Cmos reliability: A useful case history to revise extrapolation effectiveness, length and slope of the learning curve

Abstract The present paper draws a case history on the reliability growth of CMOS technology. The aim is to underline by this example the importance of the learning process of a new technology. How long does a new technology take to reach its reliability target value or the values of already consolidated technologies? In which way is it possible to shorten this period of time? How wide is the reliability gap between the first device produced on industrial scale and those produced after some years of learning? When is it right to use the accelerating factors to forecast reliability? In this paper are compared the reliability data obtained from early accelerated life tests with those from massive life test by more than 16,000 devices. All these data are differentiated following the failure analysis results. Thanks to the technological characterization of the product in a continuous way and to the definition of appropriate screening tests, it has been possible to evaluate quantitatively the reliability growth of CMOS technology.