2

Why do solid-state drives even have a limited number of read/write cycles, usually 100,000-300,000 cycles?

For hard-disk drives, it is understandable why they wear out. That is because they have moving parts that are in contact with each other. That means they have abrasion, which ultimately wears out the bearings.

However, SSDs do not have any moving parts. That means there no abrasion at all. Also, SSDs are made out of inorganic glass or ceramic, more specifically silicon metalloid doped with other elements. In chemistry, I learned that glasses and ceramics have a very simple chemical composition with their structure tightly held together by strong atomic bonds. Glasses and ceramics are distinguished between each other because the former has a disordered structure, known as amorphous, while the latter has an ordered structure, known as crystalline. Because of how highly stable they are, they do not change at all under there is an super high-energy event from an external source that causes the bond to break, such as a macroscopic force cracking the material, high temperatures melting the material, or super corrosive chemicals. Similarly, a few days ago, I did an Internet search confirming my suspicions that solid-state ceramic and glass capacitors last forever under normal operation because they do not age at all within design limits.

That contrasts with organic chemicals such as polymers. In chemistry, I learned that polymers are made up of hugely complex, varying molecules that are not inert and are only weakly held between each other by weak van der Waals forces. If a bond breaks within a molecule for example, the original molecule will almost certainly not reform. That is because there are so many possible different atoms to form new bonds with, so the original bond will almost certainly not be reinstated due to probability. Because they are unstable, they self-react, which causes them to self-destruct over time. That is greatly accelerated by external high-energy sources such as ultraviolet rays. Since electronic components such as OLEDs, liquid/gel-electrolyte lithium-ion batteries, and electrolytic capacitors are made at least partially out of polymers, that is why they break down over time.

So, shouldn't inorganic SSDs last forever (geological timescales) when its design parameters are never exceeded under operation, such as never having a power surge? Electrical conductivity is a property of electronic components (of which the transistor is a specific type, more specifically SSD bit cell), just like visible light conductivity is a property of glass. Passing light through glass never wears out glass given that it does not overheat, so shouldn't passing electric current through an inorganic SSD never wear out the SSD given that it never overheats or has dielectric breakdown?

CoastCity Lapse 00crashtest
  • 41
  • 3
  • 1
    Programming an SSD involves shooting charge using very high voltage to force it through what is basically glass. Erasing tries to reverse that process with a second voltage to pull it back out, but it isn't really reversible because lots of charge gets stuck, so theres a finite number of times you can program. – user1850479 Sep 18 '22 at 02:41
  • 2
    Such a long post for something that is well documented. There’s extensive discussions of SLC vs MLC etc and the mechsnisms involved in NAND flash wearout. – Kartman Sep 18 '22 at 02:43
  • Thank you so much. That basically answered my question! So basically, SSDs work on dielectric breakdown (arcing) whenever a write/erase cycle happens. No wonder why the process of SSDs is consumable/sacrificial, just like that for old CRT monitors, electric discharge lamps, and spark plugs. – CoastCity Lapse 00crashtest Sep 18 '22 at 04:38

1 Answers1

3

A full answer would be more than I have the time or energy to do this late at night, but this should set you in the right direction with the help of duckduckgo or google (good search terms are bolded):

Flash memory stores its data as charge in an insulated floating gate. The Fowler-Nordheim tunnelling involved in getting the charge there is, over time, destructive to the gate oxide. When the gate oxide fails, it's no longer possible for that particular FGMOS cell to store any data.

We have a couple of alternative technologies to avoid this failure mode, by storing the data some other way, but none are remotely as cheap as flash memory, at least not yet. The most promising ones are MRAM and FRAM, if you want to look those up. You can actually buy FRAM chips right now, they're just incredibly expensive (like $30 for a megabyte or two), so not very useful for SSDs.

Hearth
  • 27,177
  • 3
  • 51
  • 115
  • Flash wears out from write-erase cycles. FRAM has destructive reads so it wears out from just reading, though the number of cycles should be [long enough to be useful](https://e2e.ti.com/support/microcontrollers/msp-low-power-microcontrollers-group/msp430/f/msp-low-power-microcontroller-forum/132022/fram-endurance). – Spehro Pefhany Sep 18 '22 at 03:41
  • @SpehroPefhany FRAM endurance is so high it has to be expressed in scientific notation, and even besides that, all data from storage gets copied to a DRAM buffer (in modern computers, at least), so even program code in a tight infinite loop won't wear it out--DRAM can be written or read indefinitely with no failure. In the case of that microcontroller, which uses FRAM alone for program memory, you might have a problem, though. I can't see a hypothetical FRAM SSD wearing out before the computer it's in does, though. – Hearth Sep 18 '22 at 03:53
  • I just looked up what Fowler-Nordheim tunnelling is. It applies a very high electric field to force the electrons through the gate. So basically, SSDs work on dielectric breakdown (arcing) whenever a write/erase cycle happens. No wonder why the process of SSDs is consumable/sacrificial, just like that for old CRT monitors and electric discharge lamps. – CoastCity Lapse 00crashtest Sep 18 '22 at 04:50
  • Also, are FRAM and MRAM both specific types of NVRAM, because they both sound like they meet the non-voltatility purposes of RAM? Does the NVRAM used in routers have unlimoted read/write cycles, or is it linited to the 10^14 cycles shown in the link? – CoastCity Lapse 00crashtest Sep 18 '22 at 04:54
  • There's no arcing involved. Crt monitors did not arc in normal operation. Who said routers have NVRAM? Stop trying to draw correlation to things that are unrelated. – Kartman Sep 18 '22 at 05:42
  • You can buy MRAM also, even more expensive than FRAM. The cost can only be justified for radiation environments. https://aerospace.honeywell.com/us/en/products-and-services/product/hardware-and-systems/microelectronics/memories – Mattman944 Sep 18 '22 at 07:23
  • CRT monitors worked on the principle of arcing. The electrons had to jump between the cathode and the fluorescent screen. That is the definition of arcing. That is why cathode ray tubes are a type of electron gun. – CoastCity Lapse 00crashtest Sep 18 '22 at 10:47
  • 1
    @CoastCityLapse00crashtest That is not the definition of arcing. Arcing involves ionization and dielectric breakdown, and in a CRT there is no dielectric to break down (it's called a *vacuum* tube for a reason). There's no arcing involved in Fowler-Nordheim tunnelling either; F-N tunnelling is a type of tunnelling, a quantum mechanical effect that does not involve dielectric breakdown at all. – Hearth Sep 18 '22 at 12:31
  • @CoastCityLapse00crashtest If dielectric breakdown was involved, flash memory would only have a write endurance of 1 cycle. Gate oxide breakdown will always destroy the gate immediately; there's no reusing a part that's undergone oxide breakdown. In fact, that's the usual failure mode for flash memory--a side effect of F-N tunnelling is that it makes true breakdown easier to happen over time, and eventually it gets to where normal operating voltages will cause the gate oxide to break down. – Hearth Sep 18 '22 at 12:35