Schmitt Trigger Optimization for Neuromorphic Circuits

Abstract

In neuromorphic computing, the performance of Spiking Neural Networks (SNNs) relies heavily on the precise firing threshold and reset voltages of its neurons. In Leaky Integrate-and-Fire (LIF) models for instance, these critical voltages are inherently governed by a hysteresis comparator. In this scenario, this paper proposes a hybrid automated sizing methodology for a six-transistor CMOS Schmitt Trigger tailored for neuromorphic Leaky Integrate-and-Fire (LIF) neurons. While traditional analytical models fail to account for short-channel effects in deep submicron technologies and purely random initial solution optimizations often suffer from suboptimal convergence, our approach leverages an Equation-Based Initial Solution (EBIS) to seed stochastic algorithms. Using the SkyWater 130 nm PDK, we evaluated Hill Climbing (HC) and Simulated Annealing (SA) across Symmetric, Short, and Asymmetric hysteresis profiles under five PVT corners. Results indicate that while SA-EBIS provides the most reliable average performance, HC demonstrates significantly lower computational overhead, albeit being highly dependent on the EBIS seed to avoid stagnation in sub-optimal regions. In contrast, the SA algorithm’s ability to escape local minima allows it to consistently identify superior solutions regardless of the initial guess, justifying its higher computational cost. The presented approach successfully corrects analytical threshold deviations in extreme design spaces without compromising area or power efficiency, ensuring reliable neuromorphic hardware integration.