Stochastic Synapse with Short-Term Depression for Silicon Neurons

摘要

We report a stochastic dynamical synapse for VLSI spiking neural systems. The compactness of the circuit, real-time stochastic behavior, and probability tuning make it well suitable to implement stochastic synapses with variety of dynamics. The stochastic synapse implements short-term depression (STD) using a subtractive single release model. Preliminary experimental results show a good match with theoretical predictions. The output from the stochastic synapse with STD has negative autocorrelation and lower power spectral density at low frequencies which can remove the information redundancy in the input spike train. The mean transmission probability is inversely proportional to the input spike rate which has been suggested as an automatic gain control mechanism in neural systems. The silicon stochastic synapse with plasticity could potentially be a powerful addition to existing deterministic VLSI spiking neural systems.