An Analog In-Memory-Search Solution based on 3D-NAND Flash Memory for Brain-Inspired Computing

We present a high performance 3D-NAND-flash-based analog in-memory-searching (AIMS) chip with compact cell design, which enables novel searching/matching functionality in the analog domain. It can perform ultra-high parallel in-memory similarity searching with digital or analog output. The AIMS unit cell composition, operation algorithm, and array-level demonstration are presented in detail. Reliability concerns including data retention, read disturbance, and Program/Erase (P/E) cycling-induced issues are evaluated. Face recognition with VGGFace2 dataset is demonstrated with excellent inference accuracy up to 95%, thanks to the analog range storage in AIMS cell with gradient range boundary for approximate matching capabilities. The AIMS chip can mimic the operations in the human brain, and is highly potential to be a key component in the memory-centric computing systems.