Abstract: Large-scale search plays a crucial role in many machine learning applications, where fast and energy-efficient search is highly desirable. However, accelerating search at scale remains challenging due to massive data volumes and irregular memory access patterns. In-memory computing architectures—particularly non-volatile memory-based content addressable memory (CAM)—offer a promising solution by enabling massively parallel search with reduced energy consumption. This talk examines how architectural and system-level design choices impact the efficiency, scalability, and performance of CAM-based search accelerators. I will highlight bottlenecks and key tradeoffs, and present algorithm-hardware co-design techniques that improve performance and energy efficiency for large-scale search and retrieval workloads.

Host: Prof. L. Pozzi