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

TL;DR

This paper presents BS-KMQ, a novel nonlinear quantization method that reduces ADC resolution needs in in-memory computing, leading to significant improvements in accuracy, area, speed, and energy efficiency.

Contribution

Introduction of Boundary Suppressed K-Means Quantization (BS-KMQ), a new NL quantization approach that suppresses outliers for better ADC efficiency and system performance.

Findings

Achieves 7x area reduction in NL-ADC design.

Reduces quantization error by at least 3x over existing methods.

Provides up to 24x energy efficiency improvement in system simulations.

Abstract

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.