Holy Grail of Electronics: Failure Prediction

“There is a fundamental problem in electronic devices today and that is the inability to accurately predict when they will fail. And all electronic devices fail.”

So says UCSB electrical engineer Umesh Mishra, who with scientists around the world is searching for the golden ring of transistor development: predictability of device failure. Funded by the Office of Naval Research MURI (Multidisciplinary University Research Initiative) in 2008, the multi-university team led by Mishra was awarded $4.5 million over three years for his Design-for-Reliability Initiative for Future Technologies (DRIFT) project.

In many aspects of the commercial sector, the failure of electronic devices is not as critical because they are changed out regularly, like computers, Mishra says. Being able to predict failure becomes more important, though, when a device is deployed in space. Typically devices are tested with heat, a process called Arrhenius testing. The device is heated to some level above normal until it fails.

“In DRIFT, we are going to come up with a methodology that tries to predict failures by using techniques which are more statistical rather than purely Arrhenius. That means things can happen for reasons not due to temperature and we are going to try to track them down.”

The research is being conducted from two perspectives: a bottom-up approach in which scientists are looking at the tiniest components in a device: electrons; and a top-down approach where researchers take devices from industry and try to make them fail.

“From a bottom-up approach we have collaborators at the University of Michigan and Vanderbilt University doing things on a very basic level, looking at rogue electrons doing damage of various kinds,” Mishra explains.

“Then we bring the problem from this extremely stochastic world into trying to get from it what are called rate equations—if all these things are happening, on an average can we predict a rate of failure?”

Mishra says the top-down approach will involve studying a device’s response to individual stressors such as heat, electrical field, or mechanical stress and then observing various combinations of those stresses on the device: “Essentially, we just take a device and try to kill it.”

Because UCSB has strength in materials research, he also plans to use different materials in the testing as well. Ultimately, they hope to have a new model for testing.

“Nobody knows how to do it. So we said we’re going to get the best people together. Our team is just an unbelievable team of people—from MIT, Ohio State, Michigan, Vanderbilt, CMU and NC State—and then what we’re going to do is figure things out.”

When they solve the problem, it will be of enormous benefit both to defense agencies, like the Navy, and to the commercial sector. Take an electric motor in a hybrid car, for example.

“What engineers do now is overdesign to ensure reliability,” he says. “If you end up with testing methods which are more accurate, then maybe you will have the confidence of not asking for so much safety margin. And then there will be reduced costs.”