An analog CMOS neural network with on-chip learning and multilevel weight storage
An analog neural network with four neurons and 16 synapses, fabricated in a 1.2 μm n-well single-polysilicon, double-metal process, is presented. The circuit solutions adopted, for on-chip learning and weight storage, particularly simple and silicon area-efficient, are capable of solving the main problems to the implementation of analog neural networks.
Unable to display preview. Download preview PDF.
- Y.Arima et al., “A Refreshable Analog VLSI Neural Network Chip with 400 Neurons and 40k Synapses”, IEEE J. Solid-State Circuits, p. 1854, Vol. 27, Dec. 1992, No. 17.Google Scholar
- B.L.Barranco et al., “Modular Analog Continuous-Time VLSI Neural Networks with On-Chip Hebbian Learning and Analog Storage”, IEEE ISCAS '92, S. Diego, 10–13 May 1992.Google Scholar
- T.Morie, Y.Amemiya, “An All-Analog Expandable Neural Network LSI with On-Chip Back-Propagation Learning”, IEEE J. Solid State Circuits, Vol. 29, No. 9, September 1994, pp.1086–1093.B17.Google Scholar
- Y.K.Choi, S.Y.Lee, “Subthreshold MOS Implementation of Neural Networks with On-Chip Error Back-Propagation Learning”, Proc. of I J.C.N.N., 1993, pp. 849–852.Google Scholar
- M.R.Belli, M.Conti, C.Turchetti, “Analog Brownian Weight Movement for Learning of Artificial Neural Networks”, Proc. of the European Symposium on Artificial Neural Networks ESANN'95, Brussels, p.75–80, April 19–21 1995.Google Scholar
- A.Papoulis, “Probability, random variables, and stochastic processes”, McGraw-Hill, New York, 1965.Google Scholar
- M.Conti, S.Orcioni, C.Turchetti, “A Class of Neural Networks Based on Approximate Identity for Analog IC's Implementation”, IEICE Trans. on Fundamentals, Japan, Vol. E77-A, No. 6, June 1994, pp. 1069–1079.Google Scholar
- M.H.Cohen, A.G.Andreou, “Current-Mode Subthreshold MOS Implementation of the Herault-Jutten Autoadaptive Network”, IEEE J. Solid State Circuits, Vol. 27, No. 5, May 1992, pp. 714–727.Google Scholar