A Dummy Cell Added Neural Network Using in Pattern Recognition for Prevention of Failed Events

Abstract

Brain-inspired neuromorphic computing systems are receiving significant attention. A typical neuromorphic computing system is the neuron network, whose basic performance is the integrate-and-fire operation. However, latency issues can occur if the integrated signal is not sufficient during the integration process, the integration time is too long, or no firing occurs. In this paper, we propose a dummy cell added neural network to ensure complete I & F operation. The dummy cell compensates the weak signals to ensure a complete I & F operation and to modulate the integration time; but makes negligible influence on the strong signals. The firing rate of a weak signal increases from 80% to 100%. Finally, we analyzed the external area consumption of dummy cells, it can be reduced as small as a few thousandths with large number of input neurons. This proposed scheme can be used in pattern recognition to increase reliability and modulate the integration time.