August 17, 1998

High-temperature testing exposes failure modes

Ron Mancini

At 8:30 pm, instead of being home with the family, I was running a problematical motor drive at 100ºC ambient temperature. The engineering vice president, Buzz, came over to review my testing (and to assert his authority). He said I was foolish to run a 55ºC-rated motor drive at 100ºC because I would destroy the motor drive without gaining any constructive data.

This incident, like others discussed in my columns, happened long before I joined Harris Semiconductor; thus, I hadn't yet gathered concrete data to substantiate my position. I told Buzz that there was a problem with the motor drives, though they passed all the standard tests, so I had to do something different to find the problem. I wanted to increase the ambient temperature until failures occurred and then investigate each failure to determine whether it was related to the field problem. Buzz did not like this approach. He told me that the reason you induce failures at elevated temperatures is because parameter drift causes spec failures, which are normal. Then you waste time troubleshooting these foolish failures, according to Buzz. What is a foolish failure? I guess it's any failure that you can predict will occur at a specific elevated temperature, such as plastic melting at its melting point.

Buzz had a point, but what options did I have? Should I test a hundred motor drives and wait for one to fail? Should I wait for a field failure to find its way back to me? No. According to statistical theory, a typical motor drive designed to work at 55ºC must work at elevated temperatures because the worst-case parameters don't exist. My gut feeling said that the typical motor drive would function at elevated temperatures well beyond the design spec unless Buzz's foolish failure occurred.

I formulated a plan to test motor drives at elevated temperatures, troubleshoot the failures, continue testing by ignoring or eliminating foolish failures, find the cause for each valid failure, eliminate the cause if justified, and find a new job if the plan failed. Buzz bought the plan, but he laid down some rules: I was to report to him daily, keep logs, stop testing at 120ºC (arbitrary limit), return the motor drives to inventory at the time of test completion to keep costs low, and find a new job or eat humble pie if I proved myself wrong.

The test revealed three easily eliminated failure modes: poor die attachment on one vendor's power transistors, varnish breakdown in the coupling transformer, and circuit-board solder residue that was eating copper traces. Buzz approved elevated-temperature testing based on these results. But his money-sensitive heart was disappointed; we could not restock the test units because they looked and smelled worn out.

Years of experience have taught me to eliminate failures before shipping hardware by employing elevated-temperature testing during design characterization. Also, testing several groups, with each group at different elevated temperatures, yields an excellent prediction of when the average unit will fail in the field. Elevated-temperature testing cannot predict failures resulting from external effects, such as lightning, shock, or vibration; it predicts only temperature-related failures.

My next column discusses test techniques designed to find different unexpected failure modes.

| EDN Access | Feedback | Table of Contents |