Analog CMOS circuits for neural networks
Abstract
A compact analog complementary metal oxide semiconductor (CMOS) synaptic circuit with in-situ Hebbian learning was developed and used to construct a deterministic Boltzmann neural network. Investigations have shown that, because of the error compensating properties of the learning algorithm, neural network architectures such as the deterministic Boltzmann machine are robust even when constructed from compact, highly non-ideal analog CMOS components. The use of floating-gate field effect transistors for nonvolatile storage is also discussed.
- Publication:
-
Electrical and Computer Engineering, Volumes 1 and 2 4 p (SEE N93-30215 11-31)
- Pub Date:
- 1990
- Bibcode:
- 1990ecev.confQ....S
- Keywords:
-
- Analog Circuits;
- Cmos;
- Field Effect Transistors;
- Machine Learning;
- Neural Nets;
- Synapses;
- Algorithms;
- Data Storage;
- Errors;
- Electronics and Electrical Engineering