Chemical Vapor Deposition¶

Caveats¶

Simulating surface-atmosphere interactions typically involves quite large computations due to the following reasons.

Model Size

The employed periodic surface models have to feature large unit cells to account for the amorphous structural motifs that result from the deposition of nanometer sized films. In the case of more energetic particles (plasma, sputtering, etc.) impacting the surface, the slab model of the latter needs to be have a certain thickness to account for particle penetration phenomena.

Time Scale

To model reactions on a surface the time resolution of the simulation needs to be short enough to account for the highest vibrational modes of the system. This time scale is often orders of magnitude smaller than the average interval between successive particle impacts around any given site of the model. When studying atmospheres with low pressures such as those occuring in typical high vacuum surface science experiments, this discrepancy in time scales increases further. Furthermore, surface reactions may involve several intermediate or diffusion steps and require therefore significant simulation times too.

Pressure

Even with the accelerated dynamics simulations over millions of steps presented in this tutorial, the effective pressures achieved during the simulation are much higher than in reality. While being a compromise, this approach can be justified for cases where the chemical evolution at any given site occurs fast enough to not be affected by successive particle impacts.

Temperature

The local temperature of the slab model might be raised significantly by either a too strong flux of particles, or even by some acceleration methods. The simulation needs thus to be monitored for temperature spikes and adjusted in order to avoid unrealistic side reactions.

Evolution

The quality with which a model is described during its simulation is crucial to study a realistic evolution of the system. While the model description of such long simulations is essentially limited to empirical force fields or machine learning potentials, it is advised to carefully parametrize such methods for relevant reaction steps prior to begin any production simulations. While the example discussed below simulates realistic chemical processes with an existing ReaxFF parametrization, this cannot be assumed in the general case. Especially, relative discrepancies in reaction barrier heights can affect the chemical processes occurring during the simulation and significantly alter the evolution of the system. It is therefore the responsibility of the user to use adequately validated model descriptions for the simulations in interest. The ParAMS tutorials and documentation provide further information on how to obtain such parameterizations.

Model¶

The model used in the CVD simulations is sketched below and features the following components

Mechanical Properties

Our model mimics a surface with an underlying bulk, and should therefore have a similar mechanical resistance to particle impacts. This is achieved by fixing all atoms in a mechanical support layer at the bottom of the slab model.

Thermal Properties

The model should also approach the thermal properties of the underlying bulk material while still allowing for local hotspots around particle impact sites. This is implemented by adding a thermostat module affecting only atoms in another layer deep below the surface, directly above the mechanical support.

Molecule Gun

Particles are (formally) inserted in a random position within a zone located several nanometers above the surface and with a velocity vector pointing downwards.

Particle Translation

To save simulation steps during the particle flight time, the particle is translated close to the first impact point in its flight path. It is then actually placed into the simulation box near this impact point.

Accelerated Dynamics

In regular intervals timed to start directly after each initial particle contact with the surface, a sequence of accelerated dynamics steps using the force-bias Monte Carlo method commences.

Molecule Sink

If a particle moves away from the surface, it will likely not interact with any other parts of the model either. A safe-box is defined encompassing a large region around the entire surface model. If any stray particle leaves the safe box it is automatically removed by the molecule sink module.

Oxidized Ge(001) Surface¶

The slab model was obtained from a crystal Ge structure, whose lattice parameters were optimized with the ReaxFF parametrization CHOFeAlNiCuSCrSiGe.ff used in all subsequent simulations. After converting the crystal to a surface slab model, the oxygen centers were added to the surface and relaxed in the following. We refer to the crystals & surfaces tutorial for details on building such slab models and provide the initial structure of the partially oxidized Ge(001) surface as an xyz-file for download.

Regions¶

With the initial model imported, we begin by setting up the mechanical support and thermostat layers in the model:

1. Switch to a lateral view of the model (press Ctrl + 1 or Ctrl + 2 )

2. Select the lowest (001)-layer of atoms (see Figure below)

3. Change to the Model → Regions panel, click and name the new region as fixed

4. Repeat the above steps for the second lowest (001)-layer and name the region thermo

5. Switch to engine and then change to Model → Geometry Constraints and PES Scan, select one atom from the fixed region

6. Click on fixed (fixed position)

Equilibration¶

We can now set up a short simulation to equilibrate the system to the temperature used in the following calculations.

1. Under engine and select Force Field: CHOFeAlNiCuSCrSiGe.ff

2. Set Task: Molecular Dynamics and on

3. Set Number of steps: 100000, Sample Frequency: 1000, Initial Temperature 500 K (optional: Checkpoint Frequency 100000)

4. Change to the thermostat details by clicking on the next to it

5. Click on to add a thermostat instance

6. Set Thermostat: NHC, Temperature(s) 500 K, Damping constant: 250 fs, Atoms in region: thermo

While the above settings suffice to equilibrate the system in principle, we would also like to include the effect of the accelerated dynamics phases used in the main simulation. Indeed, the employed force bias Monte Carlo method is known to accelerate the evolution of a system by violating the reversibility of larger Monte Carlo steps. For this reason, we activate the fbMC module already for the equilibration run:

7. Change to Model → MD… → Force Bias MC (fbMC)

8. Click on to add a new fbMC instance

9. Set Frequency: 12000, Number of steps: 3000, Step length: 0.06 Å, Temperature: 450 K, Enable molecular moves Yes, Translation step length: 0.05 Å, Rotation step angle: 0.05 Radian

Simulation¶

Using the equilibrated positions and velocities allows now to set up the production simulation. We load

Load results directlyImport settings and coordinates from rkf

after the above equilibration run is complete

1. (Re)open the equilibration calculation in AMSinput

2. In the popup window, enable Use MD velocities from results for AMS MD restart

3. Click on Yes, new job

4. File → Save as… and save the input under a new name

Warning

Make sure that the original job and the ams.rkf result file is not overwritten

After either of these steps, we can start setting up the final simulation run. We begin with adapting the settings for MD and fbMC

Note that the above Start Step settings are used to align the start of the fbMC intervals with the molecule gun shots.

We proceed by loading the structures for the O₂ and AlH₃ projectiles. These are provided in the files O2.xyz and AlH3.xyz for download.

We can now use these imported regions as O₂ and AlH₃ projectiles in the molecule gun module.

1. Switch to Model → MD… → Molecule Gun

2. Click to add a new projectile

3. Select System: AlH3

4. Set Mole Fraction: 2, Frequency: 12000, Start Step: 1, Stop Step: 12500000

5. Set Fractional coords box: to 0, 1, 0, 1, 45, and 49

6. Enable Rotate

7. Velocity dummies: click , and select the top dummy atom followed by the lower one

8. Velocity direction: click

9. Set Energy: 0.05 eV, Atom Temperature: 450 K, Contact distance: 3 Å

To add the O₂ projectile

10. Click the topmost to add a new projectile

11. Select System: O2 and repeat the above steps 4.-9. (reuse the previously added dummy atoms)

12. Change to Mole Fraction: 4 for the O₂ projectile.

As a last step, we enable the molecule sink to remove any stray particles from the model.

1. Switch to Model → MD… → Molecule Sink

2. Click next to “Remove molecules”

3. Set Formula: * to remove all stray particles

4. Set Safe box to 0, 1, 0, 1, -150, and 150

5. Set Frequency: 12000 and Start step: 1

We are now ready to start the simulation using File → Run.

For the readers’ convenience, we provide the input file for this calculation, Ge200O25_O2_AlH3_fbMCCVD.ams, as well as the previous equilibrium simulation result, Ge200O25.rkf. Afterwards, we just need to follow the steps below to manually set up the initial velocities.