Compact and Energy-Efficient Memristive Spiking Neuromorphic Accelerator for Bio-inspired Interception Tasks

2026-05-29

AI Summary

This work addresses the memory wall and energy efficiency bottlenecks faced by spiking neural networks (SNNs) under the von Neumann architecture when deployed in biologically inspired interception tasks. To overcome these challenges, the authors propose a highly energy-efficient memristor-based in-memory computing accelerator. The design incorporates a novel 1T1R crossbar array and an integrated leaky integrate-and-fire neuron circuit with separated input and membrane potential nodes, effectively mitigating membrane potential drift and enhancing integration fidelity and system reliability. Fabricated in SkyWater SKY130 CMOS technology, the prototype chip achieves an ultra-low energy consumption of 10.67 pJ per spike and a compact cell area of 906 μm². Experimental results demonstrate a high correlation coefficient of 0.9622 with software SNN baselines and an interception success rate of 96%, striking an exceptional balance among energy efficiency, computational accuracy, and integration density.

Abstract

Spiking neural networks (SNNs) provide an efficient event-driven computing paradigm for bio-inspired interception tasks. However, most implementations rely on von Neumann digital computing platforms, where memory and computation bottlenecks limit energy efficiency. This work presents a compact and energy-efficient memristive neuromorphic accelerator for bio-inspired interception tasks. A novel one-transistor-one-resistor (1T1R) crossbar array is designed to emulate synaptic operations in the in-memory computing (IMC) domain, while circuit-level optimization mitigates membrane drift and improves integration fidelity. In addition, an integrate-and-fire (IF) neuron with separated input and membrane nodes is developed to improve inference robustness during array-interfaced operation. Implemented in the SkyWater SKY130 PDK, the proposed neuron achieves an energy consumption of 10.67 pJ/spike and an area of 906 um^2. System-level results show that the memristive IMC output closely matches the software SNN baseline, with a correlation coefficient of 0.9622, while achieving a 96% interception success rate. These results demonstrate the effectiveness of the proposed design for compact and reliable memristive SNN inference in bio-inspired interception tasks.

Problem

• Spiking Neural Networks
• Memristive Neuromorphic Accelerator
• In-Memory Computing
• Bio-inspired Interception
• Energy Efficiency

Innovation

• memristive neuromorphic accelerator
• in-memory computing
• spiking neural networks
• 1T1R crossbar array
• integrate-and-fire neuron