A switched-current synapse circuit for spiking neural networks based on compact structure and TLBO algorithm

An analog synapse circuit suitable for the VLSI implementation of spiking neural networks based on the compact structure switched-current (SI) filter and the teaching-learning-based optimization (TLBO) algorithm is presented. First, the spike-response function of the synaptic terminal is accurate approximated using TLBO algorithm. Then the approximated transfer function of the synapse circuit is implemented based on a compact z-domain structure filter using SI integrators and current mirrors as basic building blocks. Only by controlling the clock frequency of the designed synapse circuit, the time constant can be conveniently adapted without varying the circuit architecture. Finally, to illustrate the design procedure, a fifth-order approximated SI synapse circuit is given, and the sensitivity and imperfection of the designed SI synapse circuit are compared with the follow-the-leader-feedback (FLF) structure circuit. Simulation results confirm that the proposed method for implementing synapse circuit is effective.