A Time-Domain Verification Framework for Digitally-Trained Op Amp and Ring-Oscillator Based Analog Spiking Neural Network

Spiking Neural Networks (SNNs) implemented using analog circuits offer substantial energy efficiency compared to digital realizations, but verifying their time-domain behavior remains challenging. Analog neurons, such as op-amp and ringoscillator (RO)-based leaky integrate-and-fire models, exhibit nonlinear dynamics that can diverge from their digitally trained counterparts, particularly at high spike rates due to discretization approximations like the bilinear transform. To address these challenges, we present a time-domain verification and visualization framework that allows users to configure network architectures, neuron models, including a new RO-based neuron architecture, and spike encoding schemes, and automatically generates Simulink-based simulations that reproduce analog dynamics. The framework provides real-time visualization of spiking activity, membrane potentials, and layer-wise dynamics, enabling both educational exploration and practical debugging. Demonstrations on the Iris dataset show that the framework achieves waveform-equivalent behavior to transistor-level simulations in Cadence Spectre while providing over an order-of-magnitude speedup. By combining interactive visualization with configurable neuron models, including the novel RO neuron, the framework offers an intuitive platform for understanding and verifying analog SNNs.