This book brings together in one place important contributions and state-of-the-art research in the rapidly advancing area of analog VLSI neural networks. The book serves as an excellent reference, providing insights into some of the most important issues in analog VLSI neural networks research efforts.

Research presented in this thesis provides a substantial leap from the study of interesting device physics to fully adaptive analog networks and lays a solid foundation for future development of large-scale, compact, low-power adaptive parallel analog computation systems. The investigation described here started with observation of this potential learning capability and led to the first derivation and characterization of the floating-gate pFET correlation learning rule. Starting with two synapses sharing the same error signal, we progressed from phase correlation experiments through correlation experiments involving harmonically related sinusoids, culminating in learning the Fourier series coefficients of a square wave cite. Extending these earlier two-input node experiments to the general case of correlated inputs required dealing with weight decay naturally exhibited by the learning rule. We introduced a source-follower floating-gate synapse as an improvement over our earlier source-degenerated floating-gate synapse in terms of relative weight decay cite. A larger network of source-follower floating-gate synapses was fabricated and an FPGA-controlled testboard was designed and built. This more sophisticated system provides an excellent framework for exploring applications to multi-input, multi-node adaptive filtering applications. Adaptive channel equalization provided a practical test-case illustrating the use of these adaptive systems in solving real-world problems. The same system could easily be applied to noise and echo cancellation in communication systems and system identification tasks in optimal control problems. We envision the commercialization of these adaptive analog VLSI systems as practical products within a couple of years.

This volume contains the proceedings of a workshop on Analog Integrated Neural Systems held May 8, 1989, in connection with the International Symposium on Circuits and Systems. The presentations were chosen to encompass the entire range of topics currently under study in this exciting new discipline. Stringent acceptance requirements were placed on contributions: (1) each description was required to include detailed characterization of a working chip, and (2) each design was not to have been published previously. In several cases, the status of the project was not known until a few weeks before the meeting date. As a result, some of the most recent innovative work in the field was presented. Because this discipline is evolving rapidly, each project is very much a work in progress. Authors were asked to devote considerable attention to the shortcomings of their designs, as well as to the notable successes they achieved. In this way, other workers can now avoid stumbling into the same traps, and evolution can proceed more rapidly (and less painfully). The chapters in this volume are presented in the same order as the corresponding presentations at the workshop. The first two chapters are concerned with fmding solutions to complex optimization problems under a predefmed set of constraints. The first chapter reports what is, to the best of our knowledge, the first neural-chip design. In each case, the physics of the underlying electronic medium is used to represent a cost function in a natural way, using only nearest-neighbor connectivity.

Learning on Silicon combines models of adaptive information processing in the brain with advances in microelectronics technology and circuit design. The premise is to construct integrated systems not only loaded with sufficient computational power to handle demanding signal processing tasks in sensory perception and pattern recognition, but also capable of operating autonomously and robustly in unpredictable environments through mechanisms of adaptation and learning. This edited volume covers the spectrum of Learning on Silicon in five parts: adaptive sensory systems, neuromorphic learning, learning architectures, learning dynamics, and learning systems. The 18 chapters are documented with examples of fabricated systems, experimental results from silicon, and integrated applications ranging from adaptive optics to biomedical instrumentation. As the first comprehensive treatment on the subject, Learning on Silicon serves as a reference for beginners and experienced researchers alike. It provides excellent material for an advanced course, and a source of inspiration for continued research towards building intelligent adaptive machines.

Adaptive Analog VLSI Neural Systems is the first practical book on neural networks learning chips and systems. It covers the entire process of implementing neural networks in VLSI chips, beginning with the crucial issues of learning algorithms in an analog framework and limited precision effects, and giving actual case studies of working systems. The approach is systems and applications oriented throughout, demonstrating the attractiveness of such an approach for applications such as adaptive pattern recognition and optical character recognition. Dr Jabri and his co-authors from AT&T Bell Laboratories, Bellcore and the University of Sydney provide a comprehensive introduction to VLSI neural networks suitable for research and development staff and advanced students.

When comparing conventional computing architectures to the architectures of biological neural systems, we find several striking differences. Conventional computers use a low number of high performance computing elements that are programmed with algorithms to perform tasks in a time sequenced way; they are very successful in administrative applications, in scientific simulations, and in certain signal processing applications. However, the biological systems still significantly outperform conventional computers in perception tasks, sensory data processing and motory control. Biological systems use a completely dif ferent computing paradigm: a massive network of simple processors that are (adaptively) interconnected and operate in parallel. Exactly this massively parallel processing seems the key aspect to their success. On the other hand the development of VLSI technologies provide us with technological means to implement very complicated systems on a silicon die. Especially analog VLSI circuits in standard digital technologies open the way for the implement at ion of massively parallel analog signal processing systems for sensory signal processing applications and for perception tasks. In chapter 1 the motivations behind the emergence of the analog VLSI of massively parallel systems is discussed in detail together with the capabilities and !imitations of VLSI technologies and the required research and developments. Analog parallel signal processing drives for the development of very com pact, high speed and low power circuits. An important technologicallimitation in the reduction of the size of circuits and the improvement of the speed and power consumption performance is the device inaccuracies or device mismatch.

Engineers have long been fascinated by how efficient and how fast biological neural networks are capable of performing such complex tasks as recognition. Such networks are capable of recognizing input data from any of the five senses with the necessary accuracy and speed to allow living creatures to survive. Machines which perform such complex tasks as recognition, with similar ac curacy and speed, were difficult to implement until the technological advances of VLSI circuits and systems in the late 1980's. Since then, the field of VLSI Artificial Neural Networks (ANNs) have witnessed an exponential growth and a new engineering discipline was born. Today, many engineering curriculums have included a course or more on the subject at the graduate or senior under graduate levels. Since the pioneering book by Carver Mead; "Analog VLSI and Neural Sys tems", Addison-Wesley, 1989; there were a number of excellent text and ref erence books on the subject, each dealing with one or two topics. This book attempts to present an integrated approach of a single research team to VLSI ANNs Engineering.

Intelligent/smart systems have become common practice in many engineering applications. On the other hand, current low cost standard CMOS technology (and future foreseeable developments) makes available enormous potentialities. The next breakthrough will be the design and development of "smart adaptive systems on silicon" i.e. very power and highly size efficient complete systems (i.e. sensing, computing and "actuating" actions) with intelligence on board on a single silicon die. Smart adaptive systems on silicon will be able to "adapt" autonomously to the changing environment and will be able to implement "intelligent" behaviour and both perceptual and cognitive tasks. At last, they will communicate through wireless channels, they will be battery supplied or remote powered (via inductive coupling) and they will be ubiquitous in our every day life. Although many books deal with research and engineering topics (i.e. algorithms, technology, implementations, etc.) few of them try to bridge the gap between them and to address the issues related to feasibility, reliability and applications. Smart Adaptive Systems on Silicon, though not exhaustive, tries to fill this gap and to give answers mainly to the feasibility and reliability issues. Smart Adaptive Systems on Silicon mainly focuses on the analog and mixed mode implementation on silicon because this approach is amenable of achieving impressive energy and size efficiency. Moreover, analog systems can be more easily interfaced with sensing and actuating devices.