Alphabetti spaghetti: SanDisk adds SLC cache to TLC SSD

Acronym alert: SanDisk has upgraded its Z400 PC SSD by changing it from an MLC drive to a TLC one with an SLC cache, doubling its capacity as well making it faster. The Z400 SSD used 15nm MLC (2 bits/cell) flash when it was introduced a year ago. SanDisk has kept the 15nm NAND but changed it to TLC (3bits/cell), upping its …

[Andy Nugent]

endurance has also been extended?

"The product's endurance has also been extended.In terms of terabytes written (TBW) during its warranted life the Z400 numbers were 20 TBW (32GB), 40 TBW (64GB), 72 TBW (128 and 256GB). The Z410's equivalent numbers are 40 TBW (120GB), 80 TBW (240GB) and 120 TBW (480GB)."

so the 128GB drive used to have an endurance of 72 TBW, and now the 120GB drive has an endurance of 40TBW.

Isn't that a drop in endurance?

[JeffyPoooh]

SLC MLC TLC

The difference is 'just' * in the A/D stage. Conceptually, the 'Level' (bits/cell) could be defined and implemented in a trivial look-up table. The table column could be selected with a flag. (* There will be optimization complexities, as always.)

Caching is hard.

But the capability of fiddling with the Level (bits/cell) should have been baked into the original design, adjustable in 'firmware'. Even beyond the capabilities of the physics of the process du jour at the time.

(edit: As of this morning, "In total, your posts have been upvoted 10021 times and downvoted 3071 times." Yay! Just crossed 10k. :-) )

[Toastan Buttar]

Re: SLC MLC TLC

Each block in a MLC die can be erased as desired (MLC or SLC) by the firmware on board. Same goes for TLC. It usually makes sense to permanently partition the blocks in such a way that the important stuff (or stuff which is going to be re-written many, many times) is stored in SLC. Whilst it would make more sense to keep this partitioning static, as you say there's nothing in the physics to prevent you doing it dynamically. It's just FAR more difficult to keep track of wear-levelling, etc in the dynamic case, with very little benefit to be gained in real-world use-cases.