Ab initio scattering from random discrete charges and its impact on the intrinsic parameter fluctuations in nano-CMOS devices

摘要

This thesis is concerned with the Monte Carlo simulation of device parameter variation associated with the discrete nature and random variation of ionized impurity atoms within ultra-small conventional n-MOS devices. In particular, the Monte Carlo method is applied to accurately resolve electron interactions with individual ionized impurity atoms and in so doing capture the variation in impurity scattering associated with randomly configured dopant distributions. To date, variation in transport due to position dependent variation in Coulomb scattering has not received any attention although is expected to increase the inherent device parameter variation.A detailed methodology for the accurate treatment of Coulomb scattering within the Ensemble Monte Carlo framework is presented and verified. Improvement over existing methodologies is presented with a short-range force model that significantly reduces errors in conservation of energy during short-range attractive interactions compared with models proposed in similar work. Details of the simulated reproduction of bulk mobility are thoroughly presented to validate the method, while to date such detail is not to be found anywhere in the literature.A charge assignment method is developed to be applied to traditional 'continuously' doped regions in order to allow a consistent description of doping charge when combined with 'atomistic' doping assigned via the Cloud-In-Cell scheme. The charge assignment method also represents the only consistent description of electron charge assigned via CIC and the continuous doping charge.Trapping of a single electron in a series of scaled n-channel MOSFETs was studied with the ab initio Coulomb scattering method and is consistently seen to increase the Random Telegraph Signal, associated with the trapping and de-trapping of such charges, when compared with Drift-Diffusion simulations. It is seen that the electrostatic influence of the trapped charge is most prominent at low applied gate voltages where it accounts for nearly 70 - 80% of the total current reduction when including transport variation in devices with channel lengths of 30- nm. At high gate voltages, transport variation is the dominant factor with the electrostatic impact accounting for only 40 - 60% of the total variation in the same devices.Extending this treatment to an ensemble of atomistic devices, it is seen that the inclusion of transport variations significantly increases the distribution in device parameters and that the transport variation is significantly dependent upon the specific dopant distribution. Within an ensemble of 50 'atomistic' devices, it was seen from Drift-Diffusion simulation that the average current showed a 3.0% increase over the continuously doped device, while Monte Carlo simulations resulted in a decrease in average current of 1.5%. The standard deviation of the current distribution from Drift-Diffusion simulations was 2.4% while, significantly, Monte Carlo simulations returned a value of 6.7%. This has implications for the published data obtained from Drift-Diffusion simulations which will underestimate the variation.