Securing Emerging Nonvolatile Main Memory With Fast and Energy-Efficient AES In-Memory Implementation

Abstract

As CMOS technology approaches its scaling limit, emerging nonvolatile memory (NVM) technologies become promising alternatives to DRAM due to their low leakage power and better scalability. However, the nonvolatile main memory system suffers from a new security vulnerability. An attacker can readily access sensitive information on the memory, since the nonvolatility allows information to be retained for a long time even after the power is OFF. While real-time memory encryption during memory accesses with dedicated Advanced Encryption Standard (AES) engine is an effective solution for this vulnerability, it incurs runtime performance and energy overhead. Alternatively, in this paper, we propose a fast and efficient AES in-memory (AIM) implementation, to encrypt the whole/part of the memory only when it is necessary. Rather than adding extra processing elements to the cost-sensitive memory, we take advantage of NVM’s intrinsic logic operation capability to implement the AES algorithm. We leverage the benefits (large internal bandwidth and dramatic data movement reduction) offered by the in-memory computing architecture to address the challenges of the bandwidth intensive encryption application. Embracing the massive parallelism inside the memory, AIM outperforms existing mechanisms with higher throughput yet lower energy consumption. The experimental results show that compared with state-of-the-art AES engine running at 2.1 GHz, AIM speeds up the encryption process by $80\times $ for a 1-GB NVM.