A Multistable Circuit for Weight Storage in Analog Artificial Neural Networks

The aim of this paper is to address the problem of weight storage at the end of the learning phase, an important problem encoun- tered in analog implementation of artificial neural networks. Circuits for weight storage must satisfy essentially two requirements: they must be as simple as possible and they must be able to maintain the informa- tion indefinitely. In this paper we suggest a multilevel (or quantised) weight storage approach which can be easily implemented by standard low complexity circuitry and neither requires refreshing nor conversion.