SLC vs. MLC, Continued
Without getting too deep or technical, think of a transistor as a two-story building. The ground on which the building is erected is the substrate, the first storey is the floating gate, and the top storey above it is the control gate. Between the substrate and the floating gate is a thin layer of oxide material. Think of the oxide as the stairs people have to go up to get from the ground to the first storey. Lastly, think of electrons as the people wanting to travel from the ground level up to the first storey. If there are no people in the first storey, then the building has been evacuated, or “erased.” If there are people up there, then you have data resident and the cell has been “programmed.”
As said earlier, SLC only requires one electron per cell; MLC requires more than one. It takes more energy to move more people from place to place. The metaphor frays a bit here, but essentially it takes more voltage to move and hold electrons in that floating gate. More voltage acts like more people stomping on those stairs, wearing them out more quickly. Given enough time, the stairs will break. The electrons effectively saw a hole in the oxide insulation layer and the floating gate can no longer reliably hold its electron(s). When this happens, the SSD controller flags the problem spot as a bad block and passes its data to another block.
Because MLC uses more electrons and requires more voltage (more sawing) to move those electrons, MLC cells will wear out faster than SLC—about 10 times faster. You might see SLC drives noted with 100,000 cycle ratings and MLC with only 10,000 cycles. If you erased and reprogrammed an MLC block once per minute continuously, it would exceed its life expectancy in about one week.
Fortunately, SSDs don’t write to the same block over and over. The drive controllers use “wear leveling” to distribute the programming cycle load evenly across the drive’s million’s of NAND memory cells. An 8Gb NAND die with 4,096 blocks, if perfectly leveled, wouldn’t start exceeding its 10,000 cycle mark for 75 years. This is why having a controller with good wear leveling algorithms is so important. Moreover, the more cells are in a drive, the more real estate there is for wear leveling. This is why there’s a nearly linear relationship between SSD capacity and drive endurance. Wear is also why enterprises have historically gravitated to SLC drives. The extra performance is good, but large corporations are positively phobic about the possibility of premature failure.
