Isn't SLC the most reliable? I thought the more levels meant higher capacity at a cost of reliability?
SLC is the most reliable.
The more layers are added, the more voltages have to be dealt with. This increases the variance/drift. It's why the Samsung 840, an early planar TLC drive, had to have its data regularly rewritten to the drive to prevent read speeds from falling.
In addition to the issue of how many layers there are (single = SLC, double = MLC, triple = TLC, quad = QLC) there are others:
1) process node size (larger is more reliable, e.g. 40nm vs. 15nm)
2) 3D vs. planar. 3D is more reliable. 3D fabrication is mainly what has made TLC so much more relevant in the market than it was initially.
3) various controller design factors, like DRAM-using vs. dramless (DRAM-using is better for reliability, except with Sandforce* which is a more complex case), error correction sophistication, and the presence of a capacitor to protect against power failure
4) how much NAND is set aside
5) the grade of the NAND
6) "SLC" caching. As far as I know, no drive has actual SLC NAND in its so-called SLC cache. Instead, a portion of MLC or TLC NAND is "functioning like SLC". This improves speed but I don't know how much it affects reliability. If it were true SLC then reliability would increase, certainly.
7) firmware/controller bugs
8) cooling (e.g. metal casing, thermal pads, heat sinks). This can affect performance consistency, which is a form of reliability (performance reliability).
*In terms of Sandforce, it was a dramless design that had improved
theoretical reliability via on-the-fly data compression. This reduced the amount of data written quite a lot versus other controllers at the time; Sandforce had an excellent write amplication factor, with compressible data. Unfortunately, the design made TRIM mostly ineffective, causing the drives to slow, especially in steady state, and not recover. The design also wasn't very reliable (most severely with the Vertex 2 with 64-bit NAND in the 240 GB size) and was further worsened because Sandforce favored locking the drive to "prevent people from looking at their special firmware design". There were a lot of firmware updates for both the 2nd and 3rd generation controllers. It seems that the third generation eventually became pretty stable with 64-bit NAND but the performance, especially with a drive in need of TRIM recovery, wasn't competitive with later companies' designs. My speculation, as to the main reason why Sandforce is no longer a significant market player, is the poor compatibility with TRIM.
