Leveraging Rowhammer for Physically Unique and Non-tamperable Device Identification

Dynamic Random Access Memory (DRAM) remains the most widely used memory in computer architecture. This paper explores identifying a device using DRAM for creating Physically Unclonable Functions (PUFs) from Rowhammer-induced bit flips, e.g., for binding software to industrial control systems. Our method utilizes a device's inherent properties, similar to biometric identification, and is applicable to legacy hardware and embedded devices that lack a Trusted Platform Module. Rowhammer causes bit flips due to repeated read access. It results from electrical charges leaking into adjacent cells, leading to data corruption and security risks. Our research indicates that these bit flips are not random. We use statistical fingerprints and the Mahalanobis distance to identify systems based on the error distribution, which is dependent on other system components like the memory controller. A PUF generated by our approach can therefore not be transferred to another system just by swapping memory modules.