|Re: shelf life time of a ROM, EEPROM, EPROM vs Mask Rom|
Message #3 Posted by Eric Smith on 10 Jan 2013, 2:40 p.m.,
in response to message #2 by Jim Horn
When used at mostly room temperature, typical storage time will be in terms of centuries.
While the typical data retention time at room temperature will certainly be longer than the minimum specification of the part (usually 10 years), I very much doubt that it will be centuries. Even if the leakage rate is low enough that the charge on the floating gate in principle could last that long, there are other failure modes that are likely to cause failures more quickly than that. Not all failure modes are accurately modelled by accelerated ageing tests, so it is very difficult to predict a usable lifetime. This is one reason why most electronic component vendors rate the working lifetime of components at 5 or 10 years; the published MTBF figures are statistical measures of failure likelyhood only within the rated lifetime of a part. In other words, a part with an MTBF of 200,000 hours is NOT actually expected to last 22.8 years. What that MTBF tells you is that for a population of components, operated *within* the rated lifetime, the rate of failures for the population (not for any individual component). The MTBF doesn't tell you *anything* about the expected reliability or failure rate of the component beyond the rated lifetime. At some point past the rated lifetime, the component reaches the far end of the "bathtub curve", at which point the failure rate increased dramatically.
If a floating-gate memory device with a guaranteed data retention spec of 10 years, on average you can probably expect it to last longer than 10 years, but for the reasons above, every year that it continues to operate reliably past the 10 years should be considered a bonus, since there is no expectation of reliability past that time.
Bear in mind also that the data retention based on charge on the floating gate is itself a statistical process. The data can no longer be considered to have been retained as of the earliest time that a single bit in the device no longer reads reliably, which is likely to happen significantly sooner than average bit lifetime due to variations in the physical properties of the individual bit cells, as well as variations in the programming current and time for individual cells.
I worked on the PDP-1 Restoration Project at the Computer History Museum in Mountain View, California. The PDP-1 was made in the early 1960s. While we expect component failures, there is in principle no reason why it shouldn't be possible to maintain the PDP-1, replace failed transistors, and keep it running for the forseeable future. Probably at least another 50 years. (Because of the accelerating pace of change, it's arguably impossible to predict almost *anything* about human civilization beyond that timescale.)
On the other hand, I think it's unlikely that it will be possible to keep any electronic equipment manufactured after the mid-1990s operating for more than 25-50 years, in part due to the floating gate problem (even in places you don't expect it, because *many* chips now contain flash memory even if you don't know about it), and partly due to the general problem that modern ICs have become very specialized and have relatively short production lives.
In other words, 50 years from now we'll probably still have a working PDP-1, but sadly not too many working HP-35 calculators.