LUPIS : Latch-Up Based Ultra Efficient Processing-in-Memory System

Joonseop Sim1, Mohsen Imani1, Woojin Choi1, Yeseong Kim2, Tajana Rosing1
1UCSD, 2University of California San Diego


Abstract

Internet of Things (IoT) involve exchange and processing of massive data. This poses a huge challenge in the current computing systems due to the limited memory bandwidth. Processing in-memory (PIM) is a promising candidate to minimize this bottleneck and reduce the performance gap between processor and memory latency. We propose LUPIS (Latch-Up based Processing In-memory System) for nonvolatile memory (NVM). Unlike existing PIM techniques, which mainly focus on bitwise operation based computations and involve considerable latency and area penalty, our design facilitates computations like addition and multiplication with very low latency. This makes the system faster and more efficient as compared to the state-of-the-art technologies. We evaluate LUPIS at both circuit-level and application-level. Our evaluations show that LUPIS can enhance the performance and energy efficiency by 62X and 484X respectively as compared to a recent GPGPU architecture. Compared to the state-of-the-art PIM accelerator, our design presents 12.7X and 20.9X improvement in latency and energy consumption with insignificant overhead of 21% for area increase and one cycle for latency delay.