Reliability of ultra-thin dielectrics for giga scale silicon technologies

In this paper new insights in the relationship between the statistics and the mechanisms of intrinsic oxide breakdown for ultra thin oxides between 12 nm and 2 nm are discussed. It is shown that the statistical spread of the intrinsic TDDB-distribution, which is quantified by the slope of the Weibull distribution β, depends on the oxide thickness. A large increase in the spread on the distribution is found for decreasing oxide thicknesses. This increased spread is not determined by larger statistical variations of process parameters, such as e.g. oxide thickness uniformity, but by the nature of the breakdown mechanism itself. A statistical percolation model is presented, and it is shown that this model can explain the experimental statistical features of the breakdown distributions, which were not yet understood before, Next, it is demonstrated that for ultra-thin oxides the conventional interpretation of constant current Q_BD to evaluate tile influence of process variations on the reliability of MOS-structures can lead to erroneous conclusions. For a fixed thickness, the comparison of Q_BD. Distributions of all processes that affect the breakdown statistics becomes even meaningless. For ultra-thin oxides, the impact of different processing conditions therefore requires constant gate voltage instead of constant current density Q_BD tests