SpikySpace: A Spiking State Space Model for Energy-Efficient Time Series Forecasting
Kaiwen Tang, Jiaqi Zheng, Yuze Jin, Yupeng Qiu, Guangda Sun, Zhanglu Yan, Weng-Fai Wong
TL;DR
SpikySpace addresses the challenge of energy-efficient time-series forecasting on edge devices by introducing a fully spiking state-space model that replaces costly attention with a spike-driven, linear-time selective scan. Central to the approach are neuromorphic-friendly activations, PTSoftplus and PTSiLU, enabling spike-based gating and readout, and an ANN-to-SNN training pipeline with LSQ quantization. Empirical results on four multivariate datasets show competitive accuracy (e.g., up to $R^2=0.992$ on Electricity) while achieving substantial energy and parameter reductions (up to $98.7\%$ relative to iTransformer). The work demonstrates a practical path to on-device, real-time forecasting that preserves long-range temporal modeling via the state-space formalism while leveraging event-driven computation on neuromorphic hardware.
Abstract
Time-series forecasting often operates under tight power and latency budgets in fields like traffic management, industrial condition monitoring, and on-device sensing. These applications frequently require near real-time responses and low energy consumption on edge devices. Spiking neural networks (SNNs) offer event-driven computation and ultra-low power by exploiting temporal sparsity and multiplication-free computation. Yet existing SNN-based time-series forecasters often inherit complex transformer blocks, thereby losing much of the efficiency benefit. To solve the problem, we propose SpikySpace, a spiking state-space model (SSM) that reduces the quadratic cost in the attention block to linear time via selective scanning. Further, we replace dense SSM updates with sparse spike trains and execute selective scans only on spike events, thereby avoiding dense multiplications while preserving the SSM's structured memory. Because complex operations such as exponentials and divisions are costly on neuromorphic chips, we introduce simplified approximations of SiLU and Softplus to enable a neuromorphic-friendly model architecture. In matched settings, SpikySpace reduces estimated energy consumption by 98.73% and 96.24% compared to two state-of-the-art transformer based approaches, namely iTransformer and iSpikformer, respectively. In standard time series forecasting datasets, SpikySpace delivers competitive accuracy while substantially reducing energy cost and memory traffic. As the first full spiking state-space model, SpikySpace bridges neuromorphic efficiency with modern sequence modeling, marking a practical and scalable path toward efficient time series forecasting systems.