Predicting coating composition, deposition rate and hysteresis in a reactive PVD process

The Partners

This case study is one of examples of mutual productive co-operation of PlasmaSolve with the Belgian company Innovative Coating Solutions (ICS). ICS is a company that develops custom processes and assists industry leaders in integrating these processes into their coating equipment. To do so, ICS is operating several PVD and PECVD coaters.​

The project presented mutual benefit – ICS gained deeper understanding of their drum coater (coating uniformity, variations in coating composition) while PlasmaSolve obtained high-quality validation data for their MatSight model of reactive sputtering.

Problem Specification

The project had the following goals on the ICS and PlasmaSolve side:

1. Predict atomic composition of a nitride coating prepared by reactive magnetron sputtering using PlasmaSolve’s MatSight software
2. Validate the accuracy of spatially-resolved deposition rate predictions from PlasmaSolve’s MatSight software
3. Confirm whether MatSight predicts the correct hysteresis curve of the reactive sputtering process.
4. Proof-of-concept coupling of PlasmaSolve’s MatSIght to ICS’s Virtual Coater software and perform a combined simulation of plasma process as well as film growth

All these goals were ultimately reached and the results are a good demonstration of both PlasmaSolve’s and ICS’s joint capabilities.

Coater specification

The coating chamber – a drum coater – consists of two planar targets (red colour) connected to DC power supplies (2.2 kW at each target). Both targets are made of titanium/aluminium alloy in the ratio of 1:1. The process gas (argon and nitrogen) is fed into the chamber via 4 showerheads (orange), located along side each target.​​

The vacuum pump (purple) is located on top of the chamber, separated from the coating zone by a top plate. The silicon substrates were placed all over two (out of total six) dummy cylinders (various colours on the orange and purple cylinder).

Model performance in terms of coating composition – left part shows 8 training experiments, right part shows 4 validation experiments.

Simulation Strategy

The MatSight Reactive Sputtering App is actually a self-consistent model coupling 3 distinct physics sub-models

1. gas flow (3D, particle based DSMC method),
2. metal sputtering and transport (3D, DSMC),
3. global plasma model (solved for each cathode separately).

Each of these models has a different characteristic time scale. To reach a physically meaningful and self-consistent solutions, the three sub-models have to be coupled in a rather ingenious manner.

3D Gas flow model

The simulation describes reactive gas flow in the coater – the gases (Argon, Nitrogen) are injected at the inlet, pumped by the pump but additionally, nitrogen can be consumed at the targets, substrates and chamber walls. The consumption rate of Nitrogen does, however, depend on the 3D-spatially resolved flux of deposited metal vapors …

3D sputtering model

… the distribution and deposition of metal vapors is described by a different particle based model, which solves the partially ballistic, partially diffusive transport of metal atoms between the target and the rest of the coater. In this simulation, the sputtering yield of each metal depends on the state of poisoning of each target …

Plasma model

… and the poisoning or de-posioning of each target in the coater (note that each coater is posioned differently) is addressed by global plasma models. The global plasma models need inputs especially from 3D gas flow (instantaneous nitrogen pressure), which closes the coupling loop.

Hysteresis curve prediction

First verification measure for a reactive sputtering simulation is the hysteresis curve. The hysteresis curve is a characteristic “fingerprint” of each process and if the model reproduces it, it is most likely correct.​

Our model is complex enough to distinguish between the metallic and poisoned branches. As in real process, this only depends on the initial conditions: if we start with no nitrogen in the chamber, we end up in the metallic part, if we fill the chamber with nitrogen we simulate the poisoned part.​

In the chart, each simulated point corresponds to a 3D ab-initio simulation. We are happy to report that the simulated hysteresis curve is very close to the experimental one.

Simulating the coating composition

The ultimate verification of the model is the coating itself. Therefore, we distributed 9 probe samples over the dummy cylinders in the drum coater. The samples were always a pair consisting of a silicon coupon and a quartz crystal microbalance. That way, the coating composition and coating deposition rate can be measured simultaneously.

The rotation of the table was deliberately turned off for the purposes of the validation (one of the cylinders was directly facing a target, second was sideways) and a standard TiAlN deposition process was executed.

The probe samples are designated by a consistent naming convention, see the figures below

• All the samples that were placed on the cylinder in front of one of the targets were prefixed with “SA_” (samples on the other cylinder were prefixed with “SB_”).
• These samples were placed in five horizontal rows, marked with letters T (top row), MT (mid-top), M (middle), MB (mid-bottom), and B (bottom).
• Finally, each sample was also marked with a number corresponding to a column in which it was placed (number 8 marks column facing centre of the target).

As the figure below illustrates, the deposition rate predicted by the model matches almost exactly the measured one on all the probe samples and the variation in deposition rate between individual samples (i.e. different positions in the coater) is captured very well. What is even more remarkable is the quantitative accuracy – the mass deposition rate matches the experiment without the need to introduce a fitting constant or a scaling factor. This confirms that MatSight’s internal sputtering yield database of sputtering yields is accurate for TiAlN.​

The atomic composition of the coating is also captured accurately and the model even captured Aluminum under-stroichiometry on sample SA-M6 and SB-M3. The model only starts to deviate from the experimental measurement at probe samples with very low coating thickness – this can be improved by increasing the number of simulation particles in the model.

Measured (blue bars) and simulated (red+orange bars) absolute deposition rate of TiAlN. The mass deposition rate relates to thickness deposition rate through coating density. The percentage at each bar designates the atomic content of Al in the coating obtained from the model. For two samples, XPS measured composition was measured.

We did a step further and performed atomistic scale simulations of the film growth using the kinetic Monte-Carlo simulation tool NASCAM™ (NanoSCale Modeling) of the software Virtual Coater™ developed by UNamur and sold by Innovative Coating Solutions (ICS). Such simulations were done according to velocity distribution function computed from the MatSight sputtering simulation at individual samples’ locations. We were able to identify such properties as film thickness or columns angle.

Kinetic Monte-Carlo simulation of film growth on two probe samples compared with SEM cross-sections of the coating.

Film growth simulation with NASCAM™

We did a step further and performed atomistic scale simulations of the film growth using the kinetic Monte-Carlo simulation tool NASCAM™ (NanoSCale Modeling) of the software Virtual Coater™ developed by UNamur and sold by Innovative Coating Solutions (ICS). Such simulations were done according to velocity distribution function computed from the MatSight sputtering simulation at individual samples’ locations. We were able to identify such properties as film thickness or columns angle.

Moreover, we were able to predict coating properties as the chemical composition, the effective thermal conductivity or the porosity (pores occluded or connected to the surface).

These simulation results were benchmarked with experimental ones like relative thickness or deposition rate (profilometry), chemical composition (XPS), surface and cross section morphology (SEM) and porosity (SEM).

This study were extended to more complex objects: namely the sharp edge of cutting tools. It allowed us to highlight the influence on the coating quality of the position and the orientation of the object in the coating chamber.