Uniformity Assurance of Vacuum Coating High-Voltage Power Supplies in Anti-Reflective and Decorative Coating
An anti-reflective coating of seven alternating layers, each between twenty and two hundred nanometres, fails on a thickness error of two percent. The power supply driving the sputtering cathode does not control thickness directly, but the stability of its power, the behaviour of its arc response and the shape of its regulation determine whether the deposition rate holds steady long enough for the thickness monitor to close the loop. Uniformity problems that appear as colour shift at the edge of a lens usually trace back to the supply before they trace back to the target.
Arc handling is the first mechanism. A magnetron target with a dielectric layer, an oxide target or a contaminated surface will arc several times per second, and each arc delivers a burst of energy to a small spot. A supply that simply shuts off and restarts leaves a crater and a particle; a supply that detects the arc in under two microseconds, diverts the stored energy through a clamp, and restores power within fifty microseconds leaves a mark small enough that the next layer covers it.
Arc detection thresholds have to be tuned to the process, not to the datasheet. The current derivative during a normal discharge and during an arc differ by less than a factor of ten on some targets, so a threshold set too low trips on the process, and one set too high misses the arc. Characterising both signatures on the actual target, at the actual pressure, is the only reliable approach.
Ripple translates directly into deposition rate oscillation. A supply with one percent ripple at a few kilohertz modulates the sputtering rate, and the thickness monitor, averaging over a finite window, sees a rate that varies with the window alignment. The result is a layer that meets its thickness target on average and fails its optical specification because the index profile is not what the design assumed.
Regulation under load change is the second mechanism. Reactive sputtering of an oxide target moves between metallic and poisoned modes with a factor of three change in impedance, and the transition happens in milliseconds. A supply that responds in tens of milliseconds will follow the transition, but one that responds in hundreds will let the process settle into the wrong mode and stay there. The loop bandwidth relative to the transition time is the design parameter that matters.
Cable and target geometry affect uniformity through the supply more than through the plasma. A long cable from the supply to the cathode adds inductance, and the arc clamp sees a slower current rise, so the clamp threshold has to be adjusted for cable length. Two chambers wired with different cable lengths and the same supply settings will have different arc energies, and one of them will have a particle problem.
Pulse mode changes the uniformity mechanism again. Pulsing the target at a few hundred hertz with a duty cycle of fifty to eighty percent suppresses arcs on reactive targets, but the pulse shape determines the energy per pulse, and the pulse shape is produced by the supply. A pulse with a slow rising edge spends more time in the transition region and produces a different deposition rate than a pulse with a fast edge.
Measurement of the supply's contribution should be separate from measurement of the coating. Deposition rate monitored over an hour with the shutter closed, at constant power, reveals the supply's drift; thickness uniformity measured across a witness wafer reveals the geometry and the plasma. Blaming the supply for a uniformity problem that is caused by target erosion wastes weeks.
Target erosion is a slow uniformity mechanism that the supply can mask. As the racetrack erodes, the plasma distribution changes and the deposition rate becomes non-uniform, and a constant-power supply holds total power constant while the distribution shifts. Voltage and current monitoring, rather than power alone, shows the shift earlier because the impedance changes as the erosion deepens.
Decorative coatings have a different tolerance. A decorative film a few hundred nanometres thick tolerates thickness variation of five percent without visible colour difference, so the supply requirements are looser on ripple and tighter on throughput. Running at higher power to shorten the cycle increases arc energy, and the trade between throughput and particle count is set by the arc handling design.
Stored energy in the output filter is the parameter that ties arc handling to ripple. More capacitance reduces ripple and increases the energy available to an arc. A supply designed for decorative coating at high power typically uses a moderate capacitance with a fast clamp, while a precision optical coater uses lower power with lower capacitance. Choosing by power rating alone gets the trade wrong.
Thermal stability of the measurement chain affects long runs. A deposition cycle of four hours with a power supply whose current sense drifts by one percent over that period produces a rate drift that the thickness monitor corrects, but the correction shifts the whole process and the last layers differ from the first. Temperature compensation on the current sense, or a current sense resistor with a low temperature coefficient, keeps the run consistent.
Shielding and grounding of the cathode cable influence both arc behaviour and uniformity. A cable with a floating shield radiates, and the radiated field couples into the thickness monitor, producing noise that looks like a rate fluctuation. Bonding the shield at the chamber end, with a single connection at the supply end, removes the loop that causes the coupling.
Supply diagnostics should report arc count, arc energy and recovery time. Without these, a drifting particle count cannot be attributed to the target, the process or the supply. A log of arc events over a coating campaign, correlated with particle inspection results, is the evidence that justifies replacing a target or a supply.
Cost trade-offs are real. A supply with fast arc handling and low ripple costs more than a general-purpose unit, but the difference is small compared with the value of a rejected optical batch. In decorative coating, where margins are thinner, the same feature is often justified by particle reduction alone, without any optical argument.
Limits of the supply's influence should be acknowledged. Uniformity across a large substrate depends primarily on target geometry, throw distance and gas distribution; the supply sets the stability of the process, not its shape. A perfectly regulated supply cannot correct a target that has eroded into an asymmetric racetrack.
A workable acceptance test uses the process itself. Run a two-hour deposition at fixed power with the shutter closed and record rate, voltage, current and arc count; then run a full coating cycle and measure uniformity on witness pieces at three radii. The first test characterises the supply, the second characterises the system, and separating them is what makes the results actionable.
Wafer or lens curvature changes the uniformity story in ways the supply cannot fix. A curved substrate sits at varying distance from the target, and the deposition rate varies accordingly; the supply holds power constant while the geometry produces the gradient. Correcting with a shaped mask or a rotating fixture is standard practice, and the supply's role is limited to keeping the rate predictable so the mask design holds.
Multiple cathodes on one chamber introduce a supply interaction that is easy to overlook. Two supplies sharing a common return path can exchange current through the chamber body during arcs, and the arc in one cathode appears as a current transient in the other. Separate returns to a single star point, and a common interlock that pauses both supplies during a severe arc, keep the two processes independent.
Gas pressure and pumping speed determine how quickly an arc's particle burden is removed, and the supply's recovery time sets how much material is deposited during that window. A supply that restores power in fifty microseconds deposits almost nothing during recovery; one that takes ten milliseconds deposits a visible nodule. The recovery time specification should be quoted in the same units as the deposition rate so the comparison is direct.
Consumables and calibration drift both affect the deposited thickness, and the supply should not be blamed for either without evidence. Target thickness decreases with use and the rate rises for constant power; the thickness monitor compensates, and the compensation hides the change until the target is nearly exhausted. Periodic rate-versus-power characterisation, independent of the monitor, separates the two.
Records make the supply's contribution visible over time. Rate at fixed power, ripple, arc count and recovery time, recorded at the start and end of each campaign, form a baseline. When uniformity complaints begin, the baseline shows whether the supply changed; when it did not, attention moves to the target and the gas distribution, which is where the problem usually is.
Reactive gas control and the supply form a loop that can oscillate if the two time constants are close. The gas controller responds in tens of milliseconds and the supply in a few, so the loop is stable; a supply with a slow response and a gas controller with a fast one will hunt, and the deposited film will show alternating composition. Matching the two response times is a commissioning task that is rarely documented.
Tooling condition dominates long-run uniformity. A shield that has accumulated coating changes the plasma impedance and the deposition distribution, and the supply compensates by adjusting voltage while the distribution remains shifted. Cleaning the shield on a schedule derived from the rate trend, rather than from a calendar, keeps both the supply and the film closer to their design points.

