Reliability Assessment for an In-3D-NAND Approximate Searching Solution

The reliability issues of an In-3D-NAND approximate searching system for Hamming distance computing are evaluated and solved by optimized high VT level, recovery heating process, and reliability tracking array. The chip with optimized high Vt level can pass the commercial 3D-NAND retention criteria even after 1k cycling and 10M read stresses. Recovery heating process can mitigate over-program issue in high data update situations (up to 30K cycles) and provide robust retention capability after cycling. Reliability tracking array enables sensing level calibration for Hamming distance computing across long-term operations. Experiments showed that the proposed solution provides reliable in-memory database processing function suitable for memory centric computing and computational SSD.