Statistical Analysis of Neutron-Induced Gain Degradation of Power Transistors

Predictions of low neutron fluence failure probabilities of silicon power transistors are of particular interest since this device type is among the most vulnerable in systems operating in nuclear environments. The Weibull distribution can be used for this, but so far attempts to do so have not been too successful. This work attempts to use more statistically significant sample sizes and a more ideal neutron environment to study four representative device types, two NPN and two PNP. An important result is the extreme sensitivity of the shape of the Weibull plot to the definition of failure. Three definitions are used: (1) minimum remaining fractional gain, (2) minimum remaining absolute gain, and (3) minimum transit-time-normalized damage constant. The effects of these and of various screens on the resulting Weibull plots are noted and discussed. With few exceptions, the observed failure distributions do not fit a simple Weibull distribution suitable for extrapolation to low failure rates. Screening effects can vary over a broad range from truncation of a Weibull function to linearizing a nonlinear distribution to the point that extrapolation is made possible. The efficacy of any combination of failure criterion and screen to describe neutron-induced gain distribution is limited in applicability to certain ranges of device type and current.