A purely capacitive synaptic matrix for fixed-weight neural networks

It is shown that the synaptic function of fixed-weight neural networks can be implemented using only one capacitor. The resulting synaptic matrix, being devoid of active devices, offers very high space-power efficiency and speed along with large synapse capacity with considerable analog depth. The generic capacitor matrix is analyzed on the basis of dendritic charge conservation. The results are used to determine network limitations and to design a double-poly CMOS feedforward classifier that is capable of correcting any 3-b error occurring in a set of thirty 16-b code-patterns. Each synapse occupies 16.5 mu m*10 mu m of field-oxide space for the very conservative 3- mu m rules employed in this particular design. Electrical performance is verified through simulation. Comparison between the proposed network and other switched-capacitor neural network configurations is also included.