Synergistic Reconfiguration of Adaptive Precision Chemical Classifiers

Abstract

We present parallel implementations of a multilayer perceptron that uses reduced variable bit width hardware to improve resource utilization while still providing known levels of accuracy. We show results for a chemical classification application and introduce ways in which to take advantage of the capabilities of a reconfigurable device. We show how the optimized circuit can be used synergistically in parallel with other classifiers for added capability and alone for fault tolerance and saving power.