Design of finite impulse response filter based on single-layer CoPt spin–orbit torque devices

Abstract

Signal processing often suffers from high energy consumption and limited speed due to the separation of data storage and computation in conventional electronic systems. To harness the full potential of spin–orbit torque (SOT) devices in real-time temporal signal processing, this paper proposes a SOT-based signal processing scheme. By exploiting the multistate resistance characteristics of SOT devices and their intrinsic alignment with convolution algorithms, we develop a synergistic framework for finite impulse response (FIR) filtering circuits. A multi-channel SOT convolution kernel circuit simulation model is further constructed, which exhibits outstanding performance in denoising tasks—ranging from simple sinusoidal signals to complex speech signals. Based on this model, we implement a hardware FIR filter using CoPt-SOT devices, with experimental measurements closely matching simulation results (sinusoidal signal: SNR = 18.84 dB; speech signal: SNR = 10.29 dB). Frequency-domain analysis further confirms that the filter preserves low-frequency information while suppressing high-frequency noise. This work not only advances the application of SOT devices in neural network hardware and real-time signal processing systems but also underscores their promise for next-generation high-efficiency computing architectures.