arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.10540cs.AR
收藏

基于内存的ADC非线性激活量化用于高效内存计算

In-Memory ADC-Based Nonlinear Activation Quantization for Efficient In-Memory Computing

City University of Hong Kong(香港城市大学);Case Western Reserve University(凯斯西储大学)

Shuai Dong, Junyi Yang, Biyan Zhou, Hongyang Shang, Gourav Datta, Arindam Basu

AI总结:

本文提出一种基于内存的非线性激活量化方法,通过抑制边界异常值提升量化精度,降低ADC分辨率需求,实现高效内存计算。

摘要:

在深度网络中,ReLU和硬件驱动的钳制操作常导致激活值在分布边缘积累,导致偏置聚类和现有非线性(NL)量化方法的次优化。本文介绍了一种新的非线性量化方法,即边界抑制K均值量化(BS-KMQ),旨在减少内存计算(IMC)系统中模拟到数字转换器(ADC)的分辨率要求。通过在聚类前抑制边界异常值,BS-KMQ实现了更平衡且信息丰富的非线性量化级别。所得到的非线性参考通过可重构的内存非线性ADC实现,相比先前的非线性ADC设计实现了7倍的面积改进。在ResNet-18、VGG-16、Inception-V3和DistilBERT上的评估表明,BS-KMQ的量化误差至少比线性、Lloyd-Max、累积分布函数(CDF)和K均值方法低3倍。它还相比线性量化方法,分别提高了后训练量化精度高达66.8%、25.4%、66.6%和67.7%。经过低比特微调后,BS-KMQ在显著更少的非线性ADC级别(3/3/4/4b)内保持了具有竞争力的准确性。在ResNet-18(6/2/3b)上的系统级仿真显示,相比现有IMC加速器,达到了4倍的速度提升和24倍的能量效率提升。

英文摘要:

In deep networks, operations such as ReLU and hardware-driven clamping often cause activations to accumulate near the edges of the distribution, leading to biased clustering and suboptimal quantization in existing nonlinear (NL) quantization methods. This paper introduces Boundary Suppressed K-Means Quantization (BS-KMQ), a novel NL quantization approach designed to reduce the resolution requirements of analog-to-digital converters (ADCs) in in-memory computing (IMC) systems. By suppressing boundary outliers before clustering, BS-KMQ achieves more balanced and informative NL quantization levels. The resulting NL references are implemented using a reconfigurable in-memory NL-ADC, achieving a 7x area improvement over prior NL-ADC designs. When evaluated on ResNet-18, VGG-16, Inception-V3, and DistilBERT, BS-KMQ achieves at least 3x lower quantization error compared to linear, Lloyd-Max, cumulative distribution function (CDF), and K-means methods. It also improves post-training quantization accuracy by up to 66.8%, 25.4%, 66.6%, and 67.7%, respectively, compared to linear quantization. After low-bit fine-tuning, BS-KMQ maintains competitive accuracy with significantly fewer NL-ADC levels (3/3/4/4b). System-level simulations on ResNet-18 (6/2/3b) demonstrate up to a 4x speedup and 24x energy efficiency improvement over existing IMC accelerators.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑