CMOS Image Sensing part II

06/03/2011

Do Not Copy

Cmos Image g Sensor Jeff Raynor ST Imaging Division, Edinburgh Lionel Martin, AST, Rousset

AST Research & Innovation France

Summary

Readout & conversion

Amplification

Problems

Conclusions

Agenda Do Not Copy

        

Optics Photons  Electrons Collection Amplification Sensor Architecture CMos Sensor Optics Source of Noise Characterization Colour

previous session

AST Research & Innovation France

CMOS Sensors – Photons  Electrons

Do Not Copy

Absorption Photons separate electrons from atoms. Si – Visible light Ge – Near IR HgCdTe – Far IR

Goals: (a)Minimize reflection (b)Minimize obstructions

No electric field AST Research & Innovation France

ST Microelectronics 02/Mars/2011

1

CMOS Image Sensing part II

06/03/2011

CMOS Sensors - Collection

Do Not Copy

Need to move electrons from atoms, otherwise recombination. Use electric field Goals:





• Collect all electrons • Minimize “false” electrons – “dark current”

With electric field AST Research & Innovation France

Photons  Electrons Ephoton = h  c / 

Energy per photon:

Do Not Copy

(h=6.62E-34Js)

e.g. 550nm, Ephoton = 361E-21J Photon Flux Density:

D Wm-2

(Radiometric units)

• Photons per pixel: Nphotons = D  Apix  Tint / Ephoton Apix = Area Pixel Tint = Exposure (Integration) Time

Quantum Efficiency:

QE = Ne / Nphotons

Fill Factor:

FF = Arealight_sensitive / Areapixel

Serial

X-Decoder

Parallel Output, e.g. 10/8/5/4 bits Faster Throughput Slower Clock More pads  bigger package

   

Serial Output, 1 bit “slower” Throughput Faster Clock (200/400/600 MHz) Fewer pads  smaller package

AST Research & Innovation France

Compact Camera Port (CCP) CCP Transmitter

±1.5mA Current

±150mV Signal

Do Not Copy

CCP Receiver

DATA+ TX_DATA

RX_DATA DATA100Ω Termination Resistor

100Ω Balanced Transmission T i i Li Line

CLK+/STRB+

TX_CLK

RX_CLK

CLK-/STRB-

DATA

1

0

1

1

0

0

0

1

CLK AST Research & Innovation France

ST Microelectronics 02/Mars/2011

5

CMOS Image Sensing part II

06/03/2011

CCP Frame Structure

Do Not Copy

 32-bit data system  False Sync Codes  FFH00H00H illegal at ANY bit position within image data

Image Data: RGBXXX, YUV4XX, Raw Bayer and JPEG

LS

LE

 LS Bit transmitted 1st 0 or more Status Lines

Checksums

0 or more Status Lines

Line Blanking Period

FS

FE

Frame Blanking Period

Code

32-bit Value

Line Start Code, LS

FFH 00H 00H X0H

Line End Code, LE

FFH 00H 00H X1H

Frame Start Code, FS

FFH 00H 00H X2H

Frame End Code, FE

FFH 00H 00H X3H AST Research & Innovation France

CMOS Sensors - Optics Do Not Copy

CMOS Sensors & Optics • Fill Factor • Microlenses Mi l • Electronic Shutter

AST Research & Innovation France

CMOS Sensors – Fill Factor Do Not Copy

VRT Vdd

VRead RST

M2

M1

M3

Vpd Vint Cpd

GND

Ideal – 100% M2

M1

Circuit

M3

Add transistors: 60%

Add wiring: 50%

Add diode: 40%

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

6

CMOS Image Sensing part II

06/03/2011

CMOS Sensors - Microlens Do Not Copy

Spherical Aberration

Photosensitive Area

• Put lens on each pixel to focus light onto sensitive part • Light collection not imaging AST Research & Innovation France

CMOS Sensors – Microlens Manufacture

Before heating

Do Not Copy

After heating (constant volume)

AST Research & Innovation France

CMOS Sensors – Microlens Vignetting Do Not Copy

Centre Pixel

Off-centre Pixel

Edge Pixel

• Pixels off-centre  Lose light = Vignetting AST Research & Innovation France

ST Microelectronics 02/Mars/2011

7

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Microlens Vignetting  Anti-vignette microlens (radial shift)

Centre Pixel US6884985 Jeff US7049168 Keith

Off-centre Pixel

Do Not Copy

Edge Pixel

• Move microlens wrt pixel  NO Vignetting AST Research & Innovation France

CMOS Sensors - Cross Section Do Not Copy

2µm

10µm

7.5[6.9]µm

Not to scale AST Research & Innovation France

CMOS Sensors - Exposure Control Do Not Copy

Large Range of Illumination – 150dB

 “Real World” / Human Eye:150dB dynamic range (25 bits)  Greater than typical sensor (10 bits)  Need to control amount of light (exposure) of sensor  Use “Electronic Shutter” AST Research & Innovation France

ST Microelectronics 02/Mars/2011

8

CMOS Image Sensing part II

06/03/2011

CMOS Sensors Exposure Control – Single Pixel

Do Not Copy

Change Pixel’s Exposure Period (Integration Time) Tint1

Measure Signal

RST Vpd Tint1

RST Vpd

Saturated / "Clipped" Tint2

RST Vpd AST Research & Innovation France

CMOS Sensors Electronic Shutter Vs Film Do Not Copy

 3T CMOS Rolling blade  35mm blade shutter 1/60th (flash sync)

1/125th

1/250th

 4T CMOS / CCD  iris shutter

AST Research & Innovation France

CMOS Sensors Electronic Shutter - Theory Electronic Shutter

Do Not Copy

• Expose time = Treset - Treadout • Can reset pixels simultaneously BUT • Many pixels share column readout  Cannot readout all pixels simultaneously (3T only, 4T has storage at pixel) • If simultaneous Reset & sequential Readout Different pixels have different exposure  • T=0 reset • T=1ms Read row 1 (Tint = 1ms) • T=100ms Read row 100 (Tint = 100ms) AST Research & Innovation France

ST Microelectronics 02/Mars/2011

9

CMOS Image Sensing part II

06/03/2011

CMOS Sensors Electronic Shutter Tint=20ms Do Not Copy

T=0ms Reset Row 1

T=10ms Reset Row 2

T=20ms Reset Row 3 Readout Row 1  Tint = 20ms

= Readout = Reset T=30ms Reset Row 4 Readout Row 2  Tint = 20ms

T=40ms Reset Row 5 Readout Row 3  Tint = 20ms AST Research & Innovation France

CMOS Sensors Electronic Shutter Tint=30ms Do Not Copy

T=0ms Reset Row 1

T=10ms Reset Row 2

T=20ms Reset Row 3

= Readout = Reset T=30ms Reset Row 4 Readout Row 1  Tint = 30ms

T=40ms Reset Row 5 Readout Row 2  Tint = 30ms AST Research & Innovation France

CMOS Sensors Electronic Shutter Do Not Copy

 Webcam (streaming video) – Full Frame Integration 20ms 1024

Row

2 1 0ms

20ms 20ms

40ms

Time

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

10

CMOS Image Sensing part II

06/03/2011

CMOS Sensors Electronic Shutter Do Not Copy

 Webcam (streaming video) – More light  Shorter Integration 5ms 1024 5ms Row

2 1 0ms

5ms 5ms

20ms

40ms

Time

AST Research & Innovation France

CMOS Sensors Electronic Shutter Do Not Copy

 DSC (single image) – Less light  Very long Integration 42ms 1024

Do Nothing! (Integrate = Collect Light)

Row

2 1 0ms

42ms 5ms

20ms

40ms

Time

AST Research & Innovation France

CMOS Sensors Electronic Shutter Do Not Copy

 Examples show integer number of lines between reset and readout  NOT necessary  Integer g p part = “Coarse exposure” p  Fractional part = “Fine exposure”

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

11

CMOS Image Sensing part II

06/03/2011

CMOS Sensors Rolling Blade Shutter Do Not Copy

Photo: [email protected]

Bright Light  Short exposure  No camera shake / blur Rolling blade shutter  Not all pixels exposed at same time  Movement distorts image e.g. 10µs exposure, 50ms acquisition time. AST Research & Innovation France

CMOS Image Sensors Do Not Copy

Noise Annoys! Le bruit ennuie! (How CMOS Sensors don't work!)

AST Research & Innovation France

CMOS Image Sensors Do Not Copy

Noise Annoys! – Sources of Noise     

Light – photons are noisy!! Dark Current of the p pixel Thermal Noise of transistors Reset Noise of pixel External noise sources

 Image noise depends on several factors AST Research & Innovation France

ST Microelectronics 02/Mars/2011

12

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Non Idealities Do Not Copy

Non Ideal Features of CMOS Sensors & compensation / mitigation  Photon Shot Noise

 larger pixels

 Transistors obscure pixel  use microlenses  Fixed Pattern Noise

 compensate

 “kTC” = Reset noise

 Read twice & subtract (or depleted diode)

 Dark current = leakage through pixel, strongly dependent on temperature, dependent on time  No fix  AST Research & Innovation France

CMOS Image Sensors Do Not Copy

Dominating Factors • • • •

At high light levels At high temperature At low light levels At low temperature

Photon Shot noise, PRNU Dark Current Dark Current, External, Reset External, Reset

For both CMOS & CCD

AST Research & Innovation France

CMOS Image Sensors Light Vs Dark Do Not Copy

Dark Vpd

Max Tint

Light Max Tint

Min Tint

< Max Voltage

Max Voltage

 Apply Gain

 Apply Exposure Control

 Readout Noise

AGC

Min Tint

Max Voltage > Max Voltage

 Photon Shot Noise

AEC

= Overload!

Light Level

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

13

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Photon Shot Noise Do Not Copy

"Real" World Vs. Quantum World (macroscopic)

= 100 

100 cents

100 photons

AST Research & Innovation France

CMOS Sensors – Photon Shot Noise Do Not Copy

Light is noisy!!!

4.5%

Probab bility

4.0% 3.5%

• Quantum mechanics

3.0%

• Poisson Distribution

2.5%

• Noise = # Photons

2.0% 1.5%

• 95% 2

1.0%

• 1:44 more than 1.20

0.5%

• 1:57 less than 0.80

0.0% 0

20

40

60

80

100

120

140

160

#photons

AST Research & Innovation France

CMOS Sensors – Photon Shot Noise Do Not Copy

• Full Well Capacity limits Signal-Noise ratio Full Well = maximum #electrons: Vmax=Qmax/C Signal Noise Ratio [dB] = 20 × log10(#electrons) C

#electrons

Noise e

SNR dB

1.6E-18

10

3.16

10

16E-18

100

10.00

20

160E-18

1000

31.62

30

1.6E-15

10000

100.00

40

16E-15

100000

316.23

50

160E-15

1000000

1000.00

60

Q = C*V, Assume 1V swing AST Research & Innovation France

ST Microelectronics 02/Mars/2011

14

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Dark Current Do Not Copy

What Is Dark Current? • Ideal Diode Characteristics • Real Diode Characteristics • Ideal Photodiode Characteristics • Real Photodiode Characteristics

AST Research & Innovation France

44

CMOS Sensors – Standard Diodes (1) Do Not Copy

Ideal Diode Characteristics I

Current 0

Reverse Biased 0 Current

Current Infinite

0

Forward Biased  Current

Infinite

V

+

+

Ideal Diode Reverse Biased

Zero Current

Ideal Diode Forward Biased

Infinite Current

IV Curve

AST Research & Innovation France

45

CMOS Sensors – Standard Diodes (2) Do Not Copy

Real Diode Characteristics I

Current 0

Reverse Biased

Current Infinite

0

Infinite

Forward Biased V

+

Leakage Current

+

Real Diode Reverse Biased Some current

Leakage Current

Real Diode Forward Biased Current not infinite

Large Current

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

IV Curve

46

15

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Dark Current (1) Do Not Copy

Ideal Photodiode Characteristics Current 0

Current Infinite

0

+

Current Infinite

0

+

Ideal Photo Diode No Light, No current

Infinite

+

Ideal Photo Diode Some Light, Some current

Ideal Photo Diode More Light, More current

Photocurrent  Photon Flux

AST Research & Innovation France

47

CMOS Sensors – Dark Current (2) Do Not Copy

Real Photodiode Characteristics Current 0

Current Infinite

0

+

Infinite

+

Real (Photo) Diode No Light, Some current

Dark Current

Real Photo Diode Some Light, More current

Photocurrent  Photon Flux + Dark Current Dark Current == Leakage Current I.e. photocurrent that flows without any light. AST Research & Innovation France

48

CMOS Sensors – Dark Current (3) Do Not Copy

Pure Silicon Crystal

Si

Si Si

Si

Si Si

Si

Si

Work Function : Energy Required to liberate electron

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

49

16

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Dark Current (4) Defect in Silicon Crystal

Si

Si

Si Si

Si

Do Not Copy

Si P

Si

Si

Si

Si Si

Phosphorous Doping

Si Si

Si

Si

Si

Stress decreases Work Function  Less energy required to liberate electron  Thermal emission

AST Research & Innovation France

50

CMOS Sensors – Dark Current Do Not Copy

AST Research & Innovation France

51

CMOS Sensors – Dark Current Do Not Copy

Dark Current DOUBLES Every 8°C (approx)

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

17

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Dark Current Do Not Copy

Dark Current Vs Temperature

Dark Current [ a aA/(µm)² ]

300 250 200 150 100 50 0 0

20

40

60

80

100

Temperature [ °C ]

AST Research & Innovation France

CMOS Sensors – Dark Current Do Not Copy

Dark Current • • • • • • •

Limits low – light performance ("sensitivity") Complex problem Leakage inherent to semiconductor junctions Generated by impurities in semiconductor Extremely sensitive to temperature Simple model – doubles every 8°C Complex model – many sources, even processes which are good at low temperature are poor at high. AST Research & Innovation France

CMOS Sensors – Dark Current Do Not Copy

Dark Current 3 Issues: 1. Magnitude – it makes the image brighter  reduced dynamic range in the pixel pixel. 2. Pixel-Pixel variation – Some pixels will have higher dark current  white dots in the scene 3. Variation with time  white dots will be noisy 4. Distribution NOT mean value, important mean=1

mean=.002

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

18

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Dark Current Do Not Copy

Dark Current – Compensation Component Time invarying Time varying

Compensation Calibrate using dark image No calibration possible

Note: dark image must have • same temperature • same integration time as image AST Research & Innovation France

CMOS Sensors – Non Idealities kTC = Reset Noise Do Not Copy

Correlated Double Sampling (CDS): Vreset_noise = (kT/C) k= Boltzmann constant = 1.38 E-23 m kg s (Reset + Uncertainty) - (Measure + Uncertainty) = Signal (uncertainty removed!) 2

-2K-1

AST Research & Innovation France

CMOS Sensors – Double Sampling (1) Do Not Copy

RST

After Reset

After Reset

Vpd Before Reset

Before Reset

Output = Before Reset – After Reset • But Which “After Reset” Signal to use?

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

19

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – CDS & Double Sampling (2) Do Not Copy

RST Vpd

Correlated Double Sampling (CDS) Removes Reset Noise & Offsets L Long (F (Frame)) M Memory R Required i d

RST Vpd

Double Sampling Removes Offsets but √2 reset noise! Short Memory Required AST Research & Innovation France

CMOS Sensors – CDS & Double Sampling (3) Do Not Copy

ST (4 Transistor Pixel) Terms Vs. Classical Terminology ST

Classical

Two Capacitors

Single Capacitor (Photodiode)

• Photodiode • Sense Node

Cpd

Cpd

Csn

AST Research & Innovation France

CMOS Sensors – CDS & Double Sampling (4) Do Not Copy

ST (4 Transistor Pixel) Terms Vs. Classical Terminology ST

What

Why

Classical

CDS

Measure Floating Diffusion “quickly”

Remove system offsets. Good PSRR

Double Sampling

Measure Pixel & store in RAM

Remove reset noise

(in column)

Double Read

CDS

(frame CDS)

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

20

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Fully Depleted Diode Do Not Copy

Full Reset: No uncertainty = No kTC noise • Special manufacturing process  "Pinned" photodiode  special technology AST Research & Innovation France

CMOS Sensors – Design Do Not Copy

Sensor Design  Complex p – Separate p Section!

AST Research & Innovation France

CMOS Sensors – Characterization Do Not Copy

Characterization  Need to measure what we've designed / made  Good numbers ≠ g good p picture!

Good

Bad?

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

21

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Characterization Do Not Copy

Characterization      

Spectral Response / Quantum Efficiency Dark Current kTC Noise Readout Noise Full Well Capacity / Photon Shot Noise Fixed Pattern "Noise" / PRNU

AST Research & Innovation France

CMOS Sensors – Characterization Do Not Copy

Characterization  Acquire Images  Observe Images  Produce Statistics  PCIView / XLS  Modular Test

AST Research & Innovation France

CMOS Sensors – Characterization Do Not Copy

Characterization  Assuming Noise Sources independent (= uncorrelated)  sensor= (PSN2 + kTC2 + Readout2 + External2 + IDark2 )  Vary: ???

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

67

22

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Characterization Do Not Copy

Characterization  Assuming Noise Sources independent (= uncorrelated)  sensor= (PSN2 + kTC2 + Readout2 + External2 + IDark2 )  Vary: Light, Time (Exposure), Temperature, (Voltage)

AST Research & Innovation France

CMOS Sensors – Characterization QE/Spectral Response Quantum Efficiency

Do Not Copy

•Measure output at different wavelengths •Compare against calibrated reference

AST Research & Innovation France

CMOS Sensors – Characterization IDark Dark Current

Do Not Copy

• No Light  No Photon Shot Noise • Acquire Images with different integration times (exposures) • Acquire Images at different temperatures • Calculate Average, Tint=0, Tint=100ms • Idark = (X100ms – X0)/100m ADC codes / second • Calculate noise (standard deviation), Tint=0, Tint=100ms • Idark_noise (100ms) = (100ms2 - 02) •Dark current noise complex function of time! • Characterize & plot AST Research & Innovation France

ST Microelectronics 02/Mars/2011

23

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Characterization Read Noise Readout Noise + External Noise

Do Not Copy

• No Light  No Photon Shot Noise • Acquire Images with different 0 exposures  No Idark • Reset & Readout noise • Readout twice per frame (reset ; read ; read). {CDS} • Image1 = (kTC0 + Read1) • Image2 = (kTC0 + Read2) • Read Error Image = [Image1 – Image2] (e.g. 640*480 array) • Readout Noise =  (Read Error Image) / 2 • Can't easily separate Readout noise & External noise AST Research & Innovation France

CMOS Sensors – Characterization kTC Noise Reset (kTC) Noise

Do Not Copy

• No Light  No Photon Shot Noise • Acquire Images with different 0 exposures  No Idark • Reset & Readout noise • Readout two (or more) frames (reset0 , read1 ; reset2 , read3). • Image1 = (kTC0 + Read1) • Image2 = (kTC2 + Read3) • Image = [Image1 – Image2] (e.g. 640*480 array) • Reset Noise =  ( (Image)2 – (Readout Noise)2 ) AST Research & Innovation France

CMOS Sensors – Input Referred Noise Output Noise Vs. Input referred Noise

Do Not Copy

• Can only measure output, but wish to compare pixels • 'Reverse' signal flow & divide by gains to reach pixel noise • Vout = Vpix * 2 * 3 • Vpix = Vout /(2 * 3) • Qpix = Vpix * Cpix • Qpix = Vout * Cpix / gain  Vout = Qpix * gain / Cpix

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

24

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Characterization PSN / FW Photon Shot Noise / Full Well Capacity

Do Not Copy

• 'Constant' Light (set so ~max output at max exposure) • Acquire lots of Images with different exposures • Calculate 0ms, 10ms, 20ms, 30ms, 40ms, maxms, Measured Noise Vs Light

• FullWell2 = Dark2+ (Photon Shot)2

Noise [ADC counts]

1.2 1.0

• FullWell2 = 0ms 2 + PSN2

0.8 0.6

 PSN2 = FullWell 2 - 0ms2

0.4 0.2 0.0 0

20

40

60

80

100

Integration Time [ms]

AST Research & Innovation France

CMOS Sensors – Characterization PSN / FW #2 Photon Shot Noise / Full Well Capacity

Do Not Copy

• Q = V*C or V=Q/C or C=Q/C

[Q  electric charge, C  capacitance]

• Q = Ne * e

[Ne #electrons, e=1.6E-19 Coulombs]

• VSwingMax = NeFullWell * e / Cpix • VNoise = NeFW * e / Cpix

[Poisson distribution off photons]

 NeFW = (VSwingMax / VNoise )2  Cpix = VSwingMax * e / VNoise2 • Typical Cpix = 10fF • Minimum measurable Cpix (electrical) 10pF • Pixel Capacitance Measured using Optical Methods AST Research & Innovation France

CMOS Sensors – Characterization PSN / FW #3 Do Not Copy

Photon Shot Noise / Full Well Capacity Worked Example

• VSwingMax = 256 ADC counts = 1V

Measured Noise Vs Light

• VNoise = 1 counts

Noise [ADC counts]

1.2 1.0

• NeFW = (VSwingMax / VNoise )2

0.8 0.6

• NeFW = (256 / 1)2 = 64k electrons

0.4 0.2 0.0 0

20

40

60

80

Integration Time [ms]

100

• Cpix = 1V * e / (1V/256)2 = 64E3 * 1.6E-19 = 10.5fF

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

25

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – PRNU #1 Do Not Copy

Fixed Pattern "Noise" PRNU

1

2

• Which Image Is Better? AST Research & Innovation France

CMOS Sensors – PRNU #2 Do Not Copy

Fixed Pattern "Noise" PRNU

PRNU distributed randomly

PRNU aligned to column

• Both images have same level of PRNU AST Research & Innovation France

CMOS Sensors – PRNU #3 Fixed Pattern "Noise" FPN

Do Not Copy

a.k.a. PRNU (Photo Response Non-Uniformity)

• Fixed – not random  not really noise • Fixed on one pixel / one part, varies pixel-pixel, varies part-part • Fixed with time (seconds) • Vary with temperature, voltage • Human eye sensitive to FPN in an image especially if in a pattern (e.g. vertical column) AST Research & Innovation France

ST Microelectronics 02/Mars/2011

26

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – PRNU #4 Do Not Copy

Fixed Pattern "Noise" FPN

• Cpd = Gain, Cpd = Gain Error (Photo Response Non-Uniformity)

• Vt  Offset, Vt = Offset Error (FPN) AST Research & Innovation France

CMOS Sensors – PRNU #5 Do Not Copy

Fixed Pattern "Noise" FPN = Photo Response Non Uniformity (PRNU) Gain Error

140

140

120

120

100

100 Outpu ut

Outpu ut

Offset

80 60 40

80 60 40

20

20

0

0 0

20

40

60

80

100

0

20

40

Light

60

80

100

Light

• Offset error (FPN) much worse than Gain error (PRNU) • All CMOS sensors have offset correction (e.g. CDS) • Typical Gain error not visible to human. Lost in PSN AST Research & Innovation France

CMOS Sensors – FPN Characterization PRNU Characterization

Do Not Copy

• Flat (homogeneous) light source • Acquire lots of images  average image [reduce noise] • PRNU array: compare pixel with neighbours: PRNU(x,y) = e – Average(a,b,c,d,f,g,h,i,j) PRNUimage = Average PRNU(x,y)

AST Research & Innovation France

ST Microelectronics 02/Mars/2011

27

CMOS Image Sensing part II

06/03/2011

Misleading Datasheet (Industry Standard Practice!)

Do Not Copy

How Many Lines?

(a)

(b)

AST Research & Innovation France

CMOS Sensors – Resolution Do Not Copy

Misleading Datasheet (1) – Lines of Resolution

1

1 3 Film –2 3 line pairs

2

3

4

5

6

Semiconductor – 6 lines (always vertical, e.g. VGA = 640480  640 lines)

AST Research & Innovation France

CMOS Sensors – Resolution Vs Cost Do Not Copy

Misleading Datasheet (2) – Number of Pixels Is 4Mpix camera 4× better than 1Mpix? 5

Mp pix

4 Resolution

3

# Pix Size

2

Cost

1

2400 1200 0

500

1000

1500

2000

2500

Number of Lines

 Cost, #Pixels fn(Area) Square law  Size, Resolution fn(X) Linear  Smaller Pixels = Greater Photon Shot Noise AST Research & Innovation France

ST Microelectronics 02/Mars/2011

28

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Inch Format Misleading Datasheet (3) – One Inch  

Do Not Copy

1 inch (SI) = 25.4mm 1 inch (camera) = 16mm

25.4mm

16mm

F Format t Diagonal Di l (4:3 (4 3 aspectt ratio) ti ) 1"

16mm

½"

8mm

¼"

4mm

1/

2mm

8"

1960's technology Plumbicon Tube AST Research & Innovation France

CMOS Sensors – Photon Shot Noise Do Not Copy

Misleading Datasheet (4) – Photon Shot Noise • Photon Shot Noise not generated by sensor • Noise figures usually measured in the dark • No Photon Shot noise  Photon Shot Noise NEVER* given on Data Sheet * Occasionally SNR measured at ½ Well. (e.g. SMIA spec) AST Research & Innovation France

CMOS Sensors – Characterization Summary Do Not Copy

Characterization Summary • Can measure Spectral Response, Dark Current, Readout Noise, Reset Noise, Photon Shot Noise, FPN • Dominating Figure (light) = Photon Shot Noise • Data sheet never/rarely includes Photon Shot Noise • Difficult to correlate sensor problems with application requirements  (CMOS) Image Sensor Data Sheets useless!!!! Need to know Full Well Capacity  Sensor SNR AST Research & Innovation France

ST Microelectronics 02/Mars/2011

29

CMOS Image Sensing part II

06/03/2011

CMOS Sensors – Datasheet Specifications Datasheet Specifications • Signal / Noise Ratio (SNR)1 •Signal / Noise Ratio (SNR)2

Do Not Copy

Max Signal / Dark Noise (No Light = No PSN)!! Max Signal / Noise at 50%

• Dynamic Range (DNR) (intrascene – 1 image) = SNR1 • Dynamic Range (DNR) (interscene = 2+ images) = SNR1 + (A)EC • Input referred noise (electrons) = output noise / gain (A)EC – (Automatic) Exposure Control = Electronic Shutter AST Research & Innovation France

CMOS Sensors – Datasheet Specifications #2 Do Not Copy

Datasheet Specifications – Worked example • Sensor has 1V swing = 10 bit digital output • Exposure control – 1ms to 200ms • Pixel Capacitance of 10fF • Measurement: With no light, 2 codes ( = rms) noise • SNR = 210 / 2 = 512  20 * log(512) = 54dB • Interscene DNR = 512 * 200ms/1ms = 102400 = 100dB • Assuming Analogue Gain 1, at pixel, 1V/512 = 2mV noise • Q = C*(V/gain) = Ne * e  Ne = C*V/(e * gain) •10E-15 * 2E-3 / 1.6E-19 = 125 electrons input referred noise AST Research & Innovation France

CMOS Sensors – Typical Performance Specifications

Do Not Copy

Typical Performance Specifications* Pixel Size

7.5m

Signal/Noise

56dB

Exposure Control

81dB

Random Noise

1 17mV 1.17mV

Sensitivity (Green)

2.1V/lux.sec

Sensitivity (Mono)

6V/lux.sec

Dark Current

46mV/sec

Supply Voltage

3.0V..6.0V

Supply Current