TENDRA: Targeted Endurance Attack on STT-RAM LLC

Abstract

Spin transfer torque RAM (STT-RAM)-based last level cache (LLC) offers significant benefits like high density and low refresh energy, but faces challenges like high write latency and limited endurance. Malicious attacks in a multicore setup need access to only a single core to perform repeated attacks on specific memory locations that can lead to an accelerated lifetime degradation of the STT-RAM LLC cells. To highlight this vulnerability of STT-RAM LLC, we propose two variations of TENDRA (targeted endurance attack), namely, recurring location attack (RLA) and recurring toggle attack (RTA). Our work highlights the efficiency of these attacks on modern counter-based wear leveling techniques and also the effect of wear leveling on these attacks.