You're not imagining things – USB memory sticks are getting worse
- Reference: 1707327010
- News link: https://www.theregister.co.uk/2024/02/07/failed_usb_sticks/
- Source link:
CBL Data Recovery [1]posted that the quality of newer memory components in microSD and USB sticks is declining, and it reported that USB sticks where the NAND manufacturer's logo had been removed from the chip are increasingly turning up in its data recovery laboratory.
It suspects that flash chips from manufacturers such as SK hynix, Sandisk or Samsung that fail quality control checks are being resold into the market, but marked as components with lower memory capacities.
[2]
"When we opened defective USB sticks last year, we found an alarming number of inferior memory chips with reduced capacity and the manufacturer's logo removed from the chip," wrote CBL Managing Director Conrad Heinicke (translated from German).
[3]
[4]
Heinicke said that many of the USB sticks actually contained microSD cards that had been mounted onto the circuit board and were being managed by an external controller chip. While USB sticks like this were mostly promotional gifts, he said, there were also branded products among them, adding: "You shouldn't rely too much on the reliability of flash memory."
Heinicke's view is that the adoption of multi-level cell architectures, where a single memory cell is coaxed into storing more than just a single bit by varying the voltage, has also exacerbated the situation. With quad level cells (QLC), for example, four bits are stored per cell, which means that 16 states have to be distinguished.
[5]
This path was chosen by the NAND flash manufacturers because it delivers greater storage density, which means higher capacity drives and lower costs per GB. But it also has implications for the endurance, or longevity of the cells. In other words, the cells wear out faster the more bits they are used to store.
[6]Japan stumps up more cash for Kioxia and Western Digital to make memory chips
[7]China 'readies production' of homegrown high-bandwidth memory
[8]Seagate used bullying tactics to prevent union recognition, claims Unite
[9]Japanese government finally bids sayonara to the 3.5" floppy disk
"The downside of success in increasing storage density: writability (endurance) and storage capacity (retention) decrease," he wrote. "This means that the number of maximum possible write-read cycles decreases and the risk increases that the charge differences, which are read as bits, can no longer be read due to the aging of the insulation layer that holds these charges."
According to our sister site [10]Blocks & Files , a single-level cell supports around 100,000 cycles while a (2-bit) MLC cell wears out after around 10,000 cycles, a triple-level cell is limited to 3,000 cycles and QLC supports about 1,200. Upcoming penta-level cells (5 bits per cell) are likely to have less than 600 cycles.
"Even with high-quality memory chips, the effort that manufacturers have to put into error correction mechanisms in the controller is enormous. It is not surprising that data loss occurs with USB sticks containing decommissioned chips," Heinicke states.
He posted a few tips for flash users. If you frequently write and delete data on USB sticks, such as when using them as backup devices, for example, you should use several sticks in rotation.
[11]
If flash media is to be used to store anything for an extended period, use high-quality storage hardware, and keep it somewhere cool, as high temperatures may accelerate data loss.
Flash storage should be accessed annually or semi-annually, as this allows the error correction mechanisms to internally copy "fading" data, and devices should not be filled to capacity – leaving unallocated space allows the internal data maintenance and error correction mechanisms to function longer. ®
Get our [12]Tech Resources
[1] https://www.cbltech.de/pressezentrum/unzuverlaessige-flash-speicher
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/storage&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZcQLmE-sXZ8HhC9tuKC-@QAAAZc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/storage&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZcQLmE-sXZ8HhC9tuKC-@QAAAZc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/storage&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZcQLmE-sXZ8HhC9tuKC-@QAAAZc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/storage&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZcQLmE-sXZ8HhC9tuKC-@QAAAZc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2024/02/06/japan_western_digital_kioxia_subsidy/
[7] https://www.theregister.com/2024/02/01/china_high_bandwidth_memory/
[8] https://www.theregister.com/2024/02/01/seagate_northern_ireland_unionised/
[9] https://www.theregister.com/2024/01/29/japan_government_floppy_disks/
[10] https://blocksandfiles.com/2022/05/01/mlc/
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/storage&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZcQLmE-sXZ8HhC9tuKC-@QAAAZc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[12] https://whitepapers.theregister.com/
My SanDisk stick died last week. No, not some recent purchase, but one from 2008. I didn't write to it all that heavily, but I did read from it all the time. I was actually quite sad about it. I highly doubt its replacement will last that long, but I'll be sure to choose it very carefully!
Depending on your use case, it might be worth looking at M.2 drives. They cost more because you need an enclosure but they'll last a hell of a lot longer, and they'll be faster.
Size isn't everything
How big was your stick? most old ones were probably 8/16 or even 32GB. Since then we have found many alternatives to storing larger lumps of data. But if you just need to boot a computer or transfer some files then you may not need more today. But try and get a 8/16GB stick or SD card these days and even if you are lucky you won't find it cheaper than 32/64GB.
Paying for capacity you won't use which makes what you do use less reliable is not progress.
In other news...
... the gamers paying a premium for memory with colourful "heat spreaders" glued on top are being ripped off left, right and centre. It's really quite effective at disguising the provenance of chips and you don't even need a packaging facility or laser etching facitilies to do it.
This is the kind of thing thats as old as the hills.
None of this is new, no-name fly-by-night manufacturers (including those crappy branded ones you can order in batch) have been pumping these things out for years. I guess the main issue is that they are becoming more prevalent as unknowledgeable buyers get the cheapest 1TB super ultra hyper fast USB thumbdrives on Amazon and are surprised when they try to write more than 256MB and it starts overwtiting old data. And by then, the seller has bolted.
Simple solution?
Make Amazon liable for defective products, and don't allow them to disclaim that liability via customer license agreements.
I know, I know, there's no political will to enact that. But a girl can dream.
Re: Simple solution?
So your plan is to hold the reseller liable for wholesalers or manufacturers malicious or unintended failures or side-effects? That would certainly reduce Amazon's pervasiveness in the market, but with the added consequence of eliminating all resellers entirely!
The troubling bit is the finding: “ there were also branded products among them”…
So is that Sandisksamsung/WD et al usb memory stick really worth the label and the money?
multiple bits per flash cell
So. Not actually binary any more. I seem to recall that Bletchley Park's first offerings were decimal - approx 3.2 bits' worth per vacuum-valve "cell".
My father bought a 64GB Integral USB memory stick from a Tesco at Christmas so I could copy some videos from my PC onto it for him, and it was terrible!
I was trying to copy about 60GB of videos to it and i was getting max of 6MB/s transfer speeds, sometimes down to 3MB/s so it took me hours.
For a while I was even wondering if it was one of those dodgy ones that would report an incorrect capacity and start overwriting data after it filled up a couple of GB, but all the 60GB of data was accessible once it eventually copied over
If i compare it to my few year old 32GB Kingston Datatraveller G3 on the same PC i can get in the region of 70MB/S write speeds on that.
I have been trying to find a reliable USB stick with quality NAND to use a portable SSD, as the M.2 enclosures are a bit to big for my needs but not found one yet.
Recently had cause to want a performant usb stick, the only ones I could find were from Samsung, and surprisingly available from PC World
https://www.currys.co.uk/products/samsung-fit-plus-usb-3.1-memory-stick-256-gb-silver-10245087.html
Whilst not as fast as the SSD (only up to 400iMB/s) it was good enough to run a Windows VM from.
Trouble is what type of storage do they use and hence using it for a live VM might not be a good idea.
I've always found the Samsung ones get REALLY hot within seconds though.
Quad level is slow. I have a two TB quad level SSD drive, and it goes down to 60MB per sec.
HOWEVER it can write to the same cells as one-bit cells at 1100 MB/sec. So you can write 500GB at full speed to an empty drive. If you copy more, you go down to 60MB/sec. If you stop copying, it copies from 1 bit to 4 big cells in the background. That’s 3.6GB per minute. So after one minute without writing you can write 3.6GB at full 1100MB/sec speed. Right now, that drive has 800 GB filled, 1200GB 4bit cells or 300GB 1 Bit cells unused so I can write 300GB at full speed.
Note that today tri-level SSDs are a lot cheaper, so there is much less price advantage. But I compared it to a 2TB spinning drive, which isn’t much cheaper, which is a lot slower, and a lot heavier - that 2TB SSD is just 45 Grams.
ValiDrive
Run, don't walk, over to Steve Gibson's website and get ValiDrive.
https://www.grc.com/validrive.htm
You need this to detect deliberately misleading USB storage.
I discovered the hard way that doing an initial backup of lots of files to a mounted USB stick with LUKS encryption caused so many writes that it wore it out.
Now, I create a temporary encrypted filesystem no bigger than the stick, back up to that and then "dd" the image onto the raw device.
>With quad level cells (QLC), for example, four bits are stored per cell, which means that 16 states have to be distinguished.
I was preparing to lambast such a sentence and ridicule it to pieces; 4 "levels" of voltage would be able to store 2 bits, not 4. However, it seems that by "level" the industry does indeed mean "bits", according to the wikipedia page on the matter. I fear I have done El Reg a temporary disservice.
Now, why can't they just call it as it is, either 4 bit cells or sexdeclevel cells, or hexadecalevel cells?
I got so sick and tired of all the crappy USB sticks available. I could go around to every store in the area and not one would have a decent one. I went through several and they all had such crappy, pulsating write speeds that they were pointless.
It seemed to me they were missing some SCSI logic and buffering wasn't working correctly, in addition to it being crappy memory.
So I ordered a big hundred dollar Corsair GTX that does about 400 mb/sec (according to iotop "actual write speed") for both reads and writes.