The Impact of Random Dopant Aggregation in Source and Drain on the Performance of Ballistic DG Nano-MOSFETs: A NEGF Study

We have studied the effect of deviations from doping uniformity in source and drain on the performance of sub-20 nm silicon double-gate MOSFET devices. The study is motivated by recent direct observation of such inhomogeneities in small MOSFET devices. The model assumes a continuous charge distribution with randomly located regions of high doping. Individual dopants are not resolved. Impurity scattering in the source and drain regions is neglected. The current-voltage characteristics are computed for devices that differ only in the microscopic arrangement of the randomly distributed high doping regions. We have also studied devices that differ in the number of these clusters. An ensemble of devices with fluctuating numbers of clusters has been constructed based on a Gaussian distribution centered on the mean cluster number derived from the average doping concentration in the source/drain. The impact of high doping clusters, particularly on charge injection, has been studied in detail for the first time using full quantum simulation based on the 2D nonequilibrium Green function approach coupled self-consistently to Poisson's equation. We have assumed an average lateral symmetry for the double-gate MOSFET. Therefore, we only consider inhomogeneities in doping in the plane of the simulation. The devices studied have 12 nm channel lengths which allows the approximation of ballistic quantum transport through the self-consistent electrostatic potential landscape of the device. The computed current-voltage characteristics are monotonic and they are most strongly affected by the fluctuations in the total number of high doping regions. Fluctuations in the spatial configuration of clusters have a lesser effect. Threshold voltages shifts around 100 mV and on-current fluctuations around 50% are obtained from the simulations. However, the subthreshold slope remains almost independent of the microscopic cluster distribution. The on-current variations are considerably larger than those predicted from classical modeling.