Dynamic Symmetric Point Tracking: Tackling Non-ideal Reference in Analog In-memory Training

2026-02-24

AI Summary

In analog in-memory computing (AIMC), device update asymmetry induces systematic weight drift toward a non-ideal symmetry point, degrading training accuracy. This work proposes a dynamic symmetry-point tracking method that estimates and corrects the symmetry point in real time during training, eliminating the need for costly pre-calibration. The study provides the first theoretical characterization of the pulse complexity and estimation error associated with symmetry-point calibration, and designs a dynamically adaptive tracking mechanism with convergence guarantees. To enhance robustness, digital signal processing–inspired chopping and filtering techniques are integrated into the framework. Experimental results demonstrate that the proposed approach significantly improves training accuracy while substantially reducing calibration overhead.

Technology Category

• Machine Learning: Hardware-aware ML
• Search and Optimization: Sampling/Simulation-based Search
• Cognitive Modeling & Cognitive Systems: Analogy

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Analog in-memory computing (AIMC) performs computation directly within resistive crossbar arrays, offering an energy-efficient platform to scale large vision and language models. However, non-ideal analog device properties make the training on AIMC devices challenging. In particular, its update asymmetry can induce a systematic drift of weight updates towards a device-specific symmetric point (SP), which typically does not align with the optimum of the training objective. To mitigate this bias, most existing works assume the SP is known and pre-calibrate it to zero before training by setting the reference point as the SP. Nevertheless, calibrating AIMC devices requires costly pulse updates, and residual calibration error can directly degrade training accuracy. In this work, we present the first theoretical characterization of the pulse complexity of SP calibration and the resulting estimation error. We further propose a dynamic SP estimation method that tracks the SP during model training, and establishes its convergence guarantees. In addition, we develop an enhanced variant based on chopping and filtering techniques from digital signal processing. Numerical experiments demonstrate both the efficiency and effectiveness of the proposed method.

Problem

• Analog in-memory computing
• symmetric point
• update asymmetry
• calibration error
• weight drift

Innovation

• Dynamic Symmetric Point Tracking
• Analog In-memory Computing
• Update Asymmetry
• Chopping and Filtering
• Convergence Guarantee