Sputter transport
A magnetron sputter process is a two-scale problem: the erosion of the target is decided by the plasma (ions, some 10 eV to 1 keV, time scale ns to µs), the transport of the sputtered atoms to the substrate by the neutral gas flow (0.1 to 10 Pa, time scale ms to s). PICMC covers both, but not in the same run - the time steps differ by four orders of magnitude. The usual way is therefore a PIC-MC plasma simulation that produces the ion bombardment profile, followed by a DSMC simulation that uses this profile as the sputter source.
This page walks through the DSMC part and shows every step in both input forms. The pieces themselves are described in Particle sources, Particle emission and DSMC / PIC-MC iteration.
The physics is the same in both forms:
- the background gas is a DSMC species with collisions from the species and cross-section database,
- the target emits the film species with a Thompson energy distribution,
- the local emission flux follows an imported profile, usually the time averaged
Arplusabsorption of the preceding plasma run, - the film species sticks on the walls (absorption probability 1), the working gas is reflected.
The three ingredients
1. A case in WORKMODE = 0. The work mode is the same keyword as in the parameter file.
WORKMODE = 0 runs DSMC, so no electric field is solved and no charged species is required.
FIELD plots are skipped with a message in this mode - the relevant output are
number, density, pressure, temperature, velocity for the gas, and absorption plus
meshdata for the film species. meshdata writes the absorbed flux per mesh element (a GMSH
view on the surface mesh), which is the direct input for a thickness profile.
- RVM
- Python
WORKMODE = 0;
DT = 1e-6;
TSIM = 1e-3;
sim = PICMC.Sim(WORKMODE = 0, DT = 1e-6, TSIM = 1e-3, ...)
2. An absorption profile from the plasma run. Put Arplus into the ABSORPTION plot list of
the PIC-MC case, time average the resulting .pos files over the RF periods of interest, and
copy one file next to the case. set_profile reads a GMSH POS view and expects the
coordinates in mesh units, the value in the fourth component of each record:
- RVM
- Python
add_source("Ta", 20, "sccm");
set_profile("absorption_Arplus.pos");
set_emission_thompson("Ta", 8.1, 1.0, 500, 40);
ta_source = (
PICMC.BorderReaction(species = "Ta", reaction_type = "source",
source_value = 20.0, source_unit = "sccm")
.set_profile("absorption_Arplus.pos")
.set_emission_Thompson(species = "Ta", Ub = 8.1, alpha = 1.0,
Eion = 500.0, mion = 40.0)
)
The profile is only a distribution: it is binned onto the volume cells touching the border
and normalized over the source points of that border. The absolute flux therefore comes from
source_value, not from the numbers in the profile file. This is convenient (a profile
normalized to percent works) but also the most common mistake - a profile from a low density
test run then still gives the flux that source_value asks for.
Ub is the surface binding energy of the emitted species (8.1 eV for Ta, 4.7 eV for Si),
Eion and mion are energy and mass of the bombarding ion, here 500 eV Ar+. From
these the code computes the recoil cutoff energy of the Thompson distribution itself, so a
wrong mion silently changes the energy spectrum. This is the recommended form; the older
command set_emission_sputter still works and gives the same distribution, but it assumes the
bombarding ions (Ar, 40 u) and a cutoff of 100 times the binding energy instead of taking the
real bombardment into account.
3. Reactions for the DSMC. In the parameter file the elastic cross sections come from the
species definition file SCR/species_definition.r (SPECIESFILE selects another one), in
Python from the pydatabase:
sim = PICMC.Sim(Specs = pydatabase.SPECIES_LIST,
Reactions = pydatabase.REACTION_LIST)
Leave the reaction data out and the sputtered atoms fly collisionless, whatever the pressure says.
Order of the reactions on a border
The profile and the emission profile do not belong to the border, they belong to one wall
reaction of it - the one that was added last. Both input forms attach set_profile and
set_emission_* to the most recently declared reaction, so the list has to be ordered:
- RVM
- Python
# film species sticks, the source reaction comes last
add_absorption("Ta", 1);
add_source("Ta", 20, "sccm");
set_profile("absorption_Arplus.pos");
set_emission_thompson("Ta", 8.1, 1.0, 500, 40);
PICMC.Border(
icodec = 1,
type = "wall",
reactions = [
PICMC.BorderReaction(species = "Ta", absorption = True, probability = 1.0),
ta_source, # source reaction last - the profile attaches to it
]
)
Reversing the two puts the emission parameters on the absorption reaction, where they are accepted without complaint but never used, and the target then emits a Maxwellian at wall temperature.
Boundary conditions of the bounding box
For a deposition profile one wants a defined background pressure. Two settings are sensible:
BOUNDARY_CONDITIONS | behaviour |
|---|---|
all "specular" | closed box, the initial filling P0 is kept, static gas at rest, no pumping. Cheap and consistent with the homogeneous background a PIC-MC case assumes. |
all "outlet" | every particle leaving the box is deleted. Combined with gas inlets this gives a flow field, but the box sides then act as a very large pump: a few 100 sccm in a 0.1 m³ volume settle at 1e-3 Pa, not at the process pressure. |
The second variant only makes sense with a pump surface of realistic conductance or with gas
inlets sized for the pressure that is wanted. For the film thickness this is not a detail - at
1e-3 Pa the mean free path exceeds the target-substrate distance and the profile is purely
geometric. In Python the setting is Geometry(BOUNDARY_CONDITIONS = "specular"), per side as a
dict of the six boundaries.
A worked case
AP4/DSMC_TA_C32.py in the EOSS600 project ports the old RVM matrix case
(Si-5e-3-Plasma.par) to the FHR process module C32 with a Ta target. The mapping:
| RVM matrix case | Python case |
|---|---|
WORKMODE = 0, DT = 1e-6, TSIM = 5.0 | same mode and step, TSIM = 0.1 (the Ta inventory is at steady state after ~1 ms, the old run needed ~79 h) |
SPECIES = ["Ar", "Si"] | {"Ar": …, "Ta": …} |
NREAL_Ar = 6e12, P0_Ar = 0.6 | adapted to the process pressure 0.35 Pa and the cell size |
add_source("Si", 20, "sccm") | BorderReaction(reaction_type = "source", source_value = 20, source_unit = "sccm") |
set_emission_thompson("Si", 4.7, 1, 500, 40) | set_emission_Thompson(species = "Ta", Ub = 8.1, alpha = 1.0, Eion = 500, mion = 40) |
set_profile("…-T1-t102.5-t250.pos") per target | one profile containing both targets, each border picks its own cells |
abs=1 on all walls | BorderReaction(absorption = True, probability = 1.0) |
The grid is chosen from the mean free path, not from the plasma: 4.24 mm cells at 0.35 Pa
(λ ≈ 19 mm for argon) give 40 x 120 x 204 cells in 288 segments; NREAL_Ar = 3e11 puts about
20 simulation particles into each cell. Both numbers are far coarser than in a PIC-MC case.
Submit it like any Python case:
python3 $PICMCPATH/py/pysubmit.py DSMC_TA_C32.py -n 48 -c IST_EPYC
with -n below the number of segments (288 here); additional segments are distributed over the
available ranks. The parameter file case starts the same way through rvmslurm, with the .par
file instead of the script.
What to look at afterwards
absorptionandmeshdatafor the film species: flux and thickness profile.meshdatais a surface view, so it can be opened in GMSH on the same mesh as the case.pressureof the working gas: it should stay nearP0in a closed box. A pressure that keeps dropping means the box is a pump, not a chamber wall.- The particle counters in
<case>-out.txt: the number of sputtered atoms per second should agree withsource_value(1 sccm ≈ 4.48e17 particles/s per species). - A histogram array on a plane between target and substrate gives the angular and energy distribution the raytracer works with.