A programmable neural-fuzzy processor for handwritten digit classification
Guoxing Li, Bingxue Shi · 2000
A novel current-mode processor based on single layer perceptron and fuzzy logic for handwritten digit classification is put forward in this paper. This processor can be reconfigured as perceptron based classifier or fuzzy logic based classifier, it benefits from the fact that there is a similar architecture between single perceptron and SUM-MAX based fuzzy logic when they are used as a classifier. Both of them share the 11/spl times/10 PE (processing element) array, WTA (winner-take-all) network, switched current integrators and also I/O ports. It can give 11 binary final classifying results, of which one is regarded as rejecting signal, after the feature vector with length variability is fed into the PE array. This processor has been implemented with double metal single poly 1.2 /spl mu/m CMOS technology.