Optimizing Memory Efficiency for Deep Convolutional Neural Network Accelerators

摘要

Convolutional Neural Network (CNN) accelerators have achieved nominal performance and energy efficiency speedup compared to traditional general purpose CPU- and GPU-based solutions. Although optimizations on computation have been intensively studied, the energy efficiency of such acceleratorsremains limited by off-chip memory accesses since their energy cost is magnitudes higher than other operations. Minimizing off-chip memory access volume, therefore, is the key to further improving energy efficiency. The prior state-of-the-art uses rigid data reuse patterns and is sub-optimalfor some, or even all, of the individual convolutional layers. To overcome the problem, this paper proposed an adaptive layer partitioning and scheduling scheme, called SmartShuttle, to minimize off-chip memory accesses for CNN accelerators. Smartshuttle can adaptively switch among differentdata reuse schemes and the corresponding tiling factor settings to dynamically match different convolutional layers and fully-connected layers. Moreover, SmartShuttle thoroughly investigates the impact of data reusability and sparsity on the memory access volume. The experimental results showthat SmartShuttle processes the convolutional layers at 434.8 multiply and accumulations (MACs)/DRAM access for VGG16 (batch size = 3), and 526.3 MACs/DRAM access for AlexNet (batch size = 4), which outperforms the state-of-the-art approach (Eyeriss) by 52.2% and 52.6%, respectively.