Time-Based Analog Readout for Efficient Memristive SNN Inference

A new readout architecture for spiking neural networks (SNNs) replaces power-intensive current-mode summing with voltage-to-time conversion for vector-matrix multiplications. Post-layout simulations in 130 nm CMOS demonstrate improved area and energy efficiency for in-memory, analog SNN inference.