Wednesday, May 25, 2022

How SSDs Work – SSD Architecture, MLC vs. TLC

Hey everyone this is Steve from gamers Nexus Donna and I am joined once again by Kent Smith from LSI and today we're talking some more about the about evolving flash types and a couple of other topics so you know you've all seen mlc and TLC and and things of that nature when selecting a solid state driver when .

Reading reviews and it's beyond the spec sheet on Newegg or wherever you look at your looking there's not a whole lot of technical depth so that's where we're going today and starting with the flash types the NAND types mlc as an acronym means multi-level multi-level cell TLC is triple level cell and that's sort of where a lot of it stops so can I get you .

To tell us some more about how how that actually works what it actually means sure and just to clear everybody we skipped over SLC of course range today you really only see in the high end Enterprise SLC meaning single-level cell but what that's talking about is the actual voltage levels that are kept within a cell so single-level cell .

Meaning there's really two voltage levels effectively on-and-off it's storing for one bit when you go to mlc or two levels you actually store four bits or four levels so that means there's there's four voltage levels that are being stored in an individual cell and versus just two levels so it's a lot harder to differentiate so if this .

Particular level here suddenly has a reduced voltage if something drops it now looks like the second of those four levels and so that's a bit error and it's only because of the ECC that you'll be able to correct that the problem is ECC has to cover a number of bits a number of cells all at once and if too many of them are wrong you've got an .

Error so TLC then is 2 to the power of 3 or 8 I won't bother putting 8 fingers up here but it just gets that much harder in that same cell in that same range you're trying to put 8 different voltage levels so it's really tough and Adam and performance pretty heavily so is there anything we can do on the controller level to mitigate the impact yeah in .

Fact one of the biggest drawbacks of that of course we were kind of talking about that error rate because those will those bits will shift over time so your controller is going to try to take care of that with higher powered error correction schemes like we we announced the shield technology on the latest generation the SandForce controllers and .

So that'll be able to do a lot more air correction for future NAND flash as as it continues to evolve the the speed is also an issue as you mentioned TLC is going to be a bit slower than MLC and so you know the the pro of TLC is it's it's cheaper to make because you get more bits in the same silicon but the drawback is it gets a little sore so the .

Controller has to do other things to mitigate that you look at potentially having more die so then you want a bigger drive you know and today you know the 240 gig capacity is pretty difficult 256 you know that's a pretty common sweet spot and that for the most part is is accessing as many die as you can to keep it going as quickly as possible now .

What do we have TLC just sort of got introduced to the consumer market not long ago by what happens next is their next step after TLC the announcements that have gone on from the flash manufacturers are talking about 3d NAND and bean and this is a category of a flash memory if you think about SLC and mlc and TLC as single-family homes you .

Know in a city and then in order to take advantage of that same real estate we want to build vertically right and so there's a couple of different processes that have been announced that essentially make it sort of like high-rise apartments so you can get you know 10 or 20 additional levels for that same footprint of silica and so it has .

The advantage of you can now increase that I'm a tree you can go to a higher process node so that lowers your error rate it has the disadvantage of a little bit higher cost but you then divide that by having you know ten or twenty or thirty different levels in the same footprint and you get to go past that .

Cost curve today you know it's really not cost-effective a lot of manufacturing steps have to be worked through but you know MLC was not possible you know five ten years right now so it'll happen it'll go for programming race cycles can you give us an overview of what it actually means you know the basics are it's how many .

Times you can write to the flash and erase it before it starts decaying heavily and eventually dies right right so how does that all work okay so with in flash memory you can write to individual pages at a time but you have to erase whole blocks at a time each time you do that erase process you effectively decay the the cells now you .

You could also argue it's the program step that does the DK now that's probably more appropriate but either way you can't do a new program until you do in a race so it doesn't matter which side you're counting the count will incurring increment and you will have a decay in the cell so what happens is as a decays the cells have less ability to .

Hold a charge so at the beginning of life you know remember we were talking about on mlc you have those four levels it's really easy to keep those charges as you cycle it more and more those charges tend to start you know migrating and pretty soon charge three and four look like chart four and maybe charge one and to look like charge one and so .

Then it's harder and harder to differentiate so again the ECC does the job of correcting that but if there's too many errors in a in a single area the ECC can't can't recover from it so that's where the the cycle count comes in if I say if I have a flash that has 3000 cycles that essentially means the flash is able to go through 3,000 .

Different in that cell doing a program and a race cycle and after you do the 3000 the cell according to the factory says it can still hold those charges for whatever the spec is if it's consumer class it's typically about a year so so we actually have a cell decay exactly yeah so you could potentially set your laptop .

Computer on a shelf and you know just forget about it you're using your desktop all year long you finally go out for the next show and suddenly your laptops empty because it's been a year since you've turned it on so zap sort of like your car battery if you leave your car long enough the battery will drain it's that same idea .

Okay and another topic is preconditioning so I've mentioned this when talking about our test methodology on the website we do precondition the drive before each type of test and you guys kind of have the basics of it but can you give us a top level look at preconditioning why it's important and then and then maybe explore why our .

Readers would see lower performance or higher performance given different intervals of use sure oftentimes when you see a review or discussion around an SSD you'll see that it has a certain spec you know I can write so many megabytes per second a lot of times you'll see it's an up to number and in many cases that'll be a fresh out of box .

Number and the problem with that is the drive will only live in that lifespan within its lifespan it'll only live in that time period for less than one percent of its life you know maybe even point O one percent of its life so it's not really helpful because as soon as you start writing data to it it has to garbage collect all SSDs have to garbage .

Collect now some will be faster than others add garbage collecting but they all have to garbage collect it's during that process that the performance will drop down because during garbage collection if you're processing moving data you can't take any new data in and so it's that that that fight that internal fight if you will over the .

Bandwidth to the flash the host and the controller trying to move data around that will actually slow the drive down so when when you look at benchmarks you want to make sure you understand whoever did that analysis you know did they do the proper preconditioning and there's various ways of doing it .

There's many theories on it but at a minimum you have to do some kind of pre conditioning now we generally recommend if you really want to look at real world you want to sequentially precondition before you do a sequential test and you want to randomly precondition before you do a random test now if you believe your environment is a mix that's fine if your .

Test can have a mix of sequential and random then essentially you just need to run that test long enough to ensure garbage collection is operating with that particular configuration and you will see the performance of what the real world would look like with that same condition right and so for any of you who want to do testing on your own .

Drive at home to make sure you're getting the specs that you've seen online generally what we do with when testing SSDs is musuem etre and I can put all the links in the description below and as gamers you'll want to do 4 K random preconditioning and testing because you're gonna be using mostly very small .

Files where as sequential is more for the heavy users who are writing very large files frequently so I'll put some basic instructions on how to precondition your drive and do a real actual test that way you get a actually but the best thing to do would be tested straight out of the box or secure erase it and then do your your precondition .

Test so you can see the real difference be the difference I really understand yes you have hit that garbage collection period because you will see a drop especially on random writing right so all that again in the link in the description below and thanks again Kent and we will see you all next time thank you .

RELATED ARTICLES

Most Popular