Intel's 660p, 665p, and 670p all use a dynamic SLC cache strategy. The controller writes incoming data to QLC cells in SLC mode (1 bit per cell instead of 4), which is faster and places less stress on the cell. This creates a speed buffer that masks the inherent slowness of QLC writes.
The cache size is dynamic: on a nearly empty 1TB 660p, up to 140GB of NAND can operate in SLC mode. As the drive fills, the SLC cache shrinks. On a 75% full drive, the cache drops to around 12GB. Sustained writes that exceed the cache cause the controller to flush SLC data to QLC in the background. If the system loses power during this flush, both the SLC cache mapping and the QLC destination pages can end up in an inconsistent state. The flash translation layer, which maps logical block addresses to physical NAND locations, loses coherence.
The result: the drive either reports 0GB capacity, shows as an uninitialized disk, or disappears from BIOS. PC-3000 SSD accesses the controller through ROM mode, bypassing the corrupted firmware. The recovery process rebuilds the FTL from the NAND page metadata. QLC cells that have already worn past their rated endurance produce higher bit error rates during imaging, requiring more Read-Retry iterations.