Single-Ferroelectric FET based Associative Memory for Data-Intensive Pattern Matching

Jiayi Wang, Songyu Sun, Xunzhao Yin
Zhejiang University


Content addressable memories (CAMs) embeds parallel associative search directly into the memory blocks, thus finding widespread utility in associative memory (AM) related applications. To accommodate increasing demands of data-intensive search tasks, various efforts have been devoted to enhancing CAM density. These endeavors include the use of non-volatile memory (NVM) devices with compact structures and capitalizing on the multi-level cell (MLC) characteristics of NVM devices. In this work, we present a novel single-FeFET based CAM design, complemented by a 2-step search scheme. This design achieves ultra-compact storage density and supports dual CAM operations: binary/ternary CAM search for Hamming distance computations and multi-bit CAM for exact associative searches. Both binary/ternary CAM and multi-bit CAM operations have been illustrated and validated, and the area per bit, search latency and energy metrics have been evaluated at array level. In genome sequencing applications using hyperdimensional computing paradigm, our single-FeFET based AM engine achieves 89.9x/71.9x speedup and 66.5x/30.7x energy efficiency improvement over GPU implementations.