Simulation of charge-trapping in nano-scale MOSFETs in the presence of random-dopants-induced variability

摘要

The growing variability of electrical characteristics is a major issue associated with continuous downscaling of contemporary bulk MOSFETs. In addition, the operating conditions brought about by these same scaling trends have pushed MOSFET degradation mechanisms such as Bias Temperature Instability (BTI) to the forefront as a critical reliability threat. This thesis investigates the impact of this ageing phenomena, in conjunction with device variability, on key MOSFET electrical parameters. A three-dimensional drift-diffusion approximation is adopted as the simulation approach in this work, with random dopant fluctuations—the dominant source of statistical variability—included in the simulations. The testbed device is a realistic 35 nm physical gate length n-channel conventional bulk MOSFET. 1000 microscopically different implementations of the transistor are simulated and subjected to charge-trapping at the oxide interface. The statistical simulations reveal relatively rare but very large threshold voltage shifts, with magnitudes over 3 times than that predicted by the conventional theoretical approach. The physical origin of this effect is investigated in terms of the electrostatic influences of the random dopants and trapped charges on the channel electron concentration. Simulations with progressively increased trapped charge densities—emulating the characteristic condition of BTI degradation—result in further variability of the threshold voltage distribution. Weak correlations of the order of 10-2 are found between the pre-degradation threshold voltage and post-degradation threshold voltage shift distributions. The importance of accounting for random dopant fluctuations in the simulations is emphasised in order to obtain qualitative agreement between simulation results and published experimental measurements. Finally, the information gained from these device-level physical simulations is integrated into statistical compact models, making the information available to circuit designers.