Results from Pulse-Stream VLSI Neural Network Devices
This paper describes a novel switched-capacitor design for the implementation of artificial neural networks in VLSI using the pulse-stream signalling mechanism and dynamic weight storage. Test results are presented from a small number of chips, paying particular attention to the synaptic weight linearity and storage time. The synaptic weights are fully-programmable and the VLSI chips can be used to process analogue sensor data in real time with an accuracy equivalent to 6 or 7 bits, as demonstrated in the robotics application described in the paper.
Unable to display preview. Download preview PDF.
- Graf, H.P., Jackel, L.D., Howard, R.E., Straughn, B., Denker, J.S., Hubbard, W.E., Tennant, D.M. and Schwartz, D., “VLSI implementation of a neural network memory with several hundreds of neurons ”, in Proc. AIP Conference on Neural Networks for Computing, Snowbird, pp. 182–187, 1986.Google Scholar
- Grossberg, S., “Some Physiological and Biochemical Consequences of Psychological Postulates”, Proc. Natl. Acad. Sci USA, pp. 758–765, 1968.Google Scholar
- Hamilton, A., Murray, A.F., Reekie, H.M. and Tarassenko, L., “Working Analogue Neural Network Chips”, in this volume.Google Scholar
- Murray, A.F., Hamilton, A. and Tarassenko, L., “Programmable Analog Pulse-firing Networks’, in Advances in Neural Information Processing Systems, Touretzky, D.S., Ed., Morgan Kaufmann, pp. 671–677, 1989.Google Scholar
- Tarassenko, L., Brownlow, M.J. and Murray, A.F., “VLSI neural networks for autonomous robot navigation”, in Proc. International Neural Network Conference,Paris, pp. 213–216,1990.Google Scholar
- Vittoz, E., Oguey, H., Maher, M.A., Nys, O., Dijkstra, E. and Chevroulet, M., “Analog Storage of Adjustable Synaptic Weights”, in Proceedings of 1st Int. Workshop on Microelectronics for Neural Networks, pp. 69–79, 1990.Google Scholar