Surface chemistry as data for level-set simulation of etching and deposition

Context and methodology

This dataset accompanies the article "Surface chemistry as data for level-set simulation of etching and deposition" by Lado Filipovic, Tobias Reiter and Philipp Haslhofer, submitted to npj Computational Materials. The work belongs to feature-scale process simulation for semiconductor fabrication and was carried out at the Christian Doppler Laboratory for Multi-Scale Process Modeling of Semiconductor Devices and Sensors, Institute for Microelectronics, TU Wien.

In the article, the surface reactions of an etching or deposition process are written as a reaction file. ViennaChem compiles this file into data that the surface model of the level-set simulator ViennaPS evaluates at every surface point with ray-traced fluxes. Eight chemistries are simulated with the two example programs of ViennaPS. They are polysilicon deposition from silane, selective epitaxy on a Si/SiGe superlattice, diamond growth, metal-organic chemical vapour deposition of GaAs, argon sputtering, SF6/O2 etching of silicon, fluorocarbon etching of SiO2 with polymer deposition, and atomic layer deposition of Al2O3 from trimethylaluminium and water. The dataset contains the code, the reaction files and the stored output of every simulation. It also contains the scripts that produce every figure and computed number of the article and its Supplementary Information from that output.

Technical details

The record holds a README and four zip files. Unpacked into one folder, the zip files recreate the layout in which the scripts were run.

• ViennaPS.zip contains ViennaPS at commit 1a2dabde of branch cpc-deposit. It includes the files needed to build the surface model and its example programs, and the reaction files with their compiled data.

• ViennaChem.zip contains ViennaChem at commit 6d64b40, with the reaction-file compiler and the Python evaluator.

• reproduce.zip contains papers/reproduce. It holds run_simulations.sh, which runs every simulation, the Python scripts that produce the figures and numbers, the cached figure data, the text files with the quoted numbers, and REPRODUCE.md with step-by-step instructions.

• runs.zip contains papers/runs, with the output of every simulation.

Each run is stored under a stem made of the case and its variation. For example, pol_25 is the fluorocarbon etch at a CF2 flux of 25 × 10^15 cm^-2 s^-1, and h_y050 is the SF6/O2 hole at an oxygen fraction of 0.50. The suffixes _r2 and _r3 mark repeated runs. For each run, the surfaces at the start, at intermediate times and at the end are VTK XML PolyData files, _initial.vtp, _stepN.vtp and _final.vtp, respectively. The console output is in a .log file. For the runs drawn as figure panels, the final volume mesh is a VTK UnstructuredGrid file (_final_volume.vtu). The settings of every run are given in run_simulations.sh.

The surfaces can be opened in ParaView or any VTK reader. Building ViennaPS requires CMake, a C++20 compiler, VTK 9 and Embree 4, and CUDA for the optional GPU flux engine. CMake fetches the other dependencies in their pinned versions. The scripts require Python 3 with NumPy, Matplotlib, PyYAML and SymPy. The README lists the versions used for the article.

Further details

ViennaPS and ViennaChem are developed at github.com/ViennaTools/ViennaPS and github.com/ViennaTools/ViennaChem. The ray tracer draws new random numbers in every run, so repeated simulations differ within the ranges stated in the article. Values that depend on the ray tracing are quoted as means of three runs where the article says so. With the stored runs, every figure can be redrawn without repeating the simulations. The only exception is the verification figure, which also needs the compiled example program. Repeating all simulations takes several hours on a laptop GPU. ViennaPS is licensed under the GNU General Public License, version 3, and ViennaChem under version 3 or later. The scripts in papers/reproduce are licensed under the GNU General Public License, version 3 or later, and the data, which are the contents of papers/runs together with the figure data, results, logs and REPRODUCE.md in papers/reproduce, under CC BY 4.0. The README and the LICENSE.md files in papers/reproduce and papers/runs list which licence applies to which files.