Applying non-ideal mixed analog-digital multipliers in electronic processing circuits based on neural networks
C. Lajusticia, Nicolás J. Medrano, Guillermo Zatorre, B. Martin · 2004
This paper analyses effects of component mismatching and non-linearities in mixed analog-digital multipliers used in analog implementation of artificial neural networks. Function estimation and classification problems using a multilayer perceptron are studied, analyzing their results, accuracy requirements and network modifications required.