Feb, 2022
HiMA:一种快速可扩展的基于历史的内存访问引擎,用于可微分神经计算机
HiMA: A Fast and Scalable History-based Memory Access Engine for
Differentiable Neural Computer
TL;DRHiMA is a tiled, history-based memory access engine with distributed memories designed to improve computation speed for memory-augmented neural networks supporting almost all memory operations, including DNC-D, demonstrating a significant improvement in speed, area efficiency, and energy efficiency compared to other MANN accelerators and Nvidia 3080Ti GPU.