Architecture Considerations for Stochastic Computing Accelerators
Architecture Considerations for Stochastic Computing Accelerators
复制标题
随机计算加速器的架构注意事项
DOI：
10.1109/tcad.2018.2858338
复制
发表时间：
2018
影响因子：
2.9
通讯作者：
M. Oskin
中科院分区：
文献类型：
--
作者：
Vincent T. Lee;Armin Alaghi;Rajesh Pamula;V. Sathe;L. Ceze;M. Oskin
随机计算（SC）是嵌入式系统的一种替代计算技术，与二进制编码（BE）计算相比，它具有更低的面积和功耗，以及更好的容错能力。然而，加速器架构中SC的潜力和通用设计方法还没有得到很好的理解。在本文中，我们评估了单个SC操作和端到端加速器架构，以了解SC加速器何时以及为什么可以实现引人注目的能源效率收益。基于这些结果，我们提出了在构建能量优化的SC加速器架构时应考虑的一般设计准则。我们还评估了一个完全制造的ASIC原型-这是同类中的第一个-以经验评估SC中电压过标度（VOS）的误差容限。我们的结果表明，SC的能源效率增益主要源于SC更简单的数据路径，与等效BE相比，它需要更少的顺序元素。这使得它们可以实现高达<inline-formula> < text -math notation="LaTeX">$2.4 \times $ </ text -math></inline-formula> </inline-formula> < text -math notation="LaTeX">$30 \times $ </ text -math></inline-formula>的能源效率收益，分别为8位和4位定点精度。我们还发现，通过利用SC的容错编码，VOS可以进一步提高能源效率，最高可达<inline-formula> < text -math notation="LaTeX">$1.9\times $ </ text -math></inline-formula>。
Stochastic computing (SC) is an alternative computing technique for embedded systems which offers lower area and power, and better error resilience compared to binary-encoded (BE) computation. However, the potential of and general design methodologies for SC in accelerator architectures are not well-understood. In this paper, we evaluate individual SC operations, and end-to-end accelerator architectures to understand when and why SC accelerators can achieve compelling energy efficiency gains. Based on these results, we present general design guidelines that should be considered when building energy-optimal SC accelerator architectures. We also evaluate a fully fabricated ASIC prototype—the first of its kind—to empirically evaluate the error tolerance limits of voltage overscaling (VOS) in SC. Our results show that energy efficiency gains from SC primarily stem from SC’s simpler datapaths which require fewer sequential elements compared to BE equivalents. This allows them to achieve energy efficiency gains as high as <inline-formula> <tex-math notation="LaTeX">$2.4 \times $ </tex-math></inline-formula> and <inline-formula> <tex-math notation="LaTeX">$30 \times $ </tex-math></inline-formula> at 8-bit and 4-bit fixed-point precision, respectively. We also find that VOS can improve the energy efficiency further by up to <inline-formula> <tex-math notation="LaTeX">$1.9\times $ </tex-math></inline-formula> by exploiting SC’s error tolerant encoding.