1
\$\begingroup\$

If I have a NAND flash which is say 70°C rated, what is going to happen to that NAND flash when I heat it to 80°C or 100°C?

I'm thinking of a scenario when the 70°C rated sample actually starts producing errors at 71°C and I'm not looking at a sample which will have no problems above 100°C.

My current assumptions are that it can either the logic be corrupted (like in RAM and CPU) or the actual NAND bits can flip more often?

I'm basically trying to figure out a high level testing methodology that can run from Linux.

Thanks!

\$\endgroup\$
3
  • \$\begingroup\$ A NAND gate? Or a NAND flash? \$\endgroup\$ Commented Aug 4, 2016 at 23:38
  • 2
    \$\begingroup\$ Simply said, none of the numbers listed in the data sheet can be guaranteed. Errors may range from something as subtle as an increase in the Vlow threshold to the output never changing. \$\endgroup\$ Commented Aug 4, 2016 at 23:55
  • \$\begingroup\$ @IgnacioVazquez-Abrams NAND flash. \$\endgroup\$
    – tothphu
    Commented Aug 5, 2016 at 0:05

2 Answers 2

3
\$\begingroup\$

A few things I can think of.

  • Chance for memory to be erased (poor data retention). In flash memory, charge stored in the insulating gate of a MOSFET acts as the memory element. Under high temperature this charge leaks more quickly. From ti paper: "with sufficient thermal activation, all bits could lose their charge". Also from the same paper, gate oxide damage.
  • Trapped charges injected into the oxide due to the high temperature will build up. Over time this will add up create a voltage offset on the gate (making FET more leaky). This could create a situation where the FET is permanently on (unless the gate is driven sufficiently low) if enough charge builds up.
  • Shorter lifespan in general. Heat causes a lot of processes to speed up exponentially: (Arrhenius law). Thermal expansion can cause peeling of metal vias from their respective layers or a host of other problems.
  • Possible complete failure due to thermal runaway (unrecoverable).
  • Higher leakage overall. More power burned. From pdf on leakage, in 65nm process: leakage at 0C is ~10x less than 40C, which is ~10x less than 100C.
  • The circuit will run slower than expected, due to more collisions of free carriers.
\$\endgroup\$
1
\$\begingroup\$

You will not see anything because they are masked by the front end. When you get bit errors, you just skip to a new page. You will just notice the device get smaller, faster.

If you could see them, the failure that you will mostly likely see is soft errors due to noise, but it is different than you think. As the temperature increases, the densities of states expands and thresholds move due to drain-side charge sharing that shortens the effective channel length. This causes problems in clocked systems that gives read errors, but this is not due to the FLASH transistors, but the charge amplifiers.

There are two other scenarios that come to mind for hard errors.

1) You get more charge getting "stuck" in the oxide during injection due to these increased energy levels that would generally not trap the charge. My answer on simulating floating gates has a band diagram for this injection, and just imagine that electron get trapped in the barrier. (a note, this is the behavior that causes things to fail after millions of writes. Increasing the temperature just speeds up the process)

2) The extra energy due to heat will cause a hole to get enough energy to "jump" through the oxide. This makes a literal crater in the oxide. I could create this behavior in the lab, but I had gate control of actual device and a temperature chamber.

\$\endgroup\$
2
  • 1
    \$\begingroup\$ Your first paragraph is incorrect in many situations. If you are the developer of the flash translation layer, then you see all the errors and it is your job to map around them. Even if you're using a NAND with ECC builtin, the user is responsible for wear leveling and bad block mapping. The statement you made is only true of SD and managed NAND type devices, or if you are working in software above the flash translation layer. \$\endgroup\$
    – rjp
    Commented Aug 12, 2016 at 11:54
  • \$\begingroup\$ @rjp Good points, but using Linux or something where you have an abstraction layer, there's no way to see what's going on in the device, so I wrote from that from the standpoint of what can be seen. \$\endgroup\$
    – b degnan
    Commented Aug 12, 2016 at 13:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.