Memory Modeling with Dynamic Time Evolution Method for Neuromorphic Circuit Simulations

Abstract

For simulations of emerging neuromorphic circuits an analog memory modeling strategy with the dynamic time evolution method (DTEM) is reported. Dynamic state variables are needed to trace the physical quantities of the memory state representations. In a SPICE simulator time varying nodal voltages of the transient domain are capable to emulate changings of these physical quantities thus leveraging sub-circuits (SC) with additional nodes is one feasible method. To accommodate large scale simulation of neuromorphic circuits, the dynamic time evolution method is proposed to trace the varying memory states in the spiking-time-dependent-plasticity (STDP). Circuit matrix size is reduced with the DTEM implementations thus efficient simulation speedups are achieved.