Although Negative Bias Temperature Instability (NBTI) was first observed more than 30 years ago, and the phenomenon has been one of the focus topics of the microelectronics reliability community for some years now, there is still no generally agreed physical explanation for it. Even the interpretation of some empirical features of NBTI is a matter of ongoing debates. Most researchers in the field, however, have in the meantime acknowledged that oxide defects must play a crucial role in the explanation of NBTI.
One particularly interesting observation with NBTI can be made when transistors are subjected to alternating stress and relaxation sequences. Quite surprisingly, the amount of threshold voltage shift that recovers during the relaxation periods decreases with increasing number of cycles, or increasing number of cumulated net stress time. We carried out the according experiments on three different technologies, which show comparable behavior at high temperatures (200°C). For one technology, we also measured where there was a reduced loss of recoverable component at a lower temperature. It must be noted here that the decreasing recoverable component is incompatible with defect models with just two states (charged and neutral), as those defects should show perfectly cyclic behavior. In contrast, a defect model with at least a third state, where the formerly charged defect 'locks in' and hence will not recover in the next relaxation period, is required.
The general use-case of a Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET) is hardly a constant-bias situation, but rather a stochastic process with variable on- and off- times. Regarding NBTI in this situation means that the accurate modeling of the recovery is at least as important as correctly modeling the stress phase. Especially effects such as the described loss of the recoverable component can have a huge impact on lifetime extrapolations, as errors accumulate in the course of the simulation.