jackson6612,
The first thing to understand is the concept of "forever". "Forever" implies "All of Time", and "From now until the end of time". While we can roughly date the beginning of cosmological time as occurring 13.75 billion years ago, predicting the end of time is a far more debated topic. It is safe to assume no man-made device is older than humanity, and it is safe to assume that all computer storage media are man-made. So, what we are left with is man-made storage media that are all less than 100 years old. Most are less than 40 years old.
The scientist and engineers that develop computer storage media obviously do not have time to test the longevity of the media by direct observation (that is, it is not practical to develop an idea, build it and then wait XXX years to test its longevity), so, they use science in an attempt to predict the life span of some percentage of a batch based on their understanding of the physical properties.
This notion is similar to predicting the average age of a population at some point in the future based on the current population and age distribution. The mathematical model uses statistics guided by known factors to generate a prediction. If the model does not take into account a pandemic, and a pandemic occurs, the prediction could be way off. The further into the future the prediction attempts to calculate, the less accurate the prediction is likely to be.
With computer storage the longevity predictions are a bit more straight-forward, but there is still a lot of room for error. If you assume an 8GB flash drive contains 9 * 2^33 cells for storage alone and a fair number of additional cells for addressing, error checking and interfacing, and you consider that most cells share common characteristics with the other cells on the same die then on a single die, if one cell fails in 1 year of use then the failure rate for the cells is 1 cell out of ~100 Billion per year. If you produce 50 million dies using this architecture, and 10 dies exhibit a cell failure in one year, then your failure rate drops to 100B * 50M/10 per year, and the chances of any particular die failing in 1 year is 5 million to 1. An acceptable failure rate might be .1%, so, assuming the failures continue at the rate of 10 per year for our 50 million units, 50M * .001 = 500,000 units. 500,000 units @ 10 per year would suggest a lifetime of 50,000 years, with a failure rate of .1%.
The actual statistics for projecting useful life of a component are more complex and take into account increased failures with age, use and other variables, but that should give you some idea of how the process works. The silicone chip mfg process has incredibly high quality control. There is no other industry in the world that has a failure level as low per unit. The fact that a single chip may contain billions of cells that all function exactly as they are suppose to is a testament to engineering at its best.
To further improve the life span of things like FLASH drives and Magnetic Drives, additional internal circuitry is employed to detect failures and simply avoid those memory blocks. To lower costs, "New" memory devices frequently have "Bad Blocks" or "Bad Sectors" straight from the factory but are sold as "first quality" if the number of bad sectors/blocks are less than a specified percentage of the total capacity. The price for "perfect" storage media is considerably higher than that of "first quality" media. Military, Aero-Space, Medical and other Government Spec devices typically cost orders of magnitude more than consumer quality devices because they require strict testing and documentation of each device.
Anyway, hope that helps a bit.
Fish
