A 200x100 Array of Electronically Calibratable Logarithmic ...

A 200x100 Array Of Electronically Calibratable Logarithmic CMOS Pixels Bhaskar Choubey Satoshi Ayoma Dileepan Joseph Stephen Otim Steve Collins University of Oxford, Oxford Satoshi Ayoma is currently with Renesses Technology Corp, Japan Dileepan Joseph is currently with University of Alberta, Edmonton, Canada Introduction

Need for High Dynamic Range Imaging Linear Pixel CCD CMOS Active Pixel Sensor Techniques to improve dynamic range lead to Low Fill Factor Complex In-pixel Circuits Large Number of Bits Logarithmic Pixels Can capture Wide Dynamic Range Encode Contrast

Information Similar Fill Factor as that of APS Randomly Addressable Device Variations Cause Fixed Pattern Noise Ways to Reduce FPN On-chip Methods Electron Injection (Ricquer and coworkers) Gate Voltage Adjustment (Loose et.al.) Subtracting Reference Response (Kavadias et. al.)

Off-chip Methods Subtracting Reference Response (IMSC chips) Model Based Correction (Joseph and Collins) Electronic Calibration (Choubey and coworkers) Electronic Calibration Log Pixel Response y a b ln( x c) Parameter Extraction

y1 y2 b ln( x1 / x2 ) a y1 b ln( x1 ) yd a c exp b Photocurrent Extraction yij aij

xij exp b ij yij yij1 x1 exp ln( x1 / x2 )

y y ij1 ij 2 Chip Design 0.35 2-Poly 3-Metal Poly 3-Poly 3-Metal Metal Process Each Pixel 10 x 10 Photodiodes used : N-Poly 3-Metal diff, P-Poly 3-Metal substrate 200 x 100 Pixel array Column and Row

Scanners Experimental Procedure All Clocks were generated externally using Instrunet data acquisition board and HP Vee programming interface Analogue values from Pixels were read through semiconductor parameter analyzer (Agilent 4155B) Slow speed clock was used to remove all transient effects All electronic experiments were carried out in a stable temperature oven to remove temperature

variability. Experimental Result A uniform scene without correction, after single parameter correction and after 2 parameter correction. Experimental Result Experimental Results Uncorrected FPN was of the order of 104%

Offset FPN Correction FPN is dramatically reduced However a low error is only obtained close to the calibration point. Experimental Results Calibration Points The two points should be far from dark point to remove leakage effects. They should be well separated to remove

temporal noise effects on gain. The second point should not be too close to the moderate inversion region. FPN Correction A contrast sensitivity of 2% is obtained for more than 6 decades of intensity. Conclusion Log Pixels have the ability to match the

performance of human eye, but are crippled by high fixed pattern noise. Offset only correction fails to get a high quality image. Electronic Calibration provides an easy method of reducing the FPN caused by variation in all parameters. The residual error after correction is as low as 2% and hence matches the human eyes contrast sensitivity.