 . Sec ion 1.1 is p esen ing he o he de elopmen consis ing o la ge o ma low
noise and high dynamic ange image senso . Sec ion Fig. 142 desc ibes he senso a chi ec u e wi h he basic
blocks and he echniques used in la ge die implemen a ion. Sec ion 5.2.2 shows he ope a ion modes
a ailable in his la ge image senso ollowed by he unc ions in sec ion 5.2.3 . Yield special echniques used
in his la ge die a e desc ibed in sec ion 5.2.4 . Op ical pe o mances o he senso a e gi en in sec ion 5.2.5
. The senso has been inco po a ed in ac i e cooling came a desc ibed in sec ion 0. Finally, a low noise image
senso bench ma k compa ison is showed in sec ion 0.
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
114
5.1 LOW NOISE SENSOR WITH 5.3 MPIXEL AND 0.7ERMS DARK NOISE
5.1.1 Gene al desc ip ion
F om now on, we will e e o his CIS by he ac onym: ULN5.3: Ul a Low Noise CIS - e sion 5.3 Table 14
shows he p ima y ea u es o ULNS5.3. I is ab ica ed in Towe -Semi 180 nm CIS echnology [Towe ], whe e
he pixel has been op imized o noise and linea i y ollowing he semi-empi ical app oach in Chap e 2. This
echnology is well sui ed o machine ision, ins umen a ion, and scien i ic applica ions.
Table 14. ULN5.3 SENSOR PERFORMANCES
Pa ame e
Value
Pixel numbe [pixels2]
2,304 x 2,304
Pixel size [µm2]
6.5 x 6.5
Pixel s uc u e
4T
Shu e ypes
Rolling and Global
Row con e sion ime (TROW) [µsec.]
≥ 4.9
Mul i ROI eadou unc ion
Yes
ADC esolu ion [bi s]
14,15 and 16.
Ou pu in e ace
32 + 4 LVDS po s @
544Mbps
Full well capaci y [103 x
elec ons]
S anda d mode
15
Ex ended ull-well mode
≥ 30
Max. ame- a e [ ps]
(co esponds o TROW = 5µsec)
A ull- es. (2,304x 2,304)
88
A ROI o 2,304 x 2,048
100
Da k ixed pa e n noise [%FS]
VFPN
0.001
HFPN
0.001
B igh ield pa e n noise [%]
VFPN
< 0.2
HFPN
< 0.2
Tempo al Readou noise in
olling shu e mode. [e- ms]
14 bi s TROW ≥ 4.9µsec
≤ 1.3
16 bi s TROW ≥ 20µsec
≤ 0.9
14bi s M=16 TROW ≥ 80µsec
≤ 0.7
Tempo al Readou noise in global shu e mode wi h ex e nal
ue-CDS [e- ms. @ TROW ≥ 5µsec]
< 3.0
Dynamic Range (14b/16b/ 14b M=16) [dB]
81.24/84.44/86.6
Da k cu en [e-/pixel/sec @ 25 deg ee C]
15
Linea i y [%]
< ±1
PRNU [% ms.]
< 1
Quan um E iciency [%] o on -
side illumina ed (FSI) p ocess
@400nm
> 65
@500nm - 600nm
≥ 75
Peak @550nm
79
@700nm
≥ 65
@800nm
≥ 44
Ope a ing Consump ion [W]
< 3.4
Image lag [%]
< 0.01
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
115
ULN5.3 is a single-chip, ully digi al image senso wi h:
5.3M pixels wo king a a maximum speed o 88 ames pe second a ull esolu ion
ULN5.3 inco po a es se e al unc ions on-chip, such as:
• p og ammable exposu e ime,
• olling and global shu e ope a ion,
• windowing and lipping.
All hese unc ions a e p og ammable ia a ou -wi e Se ial Pe iphe al In e ace (SPI) s anda d in e ace.
ULN5.3´s eadou channel inco po a es he column-pa allel wo s ages ADC a chi ec u e wi h mul iple
esolu ions desc ibed in Chap e 3. The senso is op imized o applica ions equi ing e y low noise; i can
p o ide sub-elec on empo al noise a ull speed in olling shu e modes and 2.5-elec on empo al noise
wo king in global shu e mode. The noise da a in Table 14 co espond o he eadou noise pe o mance
based on he p og amming eadou mode.
ULN5.3 ou pu in e ace is based on 32 Low Vol age Di e en ial Signal (LVDS) da a po s. Fou ex a LVDS
po s a e used o ou pu ing eco e y clock and synch oniza ion da a in pa allel. A chi ec u al and physical
design is made o compa ibili y wi h:
• F on Side Illumina ion (FSI) senso implemen a ion.
• Back Side Illumina ion (BSI).
Fig. 108 shows he main di e ences be ween he illumina ion op ions a he concep ual le el. In he FSI
de ice he incoming ligh go h ow he me als up o each he silicon su ace and en e s in o he pho odiode
sensi i e a ea, while in he BSI de ice he ligh is coming di ec ly o he silicon su ace o en e in he
pho odiode. The back side illumina ed e sion is ob ained om he same wa e whe e he FSI de ice is
ab ica ed. The i s s ep o ob ain he BSI senso is o bond he de ice wa e o a suppo wa e ha can be
made by glass o silicon on o he op su ace o he senso wa e . Then, a back hinning p ocess is ealized o
educe wa e hickness o a alue ha anges in he [4m, 10m] in e al. A e ha , he PADs a e c ea ed
on o he o iginal bond pads and hey a e in e connec ed using h ough silicon ias. To do he h ow silicon
ias, he emaining [4m, 10m] a e e ched in he ia egion and hen co e ed wi h me al. Finally, addi ional
p ocess like me al g id deposi ion, colo il e s a ay and mic o-lens deposi ion should be done. To make a
design compa ible FSI and BSI he e should be some ules o be me as o example he dis ance be ween
PADs should be la ge han 150 m in he cu en echnology.
FSI AND BSI IMAGE SENSOR CONCEPTS
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
116
Fig. 109 shows he mic o-pho og aphs o ULN5.3 FSI and ULN5.3 BSI, espec i ely.
(a) (b)
PHOTO OF ULN5.3 FSI (A) AND ULN5.3 BSI (B)
BSI echnology has he ad an age o p o iding be e quan um e iciency and ill ac o , since he e a e
no me als blocking he ligh . Addi ionally o he ad an ages o BSI is he sensi i i y o UV wa e o ms can be
much mo e highe since hese ene ge ic wa e o ms a e ecombined e y close o he su ace and hen ou
o he silicon egion in FSI de ices. Fig. 110 shows he quan um e iciency o he ULN5.3 de ice, conside ing
he ill ac o o he pixel and he quan um e iciency o silicon. I is obse ed a much be e QE o BSI senso
and how o he UV wa e o ms, om 10 nm o 40 0nm, he e sion BSI wi hou glass lid has a good sensi i i y.
When using glass lid he c ys al glass lid is abso bing bellow 350nm wa eleng h.
QUANTUM EFFICIENCY OF ULN5.3 IMAGE SENSOR FSI, BSI AND BSI WITHOUT GLASS LID
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
117
ULN5.3 CIS is o ally p og ammable using SPI in e ace. The de ice is designed o suppo di e en igge
modes by p og amming.
• Con inuous F ame (CF) mode s a ed ei he in e nally (by a START command) o wi h ex e nal igge ,
• Single-edged ype Single F ame mode s a ed wi h ex e nal igge , and
• PWC ype Single F ame mode s a ed wi h ex e nal igge
• Ex e nal synch onous eadou igge mode.
• Global ese single edged ype Single F ame mode wi h ex e nal igge , and
• Global ese PWC ype Single F ame mode wi h ex e nal igge
O he ea u es is con igu able Region o In e es (RoI). In he senso can be con igu ed up o 4 RoIs. The
e ical ow di ec ion can be p og ammed. Addi ionally, he ow ime, also known as in e al ime can be
p og ammed as well inc easing he minimum one gi en in p e ious able.
5.1.2 Senso A chi ec u e
Fig. 111- op shows he ULN5.3 block diag am and highligh s he on-chip in eg a ion o mul iple unc ions o
suppo sys em au onomy and educe he ha dwa e needed o came a implemen a ion, acco ding o he
e y concep o came a-on-chip [Fossu97]. Pa icula ly, unc ions needed o acqui e images, ead and digi ize
hem, co ec e o s and ansmi digi al images a e all on-chip. The schema ics a he bo om in Fig. 111
shows he concep o he Ac i e Pixel Senso (APS) used in his CIS chip.
The ULN5.3 ci cui y di ides in o i e sec ions: Pixel a ay, Readou and con e sion channel, Digi al con ol
ci cui y, Communica ion in e ace, and Auxilia y on-chip blocks.
Pixel A ay and Readou Pa h
As Fig. 111-bo om shows, he APS is a 4-T one and employs a Pinned-Pho oDiode (PPD) o sensing
[Fossu14].The Pixel A ay con ains 2304 x 2304 ac i e pixels (5.3M px e ec i e esolu ion), and he pixel
pi ch is 6.5 m. The op imum op ical o ma is he e o e 1 1/6 inch. In addi ion, he pixel a ay con ains a
numbe o Op ically-Black (OB) columns and ows as well as a numbe o dummy columns and ows used o
doing HFPN and VFPN co ec ions, by a e aging he OB pixels and sub ac ing o he whole ow and column
espec i ely.
The senso Readou Pa h ollows a pe -column pa allel app oach. I con ains one wo-s ages ADC eadou
channel pe pixel column ha ope a es in pa allel, condi ioning and digi izing he analog da a om 1 en i e
ow o pixels simul aneously. I also con ains addi ional pe -column channels o ead Op ical- Black (OB)
columns and dummy columns. The eadou pa h has been op imized o low-noise ope a ion.
On-chip digi al ci cui y and communica ion in e ace
The on-chip Digi al Ci cui y is esponsible o he o e all con ol o he sys em and he gene a ion o he
ames o he high-speed LVDS ou pu s. The on-chip Digi al ci cui y is composed o :
• Senso con ol block, which execu es he o e all con ol o he sys em, including wa e o m
gene a ions o pixel a ay and eadou . The pixel con ol signal (T , Rs and Sel) a e d i en by on-chip
p og ammable bu e s.
• Se ializa ion block (PSER). Once digi ized by he Readou , pixel da a a e se ialized ou o he F aming
blocks.
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
118
(TOP) ULN5.3 CIS BLOCK DIAGRAM
(BOTTOM) SCHEMATICS OF THE 4T APS IN ULN5.3
Figu e is in blank because i is con aining con iden ial in o ma ion.
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
119
• The LVDS F aming block ecei es da a om Se ializa ion block, gene a es he LVDS ame, and
se ializes i ou .
• Communica ion in e ace blocks. This pa comp ises all blocks ela ed o inpu /ou pu signals, such
as SPI, ex e nal con ol signals (TRIGGER, RESETN, e c.) and da a ou pu . The da a a e ou pu ed
h ough LVDS po s. The LVDS ou pu is o med o 32 LVDS po s o image da a
(PDO[0:31]/NDO[0:31]), wo mo e o clock- eco e ing pu pose (PCLK0/NCLK0 and PCLK1/NCLK1),
and addi ional wo o synch oniza ion da a (PSYN0/NSYN0 and PSYN1/NSYN1). The clock and
synch oniza ion in o ma ion is copied in o wo LVDS po s in o de o gi e lexibili y o he ecei e o
ou e ei he one clock and synch oniza ion po along wi h he 32 da a po s o g oup one clock and
synch oniza ion po wi h 16 da a po s and ha e ano he g oup wi h he emaining clock and
synch oniza ion po wi h he es o da a po s.
Auxilia y on-chip blocks
The Auxilia y On-Chip Blocks a e auxilia y on-chip IPs, which educe he numbe o ex e nal componen s
equi ed o he ope a ion o he ULN5.3 de ice. They include he Powe -on-Rese (PoR), he empe a u e
eenso (TempS), a Clock Gene a ion Block wi h a low-ji e low-powe Phase Locked Loop (PLL), and he
Re e ence Vol age Gene a o wi h a high-accu acy band-gap.
The ULN de ice is con olled by a numbe o in e nal clocks, which a e gene a ed om a single ex e nal
c ys al e e ence (o clock sou ce) whose equency is scaled by using he PLL. The nominal e e ence ex e nal
equency is 8MHz.
5.1.3 Ope a ion modes
The pixel (see Fig. 111-bo om) is d i en by he con ol signals T , and Rs . Once he pho o-gene a ed cha ge
is con e ed o ol age a he Floa ing Di usion (FD) capaci ance, i is ead ou ia he in-pixel sou ce- ollowe
(SF) and he selec ion swi ch (Sel) ha connec s i o he da a column and cu en sou ce.
The wa e o ms applied o he con ol signals abo e a e gene a ed ia a highly lexible s a e machine ha
pe mi s p og amming a la ge numbe o combina ions in ol ing he con ol sequence as well as he du a ion
o each pulse. The exposi ion (o p ocedu e o ligh sensing and pho o-gene a ed cha ge ans e ence and
s o age a he FD node), equi es passing h ough a numbe o s a es ha can be g ouped as ollows:
• S a es B: Requi ed o p epa e he exposi ion. Du ing hose s a es, Rs and T signal a e ac i ed
simul aneously o clean he FD node and he pho odiode. I can also be e e ed o as he “Rese ”
phase.
• S a es E: Ligh sensing o Exposu e phase. Ac ually, ligh sensing begins a he alling edge o T when
Rs s ill emains a high le el, i occu s wi hin S a es B. The end o exposu e phase occu s a he
ollowing alling edge o T when he cha ge accumula ed in he pho odiode is ans e ed o he FD
node , which may happen du ing he eadou in olling shu e o du ing S a es F in global shu e .
• S a es F: Requi ed o inish he exposi ion. In global shu e , he ans e ence o he pho o-cha ge o
FD node is pe o med du ing hose s a es, which equi es a p e ious ac i a ion o he Rs signal o
e eshing he FD, ollowed by an ac i a ion o he T signal o ac ually ans e he cha ge om he
pho odiode o he FD node. In olling shu e , only he e eshing Rs is ac i a ed wi hin S a es F, as
he cha ge ans e ence akes place du ing he eadou .
The du a ion o he comple e exposi ion p ocedu e is in ac he sum o he du a ion o s a es B, E, and F,
which a e p og ammable. ULN5.3 is designed o ope a e in wo di e en shu e modes:
• olling shu e and
• global shu e .
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
120
5.1.3.1. ROLLING SHUTTER
This mode yields low noise due o he cancella ion o he ese noise hanks o on-chip “ ue” Co ela ed
Double Sampling (on-chip CDS). As i is shown in 0- op bo h he image acquisi ion (exposu e) and eadou
a e pe o med in a ow-by- ow basis in olling-shu e mode.
he pixel a ay in an "idle s a e". Fi s , du ing he B s a es o each ow, he pho odiode and he FD node
mus be cleaned and e-se , by ac i a ing he ans e (T ) and ese signal (Rs ) simul aneously. The image
acquisi ion ac ually s a s when he ans e con ol signal (T ) is deac i a ed. F om his s a e on, he pho o-
gene a ed cha ge s a s o accumula e elec ons om pho ons ecombina ion in he pinned diode.
(TOP) ILLUSTRATING ROLLING SHUTTER ROW-BY-ROW EXPOSITION AND READOUT
(BOTTOM) EXAMPLE OF ROLLING-SHUTTER PIXEL CONTROL WAVEFORMS
Fo each ow, he eadou phase s a s eading he ese alue, esul ing om lowe ing he Rs con ol
signal. This signal has been kep high du ing he whole exposu e o pulsed igh be o e he eadou (as
exempli ied in 0-bo om. Then, he pho o-gene a ed cha ge is ans e ed om pho odiode o he FD and
ead ou . A his poin , he ac ual exposi ion inishes and all he pixel in o ma ion is ans e ed o he i s
s age o he Readou channel. No e ha eadou o a ow o pixels can be made concu en ly o any o he
ope a ion in o he ows, including he F s a es, excep he Readou s a e (R).
5.1.3.2. GLOBAL SHUTTER
In global-shu e mode, image acquisi ion and pho o-gene a ed cha ge ans e ence o FD node is pe o med
by all pixels simul aneously. The iming diag am shown in Fig. 113 is an example o global shu e pixel
con ol. No e ha global shu e equi es global con ol o he ese (Rs ) and ans e (T ) signals. The
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
121
sensing s a s wi h he pixel a ay in he "idle s a e". The Rs and T pulses a e ac i a ed globally wi hin B
s a es o clean he pho odiode. Then Rs pulse is applied a he beginning o F s a es o clean he FD node.
Finally, he cha ge ans e ence om he pho odiode o he FD is ca ied ou wi h a global ans e pulse
(T ).
ULN5.3 can combine global shu e image acquisi ion wi h wo ypes o eadou p ocedu es, enabling on-
chip ( alse) CDS, o o -chip ( ue) CDS, espec i ely. These wo eadou modes a e explained in he nex
sec ions.
ILLUSTRATING ROW-BY-ROW READOUT AFTER GLOBAL F STATES IN GLOBAL SHUTTER
5.1.3.3. ON-CHIP (FALSE) CDS
When on-chip CDS is used, once he F s a es ha e inished, he eadou p ocess is pe o med ow-by- ow,
like in olling shu e mode (see Fig. 114). Fi s , he da a co esponding o he ans e ed cha ge is ead ou ,
hen, a e a pulse o he Rs signal applied o he ac i e ow, he ese le el is ead ou .
Fig. 114 shows ypical global-shu e pixel con ol wa e o ms o on-chip CDS. Once he pho o-cha ge has
been ans e ed o he FD, a he end o F s a es, i emains he e ill he eadou o he pixel is ac i a ed.
Howe e , as a gene al ema k, i should be no iced ha no eadou is allowed o a pixel du ing i s F s a es,
because a ha ime he FD o such a pixel is being e-w i en wi h he new image da a coming om he PPD.
No e ha , since his global shu e on-chip CDS ac ually equi es wo ese pulses (one global pulse be o e
he cha ge ans e ence, and one pe - ow pulse du ing he eadou ), he e o s accompanying he ese and
signal le els a e no ully co ela ed. Thus, hey canno be ully emo ed by double sampling like in olling
shu e . This is why his ope a ion i is o en e e ed o as “ alse CDS”. In ac , only ixed-pa e n noises a e
e icien ly emo ed, whe eas empo al ese noise (KTC noise) associa ed o he FD sampling mechanism is
ac ually doubled in powe by his p ocedu e. Some low- equencies empo al noises a e a enua ed, hough,
p o ided ha he ime elapsed be ween he wo eadings is no oo long. False CDS is he only way o pe o m
on-chip CDS in global shu e ope a ion.
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
128
SPATIAL NON-UNIFORMITY (LEFT) AND TEMPORAL NOISE (RIGHT) AS A FUNCTION OF THE OUTPUT LEVEL USING
THE TWO METHODS ABOVE
5.1.4.2. OVERALL SYSTEM GAIN (CONVERSION GAIN)
Measu emen o he o e all sys em gain (Con e sion Gain (CG), called also Cha ge o Vol age Fac o (CVF)),
is based on he linea dependence o he sho noise RMS wi h he signal le el. In gene al, he empo al noise
o he senso can be app oxima ed by he summa ion (in powe ) o a signal-independen e ms, like eadou
noise, quan iza ion noise, sou ce ollowe noise e c… and he sho noise, sho , con ibu ion whose a iance
is p opo ional o he elec o a e age:
)()( .
2
0
.
22
0
22
0
2,
22
0
2da kyy
da kyy
eesho y K
K
KKK
−+=
−
+=+=+=
(5. 6)
I is possible o accu a ely measu e CG by plo ing he empo al a iance as a unc ion o he signal
a e age and calcula ing he slope o he bes - i ing s aigh line. Fig. 121 illus a es his me hod and shows
he ex ac ed alue he con e sion gain o ULN5.3 when ope a ing wi h 14 bi esolu ion.
CONVERSION GAIN EXTRACTION: TEMPORAL VARIANCE AS A FUNCTION OF THE OUTPUT LEVEL
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
129
5.1.4.3. SIGNAL-TO-NOISE RATIO (SNR)
Fo gi en illumina ion, he Signal- o-Noise Ra io (SNR) is de ined as he a io o he signal inc emen (wi h
espec o da k condi ions) o he o al empo al noise a he eached ou pu le el:
y
da kyy
SNR
.
−
=
(5. 7)
which amoun s o saying
e
e
SNR
=
(5. 8)
whe e e and e a e he signal and he o al empo al noise, espec i ely, exp essed in elec ons.
When plo ed s. he a e aged digi al ou pu , see Fig. 122, SNR eaches i s meaning ul maximum alue a
he so-called “sa u a ion capaci y
7
”, which co esponds o he maximum o he empo al noise. Beyond ha
poin , SNR la gely inc eases as a consequence o he sudden d op o he empo al noise caused by sa u a ion.
The SNR a he sa u a ion capaci y is epo ed as SNRmax.
y
da kysa y
e
sa e
SNR
..
.
max
−
==
(5. 9)
SNR VS. OUTPUT LEVEL (TOP) AND THE CORRESPONDING TEMPORAL NOISE (BOTTOM)
5.1.4.4. DYNAMIC RANGE
The Dynamic Range (DR) is de ined as he a io o he sa u a ion capaci y o he absolu e sensi i i y h eshold,
co esponding o he le el a which SNR equals uni y. When exp essed in elec ons, i becomes:
7
The sa u a ion capaci y mus no be con used wi h he ull-well capaci y. The sa u a ion capaci y mus no be con used wi h he ull-well capaci y. I
is no mally lowe han he ull-well capaci y, because he signal is clipped o he maximum digi al alue, 214-1, be o e he physical sa u a ion o he
pixel is eached.
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
130
2
.
..
min.
.
1288 K
DR
da ky
da kysa y
e
sa e
+
−
==
(5. 10)
In EMVA S anda d 1288 i is shown ha he ac o K/2 comes om he ac ha a he “SNR=1” poin
he e may be a non-negligible ac ion o ligh -induced sho noise, apa om he noise in da k condi ions.
Howe e , he K/2 e m is o en o e looked. In ULN5.3, DR is speci ied in e e ence o he pixel ull-scale
ol age ange and empo al da k noise as ollows:
𝐷𝑅=𝑉𝐹𝑆
𝜎𝑣.𝑑𝑎𝑟𝑘=𝑦𝐹𝑆−𝜇𝑦.𝑑𝑎𝑟𝑘
𝜎𝑦.𝑑𝑎𝑟𝑘
(5. 11)
5.1.4.5. SPATIAL NON-UNIFORMITY (DSNU, PRNU, AND FPN)
Spa ial non-uni o mi y is cha ac e ized by he ollowing pa ame e s:
• The da k-signal non-uni o mi y de ined as
𝐷𝑆𝑁𝑈1288=𝑠𝑦.𝑑𝑎𝑟𝑘
𝐾[𝑒−]
(5. 12)
whe e sy.da k is he spa ial noise in da k condi ions (in DN) and K is he sys em gain (in DN/e-).
• The pho o- esponse non-uni o mi y, de ined as:
𝑃𝑅𝑁𝑈1288=√ 𝑠𝑦.50
2−𝑠𝑦.𝑑𝑎𝑟𝑘
2
𝑢𝑦.50−𝑢𝑦.𝑑𝑎𝑟𝑘 ∙100%
(5. 13)
wi h 𝑠𝑦.50 and 𝑢𝑦.50 being he espec i e spa ial noise and a e age le el a ound 50% o sa u a ion.
5.1.4.6. RESPONSIVITY AND FF X QE
The esponsi i y is he slope o he ligh - o-digi al cu e. I is a unc ion o he ligh wa eleng h due o he
spec al dependence o se e al ac o s like o example he pho on ene gy, he ansmi ance o he op ics
and glass lid, he p esence o no o an i e lec i e coa ings o il e s, and he pixel quan um e iciency. Among
hose, he quan um e iciency (o be e , he p oduc o he ill- ac o x quan um e iciency, since he wo
ac o s canno be dissocia ed), domina es he spec al esponse o he senso acco ding o he linea law:
exp. ··)(· E
hc
A
FFQEK pix
da kyy
+=
(5. 14)
The quan um e iciency was p e iously depic ed in Fig. 110.
5.1.4.7. NON-LINEARITY
Non-linea i y o he ligh - o-digi al esponse is measu ed by i ing da a poin s be ween 5% and 95% o
sa u a ion capaci y o a s aigh line and compu ing he i ing e o o de ia ion y. The Linea i y E o (LE)
hen de ined as he mean o he di e ence o he maximal de ia ion o he minimal de ia ion:
2
)min()max( yy
LE
−
=
(5. 15)
I is exp essed in %FS.
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
131
5.1.4.8. DARK CURRENT
The da k ou pu is no cons an bu i inc eases wi h exposu e ime as a consequence o he he mal
gene a ion o elec ons wi hin he pho odiode. Using he EMVA S anda d 1288 no a ion, we can w i e
exp0.0. ·
Id he mdd
+=+=
(5. 16)
whe e μd.0 is he da k le el co esponding o i ually ze o exposu e and μI is he da k cu en exp essed
ei he in DN/s o , p e e ably, in e-/s.
exp
20.
22 0.
2·
Id he mdd
+=+=
(5. 17)
This equa ion is alid only i images a e no da k-le el compensa ed. I such compensa ion could no be
disabled o measu e he da k cu en , he equali y gi en in (5. 18) would be used ins ead, because he mally
induced elec ons ollow a Poisson dis ibu ion simila o he sho noise.
𝜎𝑡ℎ𝑒𝑟𝑚
2=𝜇𝑡ℎ𝑒𝑟𝑚
(5. 18)
5.1.5 Op ical pe o mances in 14 bi s olling-shu e mode
The op ical pe o mances o ULN5.3 ope a ion in he olling shu e mode and 14 bi s esolu ion a e depic ed
in Fig. 123, along wi h he Pho o T ans e Cu e (PTC) g aph.
OPTICAL PERFORMANCES AND PTC IN 14 BITS ROLLING SHUTTER MODE
Fig. 124 shows he esponsi i y and he linea i y e o , kep below 1% acco ding o he speci ica ion.
Linea i y unde low le el accumula ed ligh condi ions is ele an o low-noise image senso s because a
esponse comp ession on he low side would equi e applying a linea iza ion gain, hus inc easing he noise.
The linea esponse ea u ed by ULN5.3 makes such linea iza ion unnecessa y.
Fig. 125 is a no malized his og am o he da k empo al noise displaying he da k noise dis ibu ion o
ULN5.3 in 14 bi s mode. No e ha all pixels ha e a da k noise smalle han 6 e-RMS - much be e han o he
senso s in he a [Ma17] [Xinya15] as Fig. 125 illus a es. This pe o mance is e y impo an in low-noise
applica ions since la ge ails in da k noise dis ibu ion gene a e images wi h low noise bu wi h sal and
peppe e ec yielding image wi h non-good sense o quali y [Nakam06].
UNITS
ºC
dB
dB
e-
%
DN/e-
dB
DN
e- ms
FWC
15568
Da k noise, NEE
1.38
1.31
OPTICAL
PERFORMANCES
Tjunc ion
-5.55
DR
80.03
SNRmax
41.42
DSNU
1.71
PRNU
0.24
CG
1.05
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
132
RESPONSIVITY AND LINEARITY RESPONSE IN 14 BITS ROLLING SHUTTER OPERATION MODE
(LEFT) DARK TEMPORAL NOISE DISTRIBUTION IN 14 BITS ROLLING-SHUTTER OPERATION MODE; (RIGHT)
RESULTS FOR THE SENSOR IN [MA17]
5.1.6 Op ical pe o mances in 16 bi s olling-shu e mode
Fig. 126 shows he op ical pe o mances he pho o ans e cu e in his mode. Sub-elec on eadou noise
is achie ed wi hou ex e nal mul isampling. Fig. 127 shows he esponsi i y and he linea i y e o ; he la e
emains below 1% and does no equi e linea iza ion o co ec low-ligh esponse.
OPTICAL PERFORMANCES AND PTC IN 16 BITS ROLLING-SHUTTER MODE
UNITS
ºC
dB
dB
e-
%
DN/e-
dB
DN
e- ms
OPTICAL
PERFORMANCES
Tjunc ion
-6.19
DR
83.19
SNRmax
41.77
DSNU
0.67
PRNU
0.24
CG
4.27
FWC
14573
Da k noise, NEE
3.96
0.93
No malized Pixel Coun
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
133
RESPONSIVITY AND LINEARITY RESPONSE IN 14 BITS ROLLING-SHUTTER OPERATION MODE
Finally, Fig. 128 depic s he da k noise dis ibu ion o 16 bi s mode. All pixels emain below 5.25 e-RMS
− be e han o he 14 bi s mode.
DARK TEMPORAL NOISE DISTRIBUTION IN 16 BITS ROLLING-SHUTTER OPERATION MODE
5.1.7 Pe o mances in 14 bi s mul isampling (M = 16) olling-shu e mode
Co esponding op ical pe o mances, esponsi i y/linea i y e o and da k noise dis ibu ion a e depic ed in
Fig. 129, Fig. 130 and Fig. 131, espec i ely. No e he ollowing: i) linea i y e o is below 1% acco ding o he
speci ica ion and does no equi e low-ligh co ec ion; ii) he da k noise dis ibu ion is e en be e han o
he 16bi s mode, wi h all pixels below 4 e-RMS.
OPTICAL PERFORMANCES AND PTC IN 14BITS MULTISAMPLING M =16 ROLLING-SHUTTER MODE
UNITS
ºC
dB
dB
e-
%
DN/e-
dB
DN
e- ms
OPTICAL
PERFORMANCES
Tjunc ion
-4.08
DR
85.85
SNRmax
41.59
DSNU
0.27
PRNU
0.22
CG
1.09
FWC
14739
Da k noise, NEE
0.77
0.71
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
134
RESPONSIVITY AND LINEARITY RESPONSE IN 14 BITS ROLLING SHUTTER OPERATION MODE
DARK TEMPORAL NOISE DISTRIBUTION IN 14 BITS MJLTISAMPLING M = 16 ROLLING-SHUTTER OPERATION
5.1.8 Da k cu en
Da k cu en measu emen has been done using 14 bi s olling-shu e ope a ion me hod. To ob ain he da k
cu en he ULN5.3 de ice is measu ed inside a clima e chambe , and da k images ha e been ob ained
sweeping he exposu e ime, his measu emen is epea ed in he ange o empe a u e om -40⁰C o 50⁰C.
Table 15 shows he da a o e e y measu emen , whe e T oom is he empe a u e inside he clima e chambe ,
Tjunc ion is he empe a u es in he die, μ(DN/s) is he slope o he mean alue o empo al s anda d de ia ion
o he pixels ou pu s he exposu e ime, and σ(DN/s) is he slope o he spa ial s anda d de ia ion o he
pixel ou pu s he exposu e ime.
The able also shows wo pa ame e s a e ex ac ed o he empo al(μ) and spa ial(σ) componen s,
namely:
• he doubling coe icien (Kd), and
• he alue o each pa ame e a 25⁰ C(Nd0).
The doubling coe icien is de ined as he empe a u e shi he de ice should expe ience in o de o inc ease
(o dec ease) he da k cu en by a ac o o 2. The da k cu en a an empe a u e To is gi en as:
𝜇𝑑𝑎𝑟𝑘=𝑁𝑑0∙2(𝑇𝑜−25)
𝐾𝑑
(5. 19)
Fig. 132 shows he da k cu en measu emen on a semi-loga i hmic scale whe e such ha he doubling
coe icien (Kd) is he slope o he cu e.
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
135
Table 15. DARK CURRENT MEASUREMENT DATA
Da k Cu en measu emen
T oom (ºC)
Tjunc ion(ºC)
μ (DN/s)
σ(DN/s)
-40
-11.1
0.4
0.1
-30
-1.9
1.1
0.6
-20
6.4
2.5
1.9
-10
16.6
6.5
5.5
0
25.6
16.8
13.7
10
35.1
45.0
32.2
20
44.2
125.7
72.6
30
54.1
340.5
152.0
40
62.7
916.2
306.5
50
68.4
1819.0
447.6
Ex ac ed pa ame e s
CG(DN/e)
Kd (ºC)
(doubling coe )
μ
σ
6.6
7.9
1.0670
Nd0 (e/s)
( alue a 25ºC)
μ
σ
15.8
11.0
(a) (b)
DARK CURRENT MEASUREMENT: (A) TEMPORAL DEVIATION (B) SPATIAL DEVIATION
5.1.9 Glow e ec and mi iga ion echniques
This sec ion is in blank because i is con aining con iden ial in o ma ion.
5.1.10 Low ligh benchma k compa ison
Low ligh es bench
A low ligh es bench has been used o de elop speci ic cha ac e iza ion o low-noise image senso s unde
con olled low ligh condi ions. Fig. 133 depic s he low ligh bench, which consis s o a black box, an
in eg a ing sphe e wi h shu e , a de ec o o came a, a il e holde and a calib a ed pho odiode o measu e
he ligh in ensi y.
The low ligh bench has been designed o allow:
• Con oled and ep oducible ligh condi ions: i adiance, inpu MTF and ansmi ance es cha .
• Low ligh measu emen s, image and compa isons o de ices, including om di e en echnologies
EMCMOS, EMCCD, EBCMOS and low noise CMOS image senso
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
136
• Measu emen s o ex apola ions o a iable applica ion condi ions: ligh condi ions, op ics,
a mosphe e model, a ge con as s.
• P o iding measu emen s and me i igu es (MTF, Noise, Con as h eshold [Bisog07])
• P o iding De ense and Su eillance speci ic me ics (Johnson c i e ia [Johns58]).
Fig. 134 shows he measu emen se up consis ing o KOWA lens op ic, ligh in ensi y a ia ion hanks o he
shu e ape u e, in eg a ion ime 1/60s, Gain x4 in ONYX MAX and CIS 2521 senso s and USAF es cha a
20cm om he sphe e ou pu .
LOW LIGHT TESTBENCH
MEASUREMENT SETUP
Senso pe o mance compa ison
Nigh le els images compa ison has been done using ULN5.6 and wo addi ional senso :
• ONYX is a 1.3 Megapixel senso o low ligh applica ions [Te2 21]
• CIS2521 senso [BAE]
Table 16 collec s da a ega ding senso compa ison. Fu he mo e, Fig. 135 o Fig. 139 include images
cap u ed by he h ee senso s o illuminance le els om 10mlux o 0.69 mlux.
Table 16. IMAGE SENSOR PERFORMANCES COMPARATIVE
Fea u es
Uni s
ULN5.3
ONYX1.3M
CIS2521
Resolu ion
2304 (H) x2304 (V)
1280 (H) x 1024 (V)
2560 (H) x 2160 (V)
5.1 Low noise senso wi h 5.3 Mpixel and 0.7e ms da k noise _
137
Pixel size
m
6.5
10
6.5
F ame Ra e
ps
87
95
100
Dynamic ange
dB
84.4
73
83.5
Read Noise
eRMS
1.3
3
2
Full Well Capaci y
Ke-
15
14
30000
IMAGES AT NIGHT LEVEL 2 WITH 10 MLUX AT F#1
IMAGES AT NIGHT LEVEL 3 WITH 5 MLUX AT F#1
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
144
ULN66 READOUT CHANNEL
5.2.2.1. DIGITAL CDS
The ULN66 senso can ope a e in digi al CDS. When his mode is selec ed, he 1s s age o he ADC con e s
he da a coming om he pixel e e ed o a ol age e e ence and gene a es he analog esidue a e he
con e sion. The esidue is con e ed by he 2nd ADC s age. When only he one gain le el is selec ed (ei he
high o low gain), he ADC con e s wo samples pe ow ime, one ela ed o he pixel RESET ol age and
he o he ela ed o he pixel SIGNAL ol age. The digi ized samples a e sub ac ed in he digi al domain,
which mus be done ex e nally (in he ecei e sys em). Fig. 147 shows he digi al CDS ope a ion.
ULN66 DIGITAL CDS OPERATION
When dual con e sion gain is selec ed he ADC con e s ou samples pe ow ime, one ela ed o he
RESET low gain ol age, he second o he RESET high gain, he hi d o he SIGNAL high gain and he ou h
da a column line
Pixel SF
sel
P og.
Cu en
Sou ce
+
P og.
O se
1s s age ADC 2nd s age ADC
4, 5, 6 10
MSBs LSBs
CDS
Analog
Bu e
Residue
D1D2
21 2
2DDD N
ou +=
5.2 La ge scale 66 Mpixel 10 mm pixel low noise and high _
145
ela ed o he SIGNAL low gain. The digi ized samples a e sub ac ed and p ocessed in he digi al domain,
which mus be done ex e nally. Fig. 148 shows he digi al CDS ope a ion wi h dual gain.
ULN66 DIGITAL CDS OPERATION WITH DUAL CONVERSION GAIN
5.2.2.2. ANALOG CDS
The image senso can ope a e in analog CDS. When only one gain is selec ed, low o high gain, he ADC s a s
in eg a ing he RESET le el, once he cha ge is ans e ed and he SIGNAL le el is eady, he ADC in eg a es
he pixel ou pu wi h sign in e sion, hus pe o ming he CDS ope a ion in analog domain and gene a ing he
analog esidue o he second s age o he ADC. The ADC deli e s only one wo d ou pu in one ow ime, since
CDS is done in e nally in analog domain. Fig. 149 shows he analog CDS ope a ion. In his mode he ci cui y
and a chi ec u e o eadou channel is he same, bu i is ope a ed di e en ly
ULN66 ANALOG CDS OPERATION
The analog CDS wi h dual high gain is a ailable as well. In his case, he ADC is con e ing he high gain in
he same way i is done when only one gain is selec ed, howe e he low gain componen is only sampled by
he i s s age o he ADC and hen he CDS ope a ion is pe o med p io o sen o he second s age o he
ADC, as desc ibed in Chap e 4. This ope a ion op imized he ade-o be ween p ecision, powe
consump ion and ime, since he low gain signals does no equi e ul a-low le el o noise because sho noise
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
146
is supposed o be domina ing when
low gain is used. Fig. 150 shows he
analog CDS wi h dual con e sion gain
ope a ion mode.
ULN66 ANALOG CDS
OPERATION WITH DUAL CONVERSION
GAIN
5.2.2.3. ROLLING AND GLOBAL SHUTTER MODE
Rolling-shu e mode
Like ULN5.3, he image senso ULN66 inco po a es olling and global shu e mode. The olling mode yields
low noise due o he cancela ion o ese noise by he co ela ed double sampling ope a ion. Fig. 151 shows
he olling shou e ope a ion whe e e e y ow is exposed in a ow ime shi , wi h exposu e and eadou is
done in a ow-by- ow basis.
ROLLING-SHUTTER ROW-BY-ROW EXPOSURE AND READOUT
Global-shu e mode
In global shu e mode, image acquisi ion and pho o-gene a ed cha ge ans e ence o FD node is pe o med
by all pixels simul aneously. Typical global shu e wi h alse CDS (explained in Sec ion 5.1.3.2) is a ailable in
ULN66 as well, howe e his mode in oduces he ese noise and i is no in e es ing om he poin o iew
o deli e ing low noise da k images. Ins ead, ULN66 p o ides global shu e ope a ion wi h ex e nal CDS
deli e ing he ese image, which is eadou p io o he global ans e pulse, and hen he signal image is
eadou . Since he signal is accumula ed o e he ese , he pixel ese noise is o ally cancelled ende ing a
low noise image. This ope a ion mode is call o -chip global shu e ue CDS. The low noise pe o mance can
be no so good as olling shu e ope a ion, because some eadou con ibu ion like quan iza ion noise is
added wice, he low equency noise con ibu o o sou ce ollowe is no so e ec i ely cancelled because
he ime be ween he wo samples is much highe , and inally he e is an addi ional con ibu ion coming
om he da k cu en ha is in eg a ed in he loa ing di usion node be ween he ese o he loa ing
di usion and he signal eadou . This con ibu ion is no negligible since he da k cu en gene a ed in he
loa ing di usion can be a leas one o de o magni ude la ge han pho odiode da k cu en . To mi iga e
da k cu en con ibu ion in he loa ing di usion ac i e colling sys em can be added in he inal sys em.
ow
B
0E F R
B
1E F R
2
3
4
5
...
n-2
n-1
BE F R
BE F R
BE F R
BE F R
...
BE F R
BE F R
T ame
...
5.2 La ge scale 66 Mpixel 10 mm pixel low noise and high _
147
GLOBAL SHUTTER OPERATION OFF-CHIP TRUE CDS MODE
When dual gain ope a ion is selec ed, ou ames mus be eadou , one ela ed o he ese low gain,
ano he wi h he ese high gain, he hi d o he signal high gain and inally he ou h wi h he signal low
gain. The digi ized images a e sub ac ed and p ocessed in he digi al domain, ha can be done in he ecei e
sys em.
5.2.2.4. MULTISAMPLING
Mul isampling ope a ion mode is also a ailable in olling shu e and global shu e , Fig. 153 and Fig. 154
espec i ely. In his mode, ex e nal eadou wi h non-des uc i e- eadou is used. Ex e nal p ocessing
a e aging e e y sample can be used o lowe ing pa o he eadou noise. The wide-band noise con ibu o s
like he mal noise a e educed by he squa e oo o M, being M he numbe o samples o ese and signal
le els. Howe e , he low equency noise componen s, like he licke noise 1/ , a e inc eased because he
CDS ime is inc eased wi h he numbe o samples, M. In ac , he e is a ade-o o ob ain he op imum
alue o M, whe e inc easing he numbe o samples does no dec ease he o al da k noise.
ROLLING SHUTTER OPERATION CORRELATED MULTISAMPLING (CMS) MODE
GLOBAL SHUTTER OPERATION CORRELATED MULTISAMPLING (CMS) MODE
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
148
5.2.3 Senso unc ions
5.2.3.1. WINDOWING
ULN66 can be con igu ed o eading he whole pixel a ay o a se o egion o in e es . Up o 4 di e en
ROIs can be p og ammed. E e y ROI is de ined wi h he ini ial ow and he wid h, being ead all columns. An
example o ROI selec ion is shown in Fig. 155. Addi ional o he 4 ROIs he special ows, consis ing o dummy
ows, OB ows and 2 es ows can be
selec ed. The wo es ows can be
selec ed o calib a ion and o es he
eadou pa h.
The windowing a ec s he
maximum ame a e because he
eadou ime is p opo ional o he
numbe o ows de ined o be eadou .
REGION OF INTEREST
DEFINITIONS
5.2.3.2. BINNING
ULN66 also inco po a es binning unc ion. The binning consis s o a e aging o accumula ing he pixel signal
o neighbou pixels. In his case, he senso can p o ide ow binning in he analog domain. To do ha , wo
ows a e enable a he same ime, ac i a ing he pixel con ol signals o wo consecu i e ows o ac i a e he
pixel sou ce ollowe s and sho -ci cui he sou ce ollowe ou pu . This ope a ion educes he pixel a ay
esolu ion by wo in he ow di ec ion. This ope a ion inc eases SNR and DR, as well as ame a e.
5.2.3.3. LOW POWER MODE
ULN66 includes low powe mode o educe he “glow e ec ”. In e nal blocks can be independen ly selec ed
h ough con igu a ion egis e o be powe o du ing exposu e o idle s a e, as shown in Fig. 156. The senso
au oma ically can en e in low powe mode, disabling all blocks con igu ed in low powe con igu a ion
egis e . Be o e en e ing he low powe mode (idle o exposu e), he eadou pipeline mus inish he cu en
eadou ope a ion. I means ha enable he low powe mode mus wai un il he da a is ou pu h ough he
LVDS. A p og ammable delay can be p og ammed o adjus en e ing in low powe mode.
LOW POWER MODE IN ROLLING SHUTTER
5.2.4 Senso yield imp o emen
This sec ion is in blank because i is con aining con iden ial in o ma ion.
5.2 La ge scale 66 Mpixel 10 mm pixel low noise and high _
149
5.2.5 Op ical pe o mances measu emen s
The op ical cha ac e iza ion o ULN66 is done using simila se up and p ocedu e as desc ibed in Sec ion 5.1.4
, ollowing he EMVA s anda d 1288 [EMVA21]. In he nex sec ion he op ical pe o mances o ULN66 senso
in di e en ope a ion modes a e gi en a : i) low empe a u e, −40⁰ C, and ii) oom empe a u e, 20⁰ C, wi h
he excep ion o mul isampling modes whe e only low empe a u e da a is shown.
The ULN66 senso will be inco po a ed in a scien i ic came a wi h ac i e cooling wo king a −40⁰ C ha
p o ides junc ion empe a u es in he ange 31⁰ C and 28⁰ C depending in senso powe consump ion. The
op ical senso pe o mances ha e been op imized a low empe a u e and hen measu ed a oom
empe a u e o ob ain he pe o mances de ia ions i any.
Besides op ical pe o mance me ics, esponsi i y, linea i y, spa ial noise and PTC cu es a e showed in
o de o check he main pe o mances along he senso esponse.
5.2.5.1. ROLLING SHUTTER 14 BITS INTERNAL CDS HIGH GAIN
OPTICAL PERFORMANCES 14BITS INTERNAL CDS HIGH GAIN
RESPONSIVITY, LINEARITY, SPATIAL NOISE AND PTC IN RS 14BITS INTERNAL CDS HIGH GAIN
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
150
5.2.5.2. ROLLING SHUTTER 14 BITS INTERNAL CDS LOW GAIN
OPTICAL PERFORMANCES 14BITS INTERNAL CDS LOW GAIN
RESPONSIVITY, LINEARITY, SPATIAL NOISE AND PTC IN RS 14BITS INTERNAL CDS LOW GAIN
5.2 La ge scale 66 Mpixel 10 mm pixel low noise and high _
151
5.2.5.3. GLOBAL SHUTTER 14 BITS EXTERNAL CDS HIGH GAIN
OPTICAL PERFORMANCES 14BITS IN GLOBAL SHUTTER EXTERNAL CDS HIGH GAIN
RESPONSIVITY, LINEARITY, SPATIAL NOISE AND PTC IN GLOBAL SHUTTER 14BITS INTERNAL CDS HIGH GAIN
Chap e 5 Low Noise CISs wi h Op imized Pixels, Dual Gain and O e sampled ADCs
152
5.2.5.4. GLOBAL SHUTTER 14 BITS EXTERNAL CDS LOW GAIN
OPTICAL PERFORMANCES 14BITS IN GLOBAL SHUTTER EXTERNAL CDS LOW GAIN
RESPONSIVITY, LINEARITY, SPATIAL NOISE AND PTC IN GLOBAL SHUTTER 14BITS INTERNAL CDS LOW GAIN
5.2 La ge scale 66 Mpixel 10 mm pixel low noise and high _
153
5.2.5.5. ROLLING SHUTTER 14 BITS EXTERNAL CDS HDR
OPTICAL PERFORMANCES 14BITS IN ROLLING SHUTTER EXTERNAL CDS HDR
RESPONSIVITY, LINEARITY, SPATIAL NOISE AND PTC IN ROLLING SHUTTER 14BITS EXTERNAL CDS HDR
