Northeastern University researchers introduced ROBBIN, a device-aware Rowhammer attack that injects neural-network backdoors during inference.

Northeastern University researchers published a paper titled “ROBBIN: Rowhammer-Based Backdoor Injection during Inference.” The work presents a hardware-aware attack that incorporates device-specific memory vulnerabilities into neural-network backdoor construction.

Rowhammer induces bit flips in dynamic random-access memory by repeatedly accessing selected memory rows. ROBBIN first characterizes the bit-flip patterns of the target DRAM device rather than assuming that vulnerabilities are identical across memory hardware.

The attack then uses the characterized device information to iteratively select DRAM targets for backdoor injection during model inference. The supplied abstract identifies ROBBIN as an attack method; it does not report a product deployment or operating-system vulnerability patch.