Abstract: Modern memory systems face scalability, security, and efficiency challenges as applications demand more. DRAM cells, the building blocks of memory systems, suffer from reliability issues that expose new attack vectors. One such attack is Row Hammer, which allows adversaries to flip the contents of DRAM cells by repeatedly accessing them. This compromises the security and scalability of memory systems. Additionally, limited memory and computing capacities act as bottlenecks for ML training.

This talk motivates the need for scalable, secure, and efficient memory systems. My presentation will dive deep into our latest ASPLOS 2022 and HPCA 2023 paper (best paper) and discuss a novel attack called Juggernaut. Juggernaut breaks the state-of-the-art Row Hammer mitigation. My talk will discuss developing scalable and efficient solutions to mitigate Row Hammer. Going beyond security, I will briefly discuss our system-level proposal for enabling efficient federated training (NeuRIPS 2023) for large-scale ML models. This work allows for memory, communication, and compute-aware Federated Learning. I will conclude the talk by describing how research vectors in my research group, namely Quantum Computing, AI/ML memory system architectures, and security, are critical for developing next-generation systems.

Biography: Prashant Nair is an Assistant Professor at the University of British Columbia (UBC). He is also an Affiliate Fellow at the Quantum Algorithms Institute. He leads the "Systems and Architectures (STAR) Lab." Dr. Nair has published over 25 papers in top-tier venues such as ISCA, MICRO, HPCA, ASPLOS, DSN, SC, NeuRIPS, and VLDB. He has received several awards, including the best paper award at HPCA 2023, two honorable mentions in IEEE MICRO Top-Picks, and the ECE Graduate Research Assistant Excellence Award at Georgia Tech.