Why Thin-Film Coating Thickness Varies Across Substrate Batches
Two runs, same target material, same recipe, same nominal deposition time — and yet the film thickness map on batch two doesn't match batch one. This is a familiar frustration in thin film deposition labs, and it's usually treated as a tooling or calibration issue. More often, it's neither. Thickness is the output of several time-dependent variables — deposition rate, substrate temperature field, source condition — that a fixed recipe assumes are constant within a run and rarely verifies between runs. Whether the process is Physical Vapor Deposition or thermal evaporation, the variation isn't noise; it's the process telling you something changed.
Three mechanisms account for most of what shows up as unexplained batch drift: deposition rate decay, substrate temperature non-uniformity, and target or source degradation. Each is manageable on its own in a well-controlled vacuum deposition process. Left unmonitored, they compound.
Deposition Rate Is Not a Constant — It's a Decaying Function of Run History
In sputtering coaters — magnetron sputtering systems in particular — target erosion doesn't happen uniformly across the target surface. Racetrack formation beneath the magnetron concentrates erosion in an annular zone, and as that zone deepens, sputtering yield, target density at the erosion face, and the angular distribution of ejected material all shift. The rate at hour one of a target's service life is not the rate at hour two hundred, even under identical power and sputtering pressure setpoints. Conventional direct current sputtering and ion-beam sputtering are both susceptible to this, though the erosion geometry differs between them.
Evaporation sources drift for related but distinct reasons. In thermal evaporation via thermal evaporation coaters, melt pool geometry and emission angle change as source material depletes, and e-beam gun filament aging introduces beam current instability that a fixed power setpoint won't correct for.
Chamber conditioning adds a second, easily overlooked layer. Base pressure creep, outgassing from chamber walls and fixtures between runs, and cryopump saturation mean two runs on nominally identical high vacuum chambers can start from different vacuum states. Reviewing vacuum system selection considerations is useful context here — the vacuum regime a system is built around shapes how sensitive it is to this kind of run-to-run pressure variability.
"A fixed recipe controls inputs, not outputs — thickness drift is what happens in the gap between the two."
Substrate Temperature Non-Uniformity and Its Effect on Growth Mode
Set-point temperature and actual substrate surface temperature are not the same thing. Thermal contact resistance between a substrate and its stage or platen varies with clamping method, backside roughness, and substrate flatness — all of which can differ batch to batch, even within a single supplier's tolerance band.
Local temperature gradients shift nucleation density and grain coalescence behavior, moving growth between island-type and layer-type regimes — the same substrate-temperature dependence that the structure zone model has long used to predict grain control and morphology. That shift changes the effective sticking coefficient, which alters the apparent deposition rate even when the incident flux is unchanged. Multi-substrate runs compound this further: position relative to heater elements, target-to-substrate distance across a batch, and radiative loading from neighboring substrates all introduce gradients that a single stage-level thermocouple won't capture — and none of it is corrected by simply extending deposition time to hit a thickness target.
"Substrate temperature is rarely as uniform as the set point implies — and that gap shows up first in nucleation density, long before it shows up in the thickness reading."
This is why thickness variance and film stress or density variance tend to appear together. They share a thermal root cause; treating thickness as an isolated variable misses that.
Target and Source Degradation as a Slow, Cumulative Variable
Reactive sputtering targets are vulnerable to poisoning — nitride and oxide layers forming on the target surface as reactive gases combine with sputtered material — along with redeposition of insulating material and, in more severe cases, cracking or delamination under repeated thermal cycling. A poisoned or heavily redeposited target surface is also a common precursor to arcing incidents, which introduce localized rate spikes and particulate defects that a thickness monitor downstream won't distinguish from ordinary drift. Evaporation sources have their own degradation path: crucible wetting behavior changes over repeated cycles, boat and filament resistance drifts, and shared chambers introduce contamination carryover between dissimilar runs.
None of this is visible on a per-run basis. Degradation is gradual enough that any single batch looks in-spec against its recipe. What actually shows up is a trend across a target's or source's lifecycle — a systematic thickness bias drifting across sequential batches rather than random scatter. The consequences aren't limited to thickness, either: the same degradation that drives rate drift often shows up downstream as film peeling or compromised bonding integrity, since both are sensitive to the same interfacial conditions at the point of deposition. Distinguishing a trend from noise matters because the corrective action is different: recalibrating a recipe fixes noise, but it won't fix a target that's three-quarters through its service life. For related failure patterns worth ruling out at the same time, see our breakdown of coating defect mechanisms, since target condition and defect density often share the same root cause.
Why Recipe-Based Process Control Fails to Catch This
Fixed-time or fixed-power recipes are built on the assumption that the rate is stable within a run. They control inputs — power, pressure, time — not the output that actually matters.
Quartz crystal microbalance monitoring is the common first line of defense, but it has its own failure modes: crystal aging changes the tooling factor over time, and crystal placement is rarely representative of the actual substrate position, particularly in multi-substrate or rotating-stage configurations. Optical alternatives — in-situ ellipsometry, interferometric endpoint detection — offer tighter feedback loops but add cost and complexity, and aren't universally applicable across every film and substrate combination.
Process variant matters here, too. RF sputtering, pulsed DC reactive sputtering, and high power impulse magnetron sputtering (HiPIMS) each exhibit different sensitivity to the drift mechanisms above — pulsed and HiPIMS regimes, for instance, are often adopted specifically because they reduce arcing and improve plasma stability during reactive processes, which indirectly stabilizes the rate. Some platforms address rate drift more directly by modulating power against a live rate-feedback signal rather than holding power fixed and hoping rate follows — a closer match to the actual problem than a static recipe.
"Target and source degradation rarely announce themselves in a single run. They surface as a trend across a target's lifetime, not a fault in any one batch."
The point isn't that one monitoring method is universally correct — it's that a recipe alone provides no feedback on any of the three mechanisms above. Something has to close that loop.
Practical Implications for Batch Consistency
What's worth tracking between runs is different from what most recipes log. Target and source usage hours, chamber base pressure trend over time, and substrate fixture thermal history all carry more diagnostic value than the recipe parameters themselves.
Post-process metrology — profilometry, ellipsometry, XRF thickness mapping — still has a role, but as a loop-closing check rather than a first-line control method. Labs without in-house capacity for this level of diagnostic tracking can rely on outsourced thermal performance testing to close the loop without adding equipment overhead.
Thickness Variance Is Diagnostic Information
Batch-to-batch thickness variation isn't a defect to be averaged out or tolerated within a wider spec window — it's information about where the process is drifting. Rate decay, thermal non-uniformity, and source degradation each leave a distinct signature, and distinguishing between them is what turns a recurring thickness problem into a solvable one.
For labs building out or reconfiguring deposition workflows, MSE Supplies supports the equipment side of this problem directly. If your process calls for a non-standard configuration, our custom laboratory equipment solutions are built for exactly that. To talk through your specific setup, contact us directly, or connect with our team on LinkedIn.