The multi-chip design of analog CMOS expandable modified Hamming neural network with on-chip learning and storage for pattern classification

Abstract

In this paper, a multi-chip expandable modified feedforward Hamming neural network for pattern classification is designed and implemented. In the proposed modified Hamming network, the outstar circuit is used to provide the on-chip learning capability. Moreover, the embedded ratio memory in the outstar circuit is used to store the learned pattern. The chips can be connected to form pattern, element, and pattern-and-element-mixed expansions. The experimental results have been correctly verified the operation of multi-chip expansion and classification function. The contrast enhancement characteristic of the stored pattern in the 3-chip element expansion has also been observed.