Processing in Close Proximity to Memory

In this talk, we will cover two different examples of processing in close proximity to memory. 1) Larger cache sizes closer to processor cores increase processing efficiency, but physical limitations restrict cache sizes at a given latency. Effective cache capacity can be expanded via the inline compression (processing in cache) of data as it enters a lower-level cache. Using a state-of-the-art processor cache hierarchy as a comparative baseline, we will present a custom compression scheme designed for small, line-sized data blocks, examine optimal compressor/decompressor placement, solutions to common compression drawbacks. 2) Fully homomorphic encryption (FHE) is a powerful cryptographic technique that enables computation on encrypted data without needing to decrypt it. FHE applications are both compute- and memory-intensive, owing to expensive operations on large data. Hierarchical near-DRAM processing (NDP) solution for FHE applications, harnesses the massive DRAM bank bandwidth. We observe various data access patterns in FHE that reveal distinct levels of parallelism: element-wise, limb-wise, coefficient-wise, and ciphertext-wise. We will discuss an NDP solution that exploits these levels of parallelism to map FHE operations and data onto different hierarchies of the NDP, while addressing major challenges with NDP for FHE.