BRAIN-INSPIRED SPIKING NEURAL NETWORKS FOR ULTRA-LOW POWER INTELLIGENT COMPUTING SYSTEMS

Abstract

The growing demand for intelligent computing at the edge has intensified the need for learning architectures that can deliver high accuracy with minimum energy consumption. Conventional artificial neural networks achieve strong computational performance but require dense numerical operations, continuous data transfer, and significant power resources, which limit their suitability for battery-operated and real-time embedded systems. This paper presents a brain-inspired spiking neural network framework for ultra-low-power intelligent computing systems by exploiting event-driven spike communication, temporal information encoding, and biologically inspired learning mechanisms. The proposed approach integrates leaky integrate-and-fire neuron dynamics, spike-timing-dependent synaptic adaptation, and lightweight surrogate-gradient optimization to improve classification performance while reducing redundant computation. Unlike traditional deep learning models, the network processes information only when meaningful spike events occur, enabling sparse activation and lower switching activity in neuromorphic hardware environments. The framework is evaluated on benchmark pattern-recognition and edge-intelligence tasks using accuracy, latency, spike rate, memory usage, and energy consumption as key performance indicators. Experimental results show that the proposed spiking neural network achieves a classification accuracy of 96.8%, which is comparable to conventional artificial neural networks while consuming substantially less energy. Compared with a standard convolutional neural network baseline, the proposed model reduces average energy consumption by 72.4%, decreases inference latency by 38.6%, and lowers memory utilization by 41.2%. The average spike activity is reduced by 64.7%, demonstrating the effectiveness of sparse event-driven computation. Furthermore, the system maintains stable performance under noisy input conditions, achieving an F1-score of 95.9% and a precision of 96.2%. These results confirm that brain-inspired spiking neural networks can provide an efficient balance between computational intelligence, energy efficiency, and real-time responsiveness. The study highlights the potential of spiking neural networks as a promising foundation for next-generation intelligent systems, particularly in edge AI, robotics, wearable electronics, smart sensors, and Internet-of-Things applications. The proposed framework contributes toward sustainable, adaptive, and hardware-friendly artificial intelligence by bridging biological neural principles with practical low-power computing architectures