Presentation
Victor: A Variation-resilient Approach Using Cell-clustered Charge-domain computing for High-density High-throughput MLC CiM
DescriptionMulti-level-cell (MLC) device-based Computing-in-memory (CiM) is a promising paradigm performing high-density energy-efficient matrix-vector multiplication (MVM) alleviating memory wall bottleneck. However, most MLC CiM faces challenges of low device on/off ratio and device-to-device variation, limiting computing reliability and throughput. This work proposes Victor, a variation-resilient approach using cell-clustered charge-domain computing for high-density and high-throughput MLC CiM. A cell-clustered-computing design methodology is proposed to improve MVM reliability and throughput. Taken MLC RRAM device, Victor reaches 1.93x-3.79x peak density compared with single-level-cell CiM under same area efficiency. With same macro area, Victor of 2-bit cell shows 1.55-4.59x throughput compared with current-domain crossbar-array CiM.
Event Type
Research Manuscript
TimeTuesday, July 11th3:55pm - 4:10pm PDT
Location3010, 3rd Floor
Design
In-memory and Near-memory Computing Circuits