....
How do car manufacturer deal whit memory deterioration when they design the electronic for the cars that are made today? A car should be designed to last at least 25 year’s right?
Hello,
well, I'm working in the automotive electronics industry, in the development, and have taken care about memory components their technology, reliability and failure analyses (SQM).
The reliability is described with the parameters data retention (time length of data storage) and endurance (number of Read/write/erase cycles).
Automotive electronics are generally tested for 10 years life time. (environmental tests as temp shock, vibration, etc.). But the warranty is 1-3 years only.
For every car, there's a guarantee that you can get spare parts at least 10 years after EOP, 15 or 25 years for some premium cars.
So if the data (steering program in a flash memory) in a product will be corrupted, you only have the chance to buy a spare part.
All nonvolatile memories, as PROM, EPROM, EEPROM, Flash (NOR, NAND, whatever), or hybrids as memory cards, etc. all rely on the very same semiconductor structure, that's the "floating gate", i.e. a semiconductor which can be filled with charge like a capacitor plate, and surrounded completely by an isolator, SiO2, which inhibits the drain off of the charge .
This structure determines those two reliability parameters mentioned before.
There are different techniques, how to fill that gate (by which tunneling effect), how many states it stores (single bit or multi bit), how big the gate volume is, i.e. how many elementary charge it can carry, and perhaps some others.
If you select one set of parameters, which is most reliable (single bit, big cells), the physics stays the same for each of this class of memory, and so does the reliability.
Comparing the different supplier for compatible memories i.e. same type designator, you will anyhow find very different values for the data retention, the first 10 years, the 2nd 20 years, and the fourth 200 years.
Strange, as they all use the same physics, very similar structures and production processes as they often use the same foundry, and so on.
Ok, we tried hard to find out, how the different companies define those parameters, and how they really measure it. It was/is impossible.
This information obviously is their crown jewels, their holy grail.
You can imagine, that such parameters have an underlying statistical distribution.
"Typically", in the median, 20 years are reached for data retention, but how many parts will live so long? 50%? How does the distribution look like, Gaussian, Arrhenius, width of the distribution?
Do they define data retention on 100% of the memory-bit-array, or is it that data retention defined that let's say 3%, 50% or 97% of the cells are defective?
( That's the way "lifetime" of light bulbs is defined)
That's all not defined in the specifications and ANs, and there is no standardized test procedure, and statistical calculus available.
That explains the big differences between the different suppliers, as they all use different baselines of statistics and different test methods.
This problem is exactly the same on FRAMs.
This is simply another physical effect, but they have similar reliability issues, and very similar statistics.
The only advantage - maybe- is they do not wear-off (endurance) so quickly.
But if you pose such a question, you must have the complete statistical information to calculate the reliability of your own application.
From all that said, it's clear that you cannot use one single memory, as you need near 100% safety after 20 years, but the probability that the part dies after 1, or 10 years or 18 years, is quite high.
But as you do not know the important statistical parameters, you can't tell, if one year of refresh is sufficient, as you always have the probability of sudden "early failures"..
So you must use a redundant system, i.e. several memories in parallel, and you must definitely use a memory system with error code / correction.
ECC is always done on every data storage systems, as Hard Disc Drives, DVD, BluRay, Memory Cards, (not on Audio CDs) and redundancy always on HiRel systems.
You also have to use a clever refresh procedure.
Well, and the data structure and interface is also a problem, perhaps you have to migrate your storage solution every 5 years, or so.
Sorry, I've got no recipe for you, but hope to have pointed out the problem of your question.
Search the NASA site for those terms, you will probably find stuff how THEY qualify reliability and solve that problem. Their applications run longer than cars, and cannot be serviced.
Frank