Resistive In-Memory Computing (RIMC) offers ultra-efficient computation for edge AI but faces accuracy degradation due to RRAM conductance drift over time. Traditional retraining methods are limited by RRAM's high energy consumption, write latency, and endurance constraints. We propose a DoRA-based calibration framework that restores accuracy by compensating influential weights with minimal calibration parameters stored in SRAM, leaving RRAM weights untouched. This eliminates in-field RRAM writes, ensuring energy-efficient, fast, and reliable calibration. Experiments on RIMC-based ResNet50 (ImageNet-1K) demonstrate 69.53% accuracy restoration using just 10 calibration samples while updating only 2.34% of parameters.
View on arXiv@article{dong2025_2504.03763,
title={ Efficient Calibration for RRAM-based In-Memory Computing using DoRA },
author={ Weirong Dong and Kai Zhou and Zhen Kong and Quan Cheng and Junkai Huang and Zhengke Yang and Masanori Hashimoto and Longyang Lin },
journal={arXiv preprint arXiv:2504.03763},
year={ 2025 }
}