When talking about the Mean Time To Failure (inverse or FITS, treatment

or failures per billion hours) the “bathtub” curve comes to mind. This curve represents a probability distribution for device failure, viagra 40mg

where you have a high rate of failure due to infant mortality, page

then a flat probability of random faliures, and finally an increased failure probability due to old age. National Semi has a pretty good overview of experimentally determining this distribution and the tests involved. The curiosity lies in looking at reliability reports, say for the AT49BV020, that show zero failures over a number of device hours but they are able to provide a predicted device failure rate. Since the normal MTTF formula (one involving at least one device failure, near the middle of the National link) would go to infinity, another method is employed. The Chi-squared distribution is used to determine the predicted failure rate for a device with no failures for a given operating hours and a given confidence (typically 60% or 90%.) Finally, ATMEL has a chapter (chapter 6) in their Quality and Reliability Handbook devoted to modeling device failures, especially when accelerating device failure with increased temperature or voltage.

( doc1281.pdf ) ( 49bv020.pdf ) ( calculating_mttf_with_zero_failures.pdf )