New Type NAND Chip - Sandisk QLC (TSOP48)

[Gregory]

What I found on the Internet is that it is memory is Sandisk 96L BiCS4 QLC 16K

ID: 45 73 9C B3 76
Page: 18786
Sector: 1174+1170x15

Has anyone had this type of memory and tried to recover data?

[csava]

Can't believe it, is there already a USB flash drive using this NAND flash now? What model is it from the Phison controller?

[Gregory]

PS2251-19-26, possibile new protocol, I read in WL DDR but something is not OK.
Stucture 1174_1170 can't find ECC but this should be BCH

[csava]

Okay, the 2251-19 uses a BCH error correction engine. However, I don't have a donor with the same flash memory. Could you take a photo of the appearance and model of the USB drive? I'm trying to find a matching donor

[Gregory]

Could you take a photo of the appearance and model of the USB drive? I'm trying to find a matching donor

Sorry. Flash drive without housing, I don't know what he looked like.

[arvika]

ECC for this controller was developed by our laboratory.

https://www.facebook.com/arvikadatareco ... tVB83aLP8l

[Gregory]

OK. XOR before ECC ?

[arvika]

OK. XOR before ECC ?

No. ECC before XOR.

[csava]

I consulted my flash drive supplier friend, and he mentioned that many Bics4 QLC flash chips are not considered top-tier by the kioxia company. Instead, a significant portion of them flows into third-party packaging facilities in wafer form. Therefore, I believe that Bics4 QLC is likely to appear in lower-quality SSDs and USB flash drives, but the overall quantity is not substantial. According to my friend, the quality of Bics4 QLC is quite poor, and most products using it are expected to fail after a few months of use. However, what is more noteworthy is the Bics5 QLC, as it is more widely used than Bics4 QLC. I have already obtained a supply, and I will soon conduct research and evaluation on it.

[Gregory]

Possible flashdrive with 2251-19-26: Kingston DataTraveler Exodia M, PNY Turbo Attaché R USB 3.2. I think this controller will be popular in USB 3.2 flashdrive.

[pepe]

many Bics4 QLC flash chips are not considered top-tier by the kioxia company.

how could anyone with brain somewhere in his body think that a QLC chip could be top-tier?
the whole idea is simply offending that they even started to manufacture such thing.
this is the end imo.

i also wonder why don't ssd manufacturers make it configurable, if you want to use it in SLC, MLC or TLC mode with the corresponding loss of capacity. Most if not all controllers use some portion of the capacity as an SLC cache, which suggests these chips can be used in slc mode, manufacturers just don't sell such product. I would gladly sacrifice capacity for reliability and endurance...

[csava]

many Bics4 QLC flash chips are not considered top-tier by the kioxia company.

I would like to express the quality grading system for the same model of NAND flash, similar to the Spectek company's grading system. When only a very small number of chips are classified as top-tier (premium), it indicates that the manufacturing process of the flash memory is not yet perfected.

[pepe]

Just think about what QLC means and you will say this is crazy. You have a nand cell (transistor) and put some charge on its gate and hope it will stay there. It does, ok, we are happy we can read it back. Then you think let's double the capacity and store 2 bits in a single transistor. Yeah, ok, but that requires to divide the charge level to 4 bands and introduce 2 more compare levels, ok, that's fine, we lost some SNR on the way but we will compensate it with more robust ECC. Then come the greedy butheads and say, hey, if it goes that way, why don't we work with 4 times that level??
Do it for us, friend we must increase the density on market demand.
yeah, just thinking, if the full level is converted to voltage and you get 1V as max, bellow 0.5 is 0, above it is 1, doable (SLC)
let's go QLC:
0 - 1/16 is 0
1/16-2/16 is 1 and so on, it sounds good doesn't it?
well, that makes about 62mV bands, which you need to discriminate during read. it leaves you with very low noise immunity.
in general, if you want to make an N-fold increase of density, you will have a 2 ** N impact on the tolerances.
each transistor has a transfer function that will distort the readout value as well. Yeah, they can probably make very good matching transistors, but still, when the demand makes the tolerance so braindead tight...
Luckily the engineers managed to solve all of the problems so we can start selling our new, fancy, hypercapacity NAND chips.
what? it can only be written 2-300 times? psst, let's give it a fancy name and hide that detail between the NobodyKnowsWhatItMeansButSoundsVeryGood parameters, let's call it QLC, wow it is not a boomer SLC shit, this is a QLC!!! a QVO ... a whatever.
And the crowds buy it blind.


and that's why there will be no reliable storage in some years, I would be happy with MLC but it is completely finished too...

[arvika]

The explanation that pepe described can be nicely illustrated with an example. This is what it looks like in practice with a standard reading. More bit errors than real data.

Attachments

[Gregory]

I'm not surprised I didn't find ECC, if that's what it looks like on a new donor. Does the memory use one of the available RRs for Sandisk or some new ones?

[arvika]

I'm not surprised I didn't find ECC, if that's what it looks like on a new donor. Does the memory use one of the available RRs for Sandisk or some new ones?

You will get good plots on Sandisk v2 and Sandisk v5 (except last one). But it not help so much.

[Gregory]

OK, I found donor with PS2251-19-26 but different NAND, can confirm that this contoller use ECC before XOR.
I have different structure 1138_1134 and to be more precise it is 1137_1_1133_1, so I guess this Sandisk is 1173_1_1169_1.

[arvika]

can confirm that this contoller use ECC before XOR.

No. ECC before XOR.

Different case can have different ECC algorithm, but here should be same.

I have different structure 1138_1134 and to be more precise it is 1137_1_1133_1, so I guess this Sandisk is 1173_1_1169_1.

This part I do not understand

[Gregory]

This part I do not understand

Layout naming from FE like here:
e.g. 1151_1_1143_1
http://www.flash-extractor.com/library/ ... 98_b3__2x1

[arvika]

My notes tell me about 1174_1170 structure.