USC » DEC » Computer Architecture Group » Semiconductor Device Simulation » Monte Carlo

Ensemble Monte Carlo simulator

Accurate physical modelling of the carrier transport as well as a correct description of complex 3D geometry of novel architectures is needed in order to predict their behaviour and optimise their design. Such physical modelling can be achieved via the ensemble Monte Carlo (MC) method.

In order to study statistical samples and at the same time maintain the predictive power of the simulations, trade-offs must be made between efficiency and accuracy. We have chosen the well know analytic nonparabolic anisotropic model for the dispersion relation in the valleys of the conduction band of silicon which allows very efficient MC simulations while retaining much of its accuracy for expected values of supply voltages. Also, quantum corrections are introduced "frozen", taking the value of the quantum potential from DD simulations and adding it to the self-consistent electrostatic potential updated every 0.05 fs.

To accurately reproduce the geometry of the considered 3D domains the real space discretisation is based on the use of unstructured tetrahedral elements. To minimise the impact of self-forces in the particle-mesh coupling, we have developed specific algorithms for this meshes.

The ensemble Monte Carlo technique is computationally very expensive for 3D geometries and, therefore, the use of efficient algorithms and parallel computing in order to save simulation time is desirable. We have adopted a parallelisation estrategy which mixes domain decomposition for the linear system solver with particle sharing for the free flights as shown in the figure.