We propose Multiplier-less INTeger (MINT) quantization, an efficient uniform quantization scheme for the weights and membrane potentials in spiking neural networks (SNNs). Unlike prior SNN quantization works, MINT quantizes the memory-hungry membrane potentials to extremely low precision (2-bit) to significantly reduce the total memory footprint. Additionally, MINT quantization shares the quantization scaling factor between the weights and membrane potentials, eliminating the need for multipliers that are necessary for vanilla uniform quantization. Experimental results demonstrate that our proposed method achieves accuracy that matches the full-precision models and other state-of-the-art SNN quantization works while outperforming them on total memory footprint and hardware cost at deployment. For instance, 2-bit MINT VGG-16 achieves 90.6% accuracy on CIFAR-10 with approximately 93.8% reduction in total memory footprint from the full-precision model; meanwhile, it reduces 90% computation energy compared to the vanilla uniform quantization at deployment.

翻译：暂无翻译