SRAM-Based Compute-in-Memory Accelerator for Linear-decay Spiking Neural Networks

Researchers developed an SRAM-based compute-in-memory accelerator for spiking neural networks that uses linear decay approximation instead of exponential decay, achieving 1.1x to 16.7x reduction in energy consumption. The innovation addresses the bottleneck of neuron state updates in neuromorphic computing by performing in-place decay directly within memory arrays.

• →Linear decay approximation replaces costly exponential decay in spiking neural networks with only 1% accuracy loss
• →New architecture performs neuron state updates in parallel within SRAM arrays, eliminating sequential processing bottlenecks
• →Energy consumption reduced by 1.1x to 16.7x while providing 15.9x to 69x better energy efficiency
• →Solution addresses the key latency and energy bottleneck in spiking neural network inference beyond matrix multiplication
• →Research demonstrates importance of optimizing state-update dynamics for scalable neuromorphic processing