Secondary Order RC Sensor Neuron Circuit for Direct Input Encoding in Spiking Neural Network

Abstract

In spiking neural networks (SNNs), artificial sensor neurons are crucial for converting real-world analog information into encoded spikes. However, existing SNNs face challenges due to the inefficient implementation of input sensor neurons. Here, this study proposes an SNN-compatible spike mode sensor, designed to directly convert analog current signals into real-time encoded spikes, feeding the SNN concurrently. The input sensor neuron is realized using a stable neuron circuit employing a threshold switching (TS) memristor and secondary order RC block. This design enables time delay-free spike firing, operates at low voltage, and offers a wide signal sensing range. Furthermore, this study presents an expression delineating the relationship between the pulse emission properties of the circuit and the parameters of its components, laying the basis for circuit components design and development. Analytical analysis confirms the sensor's efficacy in implementing rate-based and time-to-first spike encoding schemes. Integrating the sensor into SNNs as the input layer for image training and recognition tasks yields an impressive accuracy of 87.58% on the MNIST dataset, showcasing its applicability as a crucial interface between the physical world and the SNN framework.