2 synaptic processing unit (S²-SPE), which performs computations only on synapses associated with valid spike inputs in active (non-inhibited) neurons, thereby efficiently leveraging both neuron and spike sparsity. Eventually, a neuromorphic attention core is developed based on the S²-SPE, and its effectiveness is validated on various datasets. The experimental results demonstrate that the core can bypass over 98% of invalid calculations in the neuromorphic attention mechanism, achieving an outstanding energy efficiency of sub-0.06 pJ/SOP, which represents over 50% improvement compared to the baseline and outperforms related state-of-the-art (SOTA) works. This work is expected to advance neuromorphic hardware toward greater energy efficiency and facilitate its deployment in edge applications.">

An Energy-Efficient Neuromorphic Self-Attention Core Exploiting Dual Sparsity in Neurons and Spikes (original) (raw)

IEEE Account

Purchase Details

Profile Information

Need Help?

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity.
© Copyright 2026 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.