[NTLK] flash longevity question
newton at chichak.ca
Mon Sep 19 14:28:33 PDT 2022
> On 2022-September-18, at 18:26, Dan <dan at dbdigitalweb.com> wrote:
> Thanks Andrei. I knew most of this. And it is interesting that the NAND in say a Wii console (SLC internal NAND) does have 100,000 cycles. But the WiiU which is MLC only has 10,000. So some devices did have the 100,000 write cycles depending. I assume that is the early to late 2000's era, and later on they started with MLC and 10,000 cycles. And TLC is even worse at 3,000 write cycles (and I think most high capacity drives/chips are TLC these days).
In a handwaving way, you can treat FLASH cells as a couple of contacts with a capacitor across the top. If the capacitor is charged up with electrons, current can flow between the contacts. If the capacitor is depleted, no current flow. (for the pedants, yes, semiconductors, depletion layers, metal oxides, dopants, piranha, photolithography, gate, field effect, N-Channel)
When you erase FLASH, it charges up all of the capacitors, current can flow in all of the cells, and they all read back as 1s. When you write FLASH, the capacitors for the cells that makes up a byte gets depleted and is read as a 0. Easy stuff, flow == 1, no flow == 0. At least for SLC (single level cell) FLASH.
When you get into MLC (multi), TLC (triple) or QLC (quad), instead of the capacitor being charged or not, you have a particular voltage being stored. Now you have to read it out with an ADC (analog to digital (voltage to number)) converter to see what voltage was stored and what 2, 3, or 4 bit pattern that single cell was holding.
Now, stir in a sprig of time and the voltage can drop, changing the pattern interpretation of that cell. WAT? My data got corrupted? Yes, but it was cheap. (I see that QLC is rated with a life of 1,000 cycles, yeah don’t go there)
Capacitors are really good at holding a voltage though. You can pull apart a camera flash unit and there’s a big capacitor in there that is used to trigger the flash tube. That capacitor, after many years, will still be charged up and ready to eat the end off your screwdriver if you accidentally/purposely bridge the contacts.
The info that I can find quickly points at 1 year to failure if you erase/write 300 times at high temperatures (85C (wayyyyy hot in Freedom units)), and 1000 years if you never erase/write it. So don’t drop your Newt in close to boiling water and, remember erases kill, not time.
> On 2022-September-19, at 02:34, B <kosmicdollop at saber.net> wrote:
> But now I have another question: what about RAM? I suppose that also wears out eventually.
No, RAM works a completely different way. Some is just a bunch of transistors (SRAM (static RAM)) designed as flip-flops that are set up to be stable as either 1s or 0s and will sit there for decades (or until the power is turned off). The other type of RAM (DRAM (dynamic RAM)) is a bunch of capacitors, like FLASH, but it is designed to leak out in a few milliseconds, and has to be pumped back up (refreshed) periodically to retain the data, but the processor takes care of the refreshing automatically.
Computers have been using DRAM for decades now. Take a look at an ATM in the USA, some of those things have been in service for decades (sigh) running windows 95 or even DOS, and it’s usually the capacitors in the power supply that blow up first, followed by anything that moves (disk drives and fans). The chips are really understressed, they’ll still be viable when the boards are tossed into the toxic metal recovery burn barrels.
More information about the NewtonTalk