KLUWER ACADEMIC PUBLISHERS

Transcription

1 CMOS IMAGERS

2 CMOS Imagers From Phototransduction to Image Processing Edited by Orly Yadid-Pecht Ben-Gurion University, Beer-Sheva, Israel and Ralph Etienne-Cummings Johns Hopkins University, Baltimore, U.S.A. KLUWER ACADEMIC PUBLISHERS NEW YORK, BOSTON, DORDRECHT, LONDON, MOSCOW

3 ebook ISBN: Print ISBN: Springer Science + Business Media, Inc. Print 2004 Kluwer Academic Publishers Dordrecht All rights reserved No part of this ebook may be reproduced or transmitted in any form or by any means, electronic, mechanical, recording, or otherwise, without written consent from the Publisher Created in the United States of America Visit Springer's ebookstore at: and the Springer Global Website Online at:

4 Dedication To our loved ones.

5 Contents Dedication Contributing Authors Preface Introduction v ix xi xiii 1. Fundamentals of Silicon-Based Phototransduction HONGHAO JI AND PAMELA A. ABSHIRE 1.1 Introduction Background physics of light sensing Silicon-based photodetectors Semiconductor image sensors Information rate Summary CMOS APS MTF Modeling IGOR SHCHERBACK AND ORLY YADID-PECHT 2.1 Introduction Experimental details Physical analysis The unified model description Results and discussion Summary 72

6 viii CMOS Imagers: From Phototransduction to Image Processing 3. Photoresponse Analysis and Pixel Shape Optimization for CMOS APS IGOR SHCHERBACK AND ORLY YADID-PECHT 3.1 Introduction Photoresponse model Comparison with experimental results CMOS APS pixel photoresponse prediction for scalable CMOS technologies Summary Active Pixel Sensor Design: From Pixels to Systems ALEXANDER FISH AND ORLY YADID-PECHT 4.1 Introduction CMOS image sensors APS system-on-a-chip approach Summary Focal-Plane Analog Image Processing MATTHEW A. CLAPP, VIKTOR GRUEV, AND RALPH ETIENNE-CUMMINGS 5.1 Introduction Current-domain image processing: the general image processor Voltage-domain image processing: the temporal difference imager Mixed-mode image processing: the centroid-tracking imager Conclusions CMOS Imager Non-Uniformity Correction Using Floating-Gate Adaptation MARC COHEN AND GERT CAUWENBERGHS 6.1 Introduction Adaptive non-uniformity correction Canceling gain non-uniformity Intensity equalization Focal plane VLSI implementation VLSI system architecture Experimental results Discussion Conclusions 219 Appendix: List of Symbols 223 Index 232

7 Contributing Authors Pamela A. Abshire University of Maryland, College Park, MD, USA Gert Cauwenberghs Johns Hopkins University, Baltimore, MD, USA Matthew A. Clapp Johns Hopkins University, Baltimore, MD, USA Marc Cohen University of Maryland College, Park, MD, USA Ralph Etienne-Cummings Johns Hopkins University, Baltimore, MD, USA Alexander Fish Ben-Gurion University, Beer-Sheva, Israel Viktor Gruev Johns Hopkins University, Baltimore, MD, USA Honghao Ji University of Maryland, College Park, MD, USA Igor Shcherback Ben-Gurion University, Beer-Sheva, Israel Orly Yadid-Pecht Ben-Gurion University, Beer-Sheva, Israel

8 Preface The idea of writing a book on CMOS imaging has been brewing for several years. It was placed on a fast track after we agreed to organize a tutorial on CMOS sensors for the 2004 IEEE International Symposium on Circuits and Systems (ISCAS 2004). This tutorial defined the structure of the book, but as first time authors/editors, we had a lot to learn about the logistics of putting together information from multiple sources. Needless to say, it was a long road between the tutorial and the book, and it took more than a few months to complete. We hope that you will find our journey worthwhile and the collated information useful. The laboratories of the authors are located at many universities distributed around the world. Their unifying theme, however, is the advancement of knowledge for the development of systems for CMOS imaging and image processing. We hope that this book will highlight the ideas that have been pioneered by the authors, while providing a roadmap for new practitioners in this field to exploit exciting opportunities to integrate imaging and smartness on a single VLSI chip. The potential of these smart imaging systems is still unfulfilled. Hence, there is still plenty of research and development to be done. We wish to thank our co-authors, students, administrative assistants, and laboratory co-workers for their excitement and enthusiasm for being involved in this project. Specifically, we would like to thank Alex Belenky, Rachel Mahluf-Zilberberg, and Ruslan Sergienko from the VLSI Systems Center at Ben-Gurion University. We also would like to thank our mentors, Eric Fossum, Jan van der Spiegel, Albert Theuwissen, Mohammed Ismail, Dan McGrath, Eby

9 xii CMOS Imagers: From Phototransduction to Image Processing Friedman, Andreas Andreou, Norman Kopeika, Zamik Rosenwaks, Irvin Heard, and Paul Mueller for their support at different stages of this project. Furthermore, we would like to thank our copy-editor, Stan Backs of SynchroComm Inc. In addition, we would like to thank our publishers, Kluwer Academic Publishers, and especially Mark de Jongh for being patient with us all the way. Last but not least, we would like to thank our loved ones for their support during the process. We hope the missing hours with them are worth the result. Orly Yadid-Pecht and Ralph Etienne-Cummings

10 Introduction This book starts with a detailed presentation of the basic concepts of photo transduction, modeling, evaluation, and optimization of Active Pixel Sensors (APS). It continues with the description of APS design issues using a bottom-up strategy, starting from pixels and finishing with image processing systems. Various focal-plane image processing alternatives either to improve imaging or to extract visual information are presented. The book closes with a discussion of a completely non-traditional method for image noise suppression that utilizes floating-gate learning techniques. The final three chapters in fact provide a glimpse into a potential future of CMOS imaging and image processing, where concepts gleaned from other disciplines, such biological vision, are combined with alternative mixedsignal computation circuits to perform complex visual information processing and feature extraction at the focal plane. This benefit of CMOS imaging and image processing is still largely unexploited by the commercial sector. The first chapter reviews the background knowledge and concepts of silicon-based photo transduction, and introduces relevant concepts from semiconductor physics. Several silicon-based photo detectors are examined, including the photodiode and the photogate. This chapter also describes the operation of the charge-coupled device (CCD) imager, the predominant technology available for digital imaging. CCD technology is compared with a promising alternate technology, the APS imager. In addition, the functional performances of several basic pixel structures are compared by considering them as communication channels and determining their ability to convey information about an incident optical signal. At 30 frames per second, information rates are similar for charge-, voltage-, and current-mode pixels.

11 xiv CMOS Imagers: From Phototransduction to Image Processing Comparable trends are found for their information capacities as the photocurrent varies. The second chapter deals with the modulation transfer function (MTF) of an APS. MTF is one of the most significant factors determining the image quality. Unfortunately, characterization of the MTF of semiconductor-based focal-plane arrays (FPA) has been typically one of the more difficult and error-prone performance testing procedures. Based on a thorough analysis of experimental data, a unified model has been developed for estimation of the MTF of a general CMOS active pixel sensor for scalable CMOS technologies. The model covers the physical diffusion effect together with the influence of the geometrical shape of the pixel active area. Excellent agreement is reported between the results predicted by the model and the MTF calculated from the point spread function (PSF) measurements of an actual pixel. This fit confirms the hypothesis that the active area shape and the photocarrier diffusion effect are the determining factors of the overall MTF behavior of CMOS active pixel sensors, thus allowing the extraction of the minority-carrier diffusion length. The third chapter deals with photoresponse analysis and pixel shape optimization for CMOS APS. A semi-analytical model is developed for the estimation of the photoresponse of a photodiode-based CMOS APS. This model is based on a thorough analysis of experimental data, and incorporates the effects of substrate diffusion as well as geometrical shape and size of the photodiode active area. It describes the dependence of pixel response on integration photocarriers and on conversion gain. The model also demonstrates that the tradeoff between these two conflicting factors can lead to an optimal geometry, enabling the extraction of a maximal photoresponse. The dependence of the parameters on process and design data is discussed, and the degree of accuracy for the photoresponse modeling is assessed. The fourth chapter reviews APS design from the basics to more advanced system-on-chip examples. Since APS are fabricated in a commonly used CMOS process, image sensors with integrated intelligence can be designed. These sensors are very useful in many scientific, commercial and consumer applications. Current state-of-the-art CMOS imagers allow integration of all functions required for timing, exposure control, color processing, image enhancement, image compression, and analog-to-digital conversion (ADC) on the same die. In addition, CMOS imagers offer significant advantages and rival traditional charge-coupled devices in terms of low power, low voltage and monolithic integration. The chapter presents different types of CMOS pixels and introduces the system-on-chip approach, showing examples of two smart APS imagers: a smart vision system-onchip and a smart tracking sensor. The former is based on a photodiode APS with linear output over a wide dynamic range, made possible by random

12 CMOS Imagers: From Phototransduction to Image Processing xv access to each pixel in the array and by the insertion of additional circuitry into the pixels. The latter is a smart tracking sensor employing analog nonlinear winner-take-all (WTA) selection. The fifth chapter discusses three systems for imaging and visual information processing at the focal plane, using three different representations of the incident photon flux density: current-mode, voltagemode, and mixed-mode image processing. This chapter outlines how spatiotemporal image processing can be implemented in current and voltage modes. A computation-on-readout (COR) scheme is highlighted. This scheme maximizes pixel density and multiple processed images to be produced in parallel. COR requires little additional area and access time compared to a simple imager, and the ratio of imager to processor area increases drastically with scaling to CMOS technologies with smaller feature size. In some cases, it is necessary to perform computations in a pixelparallel manner while still retaining the imaging density and low-noise properties of an APS imager. Hence, an imager that utilizes both currentmode and voltage-mode imaging and processing is presented. However, this mixed-mode approach has some limitations, and these are described in detail. Three case studies show the relative merits of the different approaches for focal-plane analog image processing. The last chapter investigates stochastic adaptive algorithms for on-line correction of spatial non-uniformity in random-access addressable imaging systems. An adaptive architecture is implemented in analog VLSI, and is integrated with the photo sensors on the focal plane. Random sequences of address locations selected with controlled statistics are used to adaptively equalize the intensity distribution at variable spatial scales. Through a logarithmic transformation of system variables, adaptive gain correction is achieved based on offset correction in the logarithmic domain. This idea is particularly attractive for compact implementation using translinear floatinggate MOS circuits. Furthermore, the same architecture and random addressing provide for oversampled binary encoding of the image resulting in an equalized intensity histogram. The techniques can be applied to a variety of solid-state imagers, such as artificial retinas, active pixel sensors, and infrared sensor arrays. Experimental results confirm gain correction and histogram equalization in a pixel adaptive array. We hope this book will be interesting and useful for established designers, who may benefit from the embedded case studies. In addition, the book might help newcomers to appreciate both the general concepts and the design details of smart CMOS imaging arrays. Our focus is on the practical issues encountered in designing these systems, which will always be useful for both experienced and novice designers.