Edge-AI Accelerated Neuromorphic VLSI Architectures for Real-Time Sensor Fusion in Autonomous Systems

Abstract

The rapid evolution of autonomous systems, such as self-driving vehicles, unmanned aerial vehicles (UAVs), and industrial robots, has led to an exponential increase in the volume, velocity, and variety of sensor data that must be processed in real-time. Traditional von Neumann architectures and centralized cloud-based processing introduce latency, energy inefficiencies, and scalability bottlenecks that are unsuitable for latency-sensitive and mission-critical applications. To address these challenges, this paper proposes a novel edge-AI accelerated neuromorphic VLSI architecture designed for real-time sensor fusion in autonomous systems. Inspired by the efficiency of biological neural networks, the proposed architecture integrates neuromorphic computing principles with edge-based artificial intelligence to process heterogeneous sensor data at the hardware level, ensuring ultra-low latency, energy efficiency, and high reliability.

The architecture employs spiking neural networks (SNNs) implemented on custom VLSI circuits to facilitate asynchronous, event-driven processing. Sensor inputs ranging from LiDAR, radar, inertial, to visual modalities are fused using adaptive learning mechanisms embedded within the chip, allowing context-aware decision-making without reliance on external computation. The use of edge AI algorithms, such as lightweight convolutional and recurrent neural models, complements the neuromorphic substrate for dynamic pattern recognition and anomaly detection. Simulated and real-world case studies demonstrate significant improvements in energy efficiency (up to 60%), reduced decision latency (up to 40%), and robust performance in noisy environments.

This research bridges a critical gap between neuromorphic theory and embedded system implementation, paving the way for future intelligent autonomous platforms capable of making real-time decisions locally, securely, and efficiently.