From Kybernetes, vol. 33, no. 1, 2004, pp. 155-157
Analog VLSI: Circuits and Principles
by Shih-Chii Liu, Jörg Kramer, Giacomo Indiveri, Tobias Delbrück, and Rodney Douglas
with contributions from Albert Bergemont, Chris Diorio, Carver A. Mead, Bradley A. Minch, Rahul Sarpeshkar, and Eric Vittoz, and foreword by Carver A. Mead
MIT Press, Cambridge, Massachusetts (Bradford Book), 2002, xx + 434 pp., ISBN 0-262-12255-3, hardback, £39.95
This book introduces an important area of technology that will be unfamiliar to most readers- at least, I have to admit it was new to me. The book not only introduces the technology and explains its significance, it also reviews the underlying physics and the technical developments and possibilities in considerable detail. There is even reference to means of fabrication of the necessary VLSI circuit boards, including their production in modest quantities for research and development.
Despite the large number of contributing authors the style is uniform as though from a single author, suggesting that rather firm discipline was imposed during its preparation, and resulting in a coherent text that is well suited to be the basis for a taught course. The material has in fact been used in teaching in the Institute for Neuroinformatics associated with the University and Technical High School in Zurich, and in a series of Neuromorphic Engineering Workshops held in Telluride, Colorado.
In the Introduction it is mentioned that much relevant information, including examples of simulations and layouts for simple circuits, can be found on the Institute of Neuroinformatics website. The circuits were fabricated and used in taught laboratory courses, using design tools available commercially from a company Tanner Research Inc. There is also public domain software produced by John Lazzaro and David Gillespie.
The Internet addresses are not given in the book, but can easily be found using the Google search facility. For the Institute of Neuroinformatics it is: http://www.ini.unizh.ch/ and a link from this leads to information on the Workshops in Telluride (as well as on this attractive resort at 2670 metres in the Rockies). The company has address: http://www.tanner.com/ and details about the public domain software can be found at: http://www.cs.berkeley.edu/~lazzaro/chipmunk/.
The main reason that analog VLSI is exciting is indicated by the references to neuroinformatics and neuromorphic engineering. Biologically-inspired schemes for information processing generally demand distributed analog memory, which can be provided only clumsily by digital circuitry. Analog VLSI offers the possibility of compact and economic implementation of very large artificial neural nets, or "natural" computing systems. This is not the only reason for interest, since the transistor was first introduced as an analog amplifier and continues to be used as such in many devices that are likely to embody VLSI. The point is also made that circuitry designed for digital operation inevitably also has analog character, and as operating speeds are pushed higher and higher the fraction of total time that such circuits spend in transition between stable states must also increase, and analysis of behaviour increasingly requires attention to analog aspects.
The past emphasis on digital operation in VLSI development may have been encouraged by the observation that any computation can be approximated digitally with arbitrary precision. This is tacitly acknowledged by the provision of digital simulation software, but a serial digital implementation is likely to be slow and fragile as a means of achieving an effect best represented by a neural net.
The book has four parts. The first has the heading "Silicon and Transistors" and in its first chapter the underlying physics of semiconductor devices is reviewed. In the next chapter specific attention is given to metal-oxide-silicon field effect transistors, or MOSFETs, which provide a suitable basis for analog operation. In the next-again chapter there is further specialisation to what are termed floating-gate MOSFETs, which embody a conducting element that is totally surrounded by what is normally insulating material, but sufficiently thin that charge can pass by either the quantum "tunnel" effect or another form of injection.
The floating-gate MOSFET allows non-volatile analog storage with the possibility of incrementation (and decrementation). Not surprisingly, the term ôsynapse transistorö is applied and silicon learning arrays are described.
The second and third parts of the book describe, in full detail with the necessary mathematical analysis, a great many useful and ingenious circuits using analog transistors. The second part deals with static circuit behaviour and includes a systematic means of designing circuits with a given transfer characteristic. The third part treats dynamic behaviour, including techniques for integration and differentiation. A final chapter within this part deals with photosensors, which are a common form of analog device, often fabricated in arrays for use in electronic photography (which becomes digital photography with subsequent processing).
An interesting point is that imaging arrays are invariably scanned since the number of output leads is much smaller than the number of receptors. However, the use of analog VLSI can allow what is termed "focal-plane processing" of the unscanned data. This could emulate the processing that is known to occur in the layers of the human or animal retina, so as to perform such operations as contour enhancement and movement detection.
The fourth part of the book treats some special topics, including noise in transistor circuits, and means of fabrication of analog VLSI boards.
The treatment is definitely "meaty" with quite a sprinkling of equations, and clearly provides a solid foundation for further development. It is a little strange to find neural circuitry discussed without reference to McCulloch and Pitts, or Bernard Widrow or Frank Rosenblatt or various other names that come to mind, though it is probably reasonable since the emphasis here is on the means of realisation rather than the broad principles.
The presentation of clever circuit arrangements is nostalgically reminiscent of the days when the possibilities of pulsed electronic circuitry were first explored to meet the needs of radar, and then of computers and television, when the books of Puckle (1943) and Chance (1949) were standard works. I have the feeling that the new book is similarly opening up an exciting new era.
Alex M. Andrew
Chance, Britton (1949)(ed) Waveforms, Radiation Laboratory Series, McGraw-Hill, New York
Puckle, O.S. (1943) Time Bases (Scanning Generators), Chapman & Hall, London
Back to Reviews