Mega: A 22 nm Convolutional Spiking Neural Network Accelerator Achieving 0.375 pJ/SOP for Efficient Edge Vision

2026-06-29

AI Summary

Existing convolutional spiking neural network (SNN) accelerators struggle to efficiently exploit convolutional parallelism and lack flexibility in accommodating varying memory demands and input sparsity across layers. To address these challenges, this work proposes a digital SNN accelerator tailored for edge vision applications, featuring a highly parallel 3×3 convolution architecture, a unified memory system that integrates spikes, neuron states, and weights, and a low-overhead event-driven spike processing mechanism. Implemented in GlobalFoundries 22 nm FDSOI technology, the design achieves an energy efficiency of 0.375 pJ/SOP—four times better than the state-of-the-art at the time—demonstrating significant advances in both energy efficiency and architectural flexibility.

Abstract

Convolutional Spiking Neural Networks (SNN) offer the potential for highly energy-efficient vision processing by exploiting sparse, event-driven computation. However, existing SNN accelerators underutilize the inherent parallelism of convolutional layers and lack the flexibility to accommodate varying memory demands and input sparsity across layers. This paper presents Mega, a digital architecture for convolutional SNNs that addresses these limitations through three key contributions: (1) highly parallel acceleration of $3 \times 3$ convolutions, (2) a unified data memory for spikes, neuron states, and weights, and (3) efficient spike map processing with low-overhead spike detection. Fabricated in GlobalFoundries 22 nm FDSOI technology, Mega achieves an energy efficiency of 0.375 pJ/SOP, improving the state of the art by $4\times$.

Problem

• Convolutional Spiking Neural Networks
• SNN accelerators
• parallelism
• memory demands
• input sparsity

Innovation

• Convolutional Spiking Neural Network
• Energy Efficiency
• Parallel Acceleration
• Unified Memory Architecture
• Spike Map Processing