Million Atom Electronic Structure and Device Calculations on Peta-Scale Computers

TL;DR

This paper introduces OMEN 3-D, an advanced atomistic modeling tool for semiconductor devices, leveraging peta-scale computing with improved parallelization to simulate complex quantum effects and impurity bands.

Contribution

The paper presents OMEN 3-D, a new computational tool with enhanced parallelism and domain decomposition for large-scale atomistic semiconductor simulations.

Findings

Successfully demonstrated full 3-D parallelized Schrödinger-Poisson solver

Calculated impurity bands in δ-doped phosphorus layers in silicon

Showcased scalability on peta-scale computing systems

Abstract

Semiconductor devices are scaled down to the level which constituent materials are no longer considered continuous. To account for atomistic randomness, surface effects and quantum mechanical effects, an atomistic modeling approach needs to be pursued. The Nanoelectronic Modeling Tool (NEMO 3-D) has satisfied the requirement by including emprical $sp^{3}s^{*}$ and $sp^{3}d^{5}s^{*}$ tight binding models and considering strain to successfully simulate various semiconductor material systems. Computationally, however, NEMO 3-D needs significant improvements to utilize increasing supply of processors. This paper introduces the new modeling tool, OMEN 3-D, and discusses the major computational improvements, the 3-D domain decomposition and the multi-level parallelism. As a featured application, a full 3-D parallelized Schr\"odinger-Poisson solver and its application to calculate the bandstructure of $\delta$ doped phosphorus(P) layer in silicon is demonstrated. Impurity bands due to the donor ion potentials are computed.