Efficient In-Memory AES Encryption Implementation Using a General Memristive Logic: Surmounting the data movement bottleneck

Abstract

Emerging nonvolatile main memory (NVMM) suffers from secure vulnerability due to its nonvolatility. To address this issue, existing methods tend to employ the encryption engine on the CPU side for encryption. However, this incurs large energy and latency overhead due to the massive data movement between the CPU and NVMM. On the other hand, popular encryption algorithms like the Advanced Encryption Standard (AES) usually involve massive bit-level parallelism. As a result, an emerging technology named logic-in-memory (LiM), which leverages the electrical characteristics of nonvolatile devices to enable efficient in-memory Boolean operations in parallel, is a promising solution to eliminating data movement overhead and enables faster and more energy-efficient encryption.