Join us for our upcoming SAFARI Live Seminar
Title: Fast Reliable Digital Processing-in-Memory
As the memory-wall is becoming a fundamental bottleneck to computing systems, an emerging solution is processing-in-memory (PIM): inserting basic logic functionality into the underlying physical devices of the memory. Digital PIM enables massive parallelism for aligned bitwise operations within the memory, thereby providing a highly unique computational model with massive potential. In this talk, we begin by presenting recent works that seek to extend this bitwise parallelism into high-throughput vectored arithmetic. We present the first in-memory floating-point algorithms that are compatible with nearly all digital PIM technologies as they are based solely on data-flow, and also provide significant latency acceleration through an emerging concept of partitions. We continue with recent works that have exploited this vectored arithmetic towards larger applications such as matrix operations (multiplication and convolution) and high-throughput Fast-Fourier-Transform (FFT). Lastly, we address reliability, a significant barrier to the emergence of digital PIM, through a novel concept of in-memory error-correcting-codes (ECC) that exploits the digital PIM itself towards ECC encoding/decoding.
Orian Leitersdorf is currently studying towards the B.Sc, M.Sc., and Ph.D. degrees at the Technion, Haifa, Israel. He is a scholar at both the Technion Excellence Program and the Technion Lapidim CS Excellence Program, and is also a recipient of the Gutwirth Excellence Scholarship. His current research aims to advance digital processing-in-memory (PIM) towards fundamental applications (e.g., matrix operations, graph algorithms, cryptography) while also addressing challenges such as reliability. Further, his research interests also include several topics in theoretical computer science, including coding theory.