An Electronic Calibration Scheme for Logarithmic CMOS Pixels

An Electronic Calibration Scheme for Logarithmic CMOS Pixels Bhaskar Choubey, Satoshi Ayoma*, Stephen Otim, Dileepan Joseph**, Steve Collins, University of Oxford, UK *Now with Renessas Technology Corp, Japan ** Now with University of Alberta, Edmonton, Canada Outline Introduction

FPN reduction Techniques High Gain Readouts to reduce effects of gain variations Electronic Calibration Scheme Experimental Results Conclusion Introduction Limited range of CCDs and CMOS APS Linear high dynamic range pixels require large number of bits with complex circuitry and/or post processing Logarithmic Pixels Human Eye like response

Encode Contrast Information Logarithmic Pixel Circuit Fixed Pattern Noise Device Variations cause Fixed Pattern Noise FPN reduces the contrast sensitivity of the image sensor, resulting in inferior images, compared to linear sensors Joseph and Collins Model of Logarithmic Pixel y a b log (x c) FPN Reduction Methods

In Pixel Techniques Hot Electron Effects (Ref. Ricquer and others) Gate Voltage Adjustment with feedback (Ref. Loose and others) Off Pixel Technique Subtraction of Uniform Scene Response (Ref. IMS Chips) Convenience Error Possibility Off Chip Technique On Chip current source for Double Sampling (Ref. Kavadias and others) Offset only

Remaining FPN ~ 15 % of one decade response FPN Reduction Methods Corrected Response ycorr b log( xin / xcorr ) Error in Corrected response ycorr b log( xin / xcorr ) Equivalent Percentage Error 100b k log( xin / xcorr )

b Electronic Calibration Joseph and Collins used 24 images and an iterative scheme Computation of thee parameters y2 y1 b log( x2 / x1 ) yd a c exp b

a y1 b log( x1 ) Circuit Implementation The Two Readouts Parameter Mean/SD SF Diff Amp

Offset (V) Mean SD 0.5524 0.0165 1.7430 0.0174 Gain (mV/decade)

Mean SD -56.87 1.084 -66.7 1.023 Bias (fA) Mean

SD 28.163 4.4974 26.188 6.3040 Residual FPN Pixel Current (A) Residual FPN

Contrast Sensitivity 3.55e-13 1.07e-12 3.91e-10 8.80e-10 3.25e-9 3.45e-8 1.57e-7 0.922 0.694 0.397

0.440 0.429 0.501 1.089 2.11 1.62 0.92 1.10 0.99 1.16 2.53 Comparison of Residual FPN

Conclusion Offset only FPN correction can not produce high quality image. Impact of gain variation can be reduced by having a high gain readout like differential readout. FPN can be reduced to a contrast sensitivity of <2% by electronic calibration scheme, comparable to the human eye.