A Compact and Fast Single-Spike TTFS Encoder With Ultralow Operating Voltage for Efficient Neuromorphic Fourier Transform.

Fourier transform (FT) is an indispensable mathematical foundation for frequency-domain analysis in modern signal processing. However, conventional FT hardware relies on complex digital logic circuits, leading to high power consumption and poor adaptability. Spiking neural networks (SNNs) with time-to-first spike (TTFS) encoding offer advantages in energy efficiency, yet existing TTFS encoders suffer from high operating voltage, slow encoding speed, complex peripheral circuits, and redundant spike outputs. Here, we propose a compact, ultralow-voltage, and ultra-fast single-spike TTFS encoder based on a Pt/TaOx/V device. The device exhibits reliable threshold switching, for which the combined experimental and computational results support a proposed mechanism coupling oxygen vacancies-assisted conduction in TaOx with a thermally driven conductivity change in a VO2-like interfacial region. Moreover, it achieves single-spike TTFS encoding at an ultralow voltage of ∼0.35 V with sub-microsecond (∼100 ns) encoding speed in a simplified circuit configuration. Systematic characterizations on multi-tone signals, digitized speech segments, and linear frequency-modulated non-stationary signals demonstrate that the SNN-FFT system achieves spectral fidelity highly consistent with standard FFT, while maintaining strong sparsity and low power consumption. This work not only develops a high-performance TTFS spike encoder but also provides a feasible and efficient solution for low-power, real-time, miniaturized neuromorphic FT hardware.