A Defense Against Rowhammer Attacks Based on Dual-Layer RRAM Counting Method
The security issues in high-density memory have attracted increasing attention, and how to defend Row Hammer (RH) has become a research hotspot. To address the issue of significant storage overhead in defending RH attacks, this paper proposes a defense method based on dual-layer counting with Resistive Random Access Memory (RRAM). Initially, this method investigates the transition from low-resistance state to high-resistance state in RRAM, and utilizes the resistance variation process of RRAM to represent the number of row activations, achieving the functionality of counting row activations using a single RRAM; Secondly, a group counter is formed by an RRAM that one end connected to multiple memory rows and a sensitive amplifier that monitors the resistance variation of the RRAM. This group counter completes the aggregated counting of the row groups formed by multiple memory rows, completing the aggregation counting of the row groups composed of multiple memory rows; Finally, a row counter is constructed using RRAM and a sensitive amplifier placed at the end of a single memory row to accurately count the memory rows that exceed the counting range of the previous stage, achieving protection against RH. The dual-layer counter composed of row counter and group counter is connected to DRAM and tested under TSMC28nm technology. The experimental results demonstrate that the two counters can work together to protect against RH attacks without affecting the original Dynamic Random Access Memory (DRAM) timing logic.
Authors
- Zhuo Ruan
- Pengjun Wang (ORCID: https://orcid.org/0000-0002-1461-3719)
- Yuejun Zhang (ORCID: https://orcid.org/0000-0003-1132-6332)
- Shuaicheng Guo
Institutions
- Ningbo University (CN)
- Wenzhou University (CN)
Publication Details
- Published
- 2024-10-25
- DOI
- https://doi.org/10.1109/icicm63644.2024.10814450
- Primary Topic
- Adversarial Robustness in Machine Learning
- Type
- article
- Field-Weighted Citation Impact
- 0.00