2026-09-18T21:53:39Zhttps://repository.rice.edu/server/oai/request

oai:repository.rice.edu:1911/1181642025-09-16T20:45:55Zcom_1911_8299col_1911_13110

Yang, Kaiyuan Pakala, Akhil 2025-01-16T19:31:45Z 2024-12 2024-08-12 December 2024 2025-01-16T19:31:45Z https://hdl.handle.net/1911/118164 The adoption of Artificial Intelligence (AI) and IoT devices has seen unprecedented growth in recent times. AI engines demand more processing capabilities while simultaneously handling sensitive user information and proprietary IP. Concurrently, IoT devices generate vast amounts of sensitive data, necessitating robust security measures to safeguard against breaches and ensure data privacy. Hence, there is a definitive need for faster and more secure systems. We present two key implementations that address the problem. Firstly, we introduce MBSNTT (Multi-Bit Serial Number Theoretic Transform Accelerator) to accelerate the NTT operation in Homomorphic Encryption. In this implementation, we apply processing-in-memory techniques to the NTT operation, thereby achieving high parallelism. In the second design, namely HDCIM (Hybrid Security-based Digital Compute in Memory Accelerator for Protected Inference), we propose hybrid security by applying mathematical masking techniques to neural network operations and closing the security gaps with low-overhead physical security measures. application/pdf eng Copyright is held by the author, unless otherwise indicated. Permission to reuse, publish, or reproduce the work beyond the bounds of fair use or other exemptions to copyright law must be obtained from the copyright holder. In-Memory Computing Cryptography Homomorphic Encryption NTT SRAM Masking Neural Networks Machine Learning Physical security Secure Circuits: Efficient hardware countermeasures against physical side-channel attacks Thesis Text 2025-06-01 2025-06-01 Electrical and Computer Engineering Engineering Rice University Masters Master of Science open.access