The boost in switching speed and the perpetual reduction of device size are two of the major driving forces of the semiconductor industry. As a side effect, reliability problems have become more and more important, due to the shortening of the channel length, which raises the electric field in the device. Reliability issues like Hot Carrier Stress (HCS) and Negative Bias Temperature Instability (NBTI) become major topics. Although theoretical models have been developed from the very beginning, they still do not satisfy all needs.
Up to now it is not quite certain, judging from the contradicting reports, what kinds of defects are to blame for the NBTI phenomenon. Therefore it is vital to collect measurement data in a systematic manner. Empirical research and data analyses should finally lead to an applicable model. One of the most established models, the reaction-diffusion-model, assumes that some hydrogen species is first released from previously passivated interface defects and then diffuses into the oxide. Unfortunately this model is not capable of explaining the relaxation behavior correctly.
A growing number of recent publications attribute at least a part of the degradation to hole traps. Eventually a combination of these two mechanisms may lead to a better description of the phenomenon.
Although most applications operate in AC mode, due to its simplicity DC-stress has become the common test routine for reliability. Here, one drawback is the missing consideration of the relaxation of the threshold voltage during and after stressing of the device. This is of considerable importance for accurate lifetime prediction.
Factors like the stress voltage (high electric field) and high temperature, as well as time and duty cycle of the bias stress, have an enormous impact on the recovery behavior. An intriguing feature of relaxation is its universality, which is shown in the picture. The exact physical mechanisms remain unresolved at the moment. One of the main tasks is thus the development of a model which is able to anticipate degradation precisely, in particular the long logarithmic tails covering more than 10 decades of time.