Adaptive Quantization as a Device-Algorithm Co-Design Approach to Improve the Performance of In-Memory Unsupervised Learning With SNNs
Adaptive Quantization as a Device-Algorithm Co-Design Approach to Improve the Performance of In-Memory Unsupervised Learning With SNNs
复制标题
DOI：
10.1109/ted.2019.2898402
复制
发表时间：
2019-02
影响因子：
3.1
通讯作者：
Yuhan Shi;Zhisheng Huang;Sangheon Oh;Nathan Kaslan;Jungwoo Song;D. Kuzum
中科院分区：
文献类型：
--
作者：
Yuhan Shi;Zhisheng Huang;Sangheon Oh;Nathan Kaslan;Jungwoo Song;D. Kuzum
片外存储器访问是加速神经网络运算和降低能耗的主要瓶颈。已经提出了使用新兴的非易失性存储器（eNVM）的存储器内训练和计算来解决这个问题。然而，少量的电导状态限制了内存在线学习性能。在这里，我们介绍了一种设备算法协同设计方法，并将其应用于相变存储器（PCM），以提高学习精度。我们提出了一种自适应量化方法，该方法补偿了由于有限的电导水平而导致的精度损失，并使低精度eNVM设备能够实现高精度无监督学习。我们为NeuroSim平台开发了一个尖峰神经网络框架，以比较模拟和数字实现的PCM阵列的在线学习性能，并对能耗、延迟和面积进行基准权衡。
Off-chip memory access is the primary bottleneck toward accelerating neural network operations and reducing energy consumption. In-memory training and computation using emerging nonvolatile memories (eNVMs) have been proposed to address this problem. However, a small number of conductance states limit in-memory online learning performance. Here, we introduce a device-algorithm co-design approach and its application to phase change memory (PCM) for improving learning accuracy. We present an adaptive quantization method, which compensates the accuracy loss due to limited conductance levels and enables high-accuracy unsupervised learning with low-precision eNVM devices. We develop a spiking neural network framework for NeuroSim platform to compare online learning performance of PCM arrays for analog and digital implementations and benchmark the tradeoffs in energy consumption, latency, and area.