Improved nearest-grid-point and cloud-in-cell particle-mesh schemes are suggested, and a new nearest-element-center scheme proposed, to help reduce self force and improve the spatial accuracy of forces in Monte Carlo semiconductor device simulation. These schemes are exercised on both one-and two-dimensional model problems. An attempt to design a scheme with reduced self force for unstructured triangular meshes is unsuccessful.

This paper describes the two-dimensional heterojunction device simulator MONTE. Drift-diffusion forms the basis for the transport of electrons and holes. Finite differences and Gummel's algorithm have been adopted to solve the coupled equations. Internal electrodes, capping dielectrics, recessed gates, Fermi-level pinning at the device surface, and deep traps in the substrate can all be treated. A heuristic approach has been devised for evaluating the mobility of electrons based on the quasi-Fermi level. Artifacts such as carrier heating in the built-in field are eliminated, but the model does underestimate electron mobility wherever overshoot occurs. Examples of simulation results on Fermi-level pinning, heterojunction Schottky diodes, a recessed gate MODFET, and a GaAs gate heterojunction FET will be presented.

Techniques for ascertaining the small-signal behavior of semiconductor devices in the context of numerical device simulation are discussed. Three standard approaches to this problem will be compared: (i) transient excitation followed by Fourier decomposition, (ii) incremental charge partitioning, and (iii) sinusoidal steady-state analysis. Sinusoidal steady-state analysis is shown to be the superior approach by providing accurate, rigorously correct results with reasonable computational cost and programming commitment.

A method for incorporating impact ionization into a general control-volume formulation of semiconductor transport is described. The methods for electric-field and current-density vector evaluation and generated charge partitioning within a two-dimensional triangular element are given. The techniques employed allow device breakdown to be accurately determined independent of mesh orientation to current flow direction. The avalanche breakdown of a 1-μm n-MOSFET illustrates the approach. Regions where the drain current is a multivalued function of drain voltage are directly and self-consistently calculated for this device.