Gait recognition (GR) has emerged as a promising biometric technology due to its non-contact nature and long-range identification capabilities, offering transformative potential in surveillance, healthcare, and intelligent systems. Despite significant progress in gait biometrics, current GR algorithms continue to struggle against dynamic gait variations, such as diverse viewpoints, appearance changes, and occlusion. To alleviate the aforementioned issues, this work introduces a spiking neural networks (SNNs)-based deep architecture, designed to optimize spatio-temporal gait feature modeling. The proposed architecture features two principal innovations: (1) a novel gait recognition framework that leverages the sparse activation and event-driven processing of SNNs to model temporal dynamics directly from gait sequences, marking the first application of SNNs in silhouette-based gait recognition; and (2) an SNNBlock that integrates 3D convolution and Leaky Integrate-and-Fire (LIF) neurons in a parallel pathway, capturing gait periodicity and spatial correlations effectively via dual branch structure. Experimental evaluations on multiple outdoor gait datasets demonstrate that the threshold dynamics inherent to LIF neurons enable superior temporal encoding of gait patterns, while the SNNBlock’s hierarchical spiking mechanism enhances regional perceptual weighting. This research not only establishes the novel framework for high-performing gait recognition but also advances the theoretical development of SNNs in computer vision, providing a scalable solution for real-world deployment.