ReRAM-aware Model Finetuning addressing I-V Non-linearity and Retention Errors

2026-06-15

AI Summary

This work addresses the degradation in neural network accuracy caused by I–V nonlinearity and data retention errors in ReRAM-based in-memory computing. To avoid the prohibitive cost of full retraining, the authors propose an efficient fine-tuning method that jointly models these two non-idealities for the first time. Specifically, they correct I–V nonlinearity using a range-compressed sinh transformation and incorporate retention-induced errors into a regularized loss function within a standard fine-tuning pipeline. The approach achieves zero accuracy loss on ResNet18 and DeiT-Tiny, incurs less than 2% accuracy drop on MobileNetV3 over ImageNet, and reduces the F1 score by only 1 point on the SQuAD v2 benchmark, substantially lowering the hardware adaptation overhead.

Abstract

Traditional CPU, GPU, and NPU architectures are increasingly limited by the von Neumann bottleneck. While In-Memory Computing (IMC) using ReRAM crossbar arrays offers a high-density, energy-efficient alternative, its practical deployment is constrained through their non-idealities. Existing hardware-aware training frameworks often require training from scratch, which is computationally prohibitive for modern large-scale models. In this work, we propose a finetuning-based hardware-aware training algorithm that enables robust DNN deployment on ReRAM with minimal training overhead. Our approach mitigates I-V non-linearity by applying a range-shrunk sinh transformation and incorporates retention errors directly into a regularization loss during the finetuning process. We evaluate our framework across models and tasks such as image classification and question-answering (QA). Experimental results demonstrate that our method achieves similar accuracy on large-scale models like ResNet18 and DeiT-Tiny as the base model. In-case of ImageNet for MobileNetV3 families the technique has only less than 2% accuracy degradation. Further, applying the technique on the SQuAD v2 dataset results in only 1 point degradation of F-1 score.

Problem

• ReRAM
• I-V non-linearity
• retention errors
• hardware-aware training
• von Neumann bottleneck

Innovation

• ReRAM
• in-memory computing
• hardware-aware finetuning
• I-V non-linearity
• retention error