Competitive Spike Pattern Detection for Neuromorphic systems

抄録

Development of neuromorphic chips with biologically plausible learning mechanisms is vital for investigating brain-like learning processes. One such mechanism is Spike-timing dependent plasticity (STDP), but implementing its multi-bit circuitry requires significant silicon area. In a prior study, we introduced a hardware-friendly learning rule named adaptive STDP. Through experiments, we demonstrated its performance similarity to the ideal STDP rule in a basic biologically plausible spike pattern detection task involving a single neuron. Building upon this, our present study extends the adaptive STDP learning rule to encompass lateral inhibition, a prevalent motif in the brain. We apply it to a spike pattern detection model featuring multiple neurons that engage in competition to detect multiple patterns. Furthermore, we investigate the performance of the ideal STDP rule using 4-bit and 6-bit synapse resolution and present a comparative analysis of the results.