真空镀膜高压电源在光学多层膜与太阳能薄膜中的层控

  A multilayer optical coating is a stack of sixty to two hundred layers, each a quarter wavelength thick, and the stack works only if every layer is within a percent of its design thickness. The supply that drives the deposition sets the rate, and the rate sets the thickness.

  Two processes dominate. Thermal evaporation melts a source in a crucible and the vapour deposits on the substrate, with the rate set by the source temperature. Sputtering bombards a target with ions and the rate is set by the target power, which the supply controls directly.
  Evaporation puts the supply at a distance. The electron beam that heats the crucible is driven by a high-voltage supply of a few kilovolts, and the beam current sets the power. A drift of one percent in the beam current is a drift of roughly one percent in the rate, and a stack of a hundred layers integrates that drift into a colour shift visible to the eye.
  Sputtering puts the supply in the loop. The target voltage and current together set the power, and the supply's regulation determines how fast the power responds to the plasma's changes. A magnetron plasma is inherently noisy, and a supply with a slow loop will let the noise modulate the rate.
  Reactive sputtering is the harder case. A metal target sputtered in an oxygen atmosphere forms an oxide on the target surface, and that oxide changes the sputtering yield. The transition between metallic and oxidic modes is sharp, and a supply that cannot hold power through the transition will oscillate between the two modes and produce a layer with a graded composition.
  The standard cure is a supply that regulates power rather than voltage. In power mode, the supply adjusts the voltage as the plasma impedance changes, and the rate stays constant through the transition. In voltage mode, the same transition produces a rate change of a factor of two.
  Arc handling is the second requirement in reactive sputtering. Oxide flakes on the target surface charge, then discharge in a microarc that lasts microseconds and ejects particles into the coating. The supply must detect the arc and cut the power before the arc energy grows.
  Arc detection in this role is done by watching the voltage fall rather than the current rise. A microarc pulls the voltage down sharply, and a comparator that fires within a microsecond can blank the output for a few hundred microseconds, which is enough to extinguish the arc without disturbing the plasma.
  A blanking cycle that is too long extinguishes the plasma and forces a restart, and a restart produces a rate transient. The blanking time is therefore tuned to the target's recovery, typically a few hundred microseconds, and it is a parameter that changes with the target material.
  Thickness control is the third requirement. A quartz crystal monitor measures the mass deposited, and the shutter closes when the thickness is reached. The monitor's accuracy and the supply's stability combine to produce the layer's tolerance, and neither alone is sufficient.
  A rate that is constant to one percent and a monitor that is accurate to half a percent give a layer that is within one and a half percent, which is the practical limit for a production coater. The optical design must be tolerant to that, and a design that demands a tenth of a percent belongs in a different machine.
  Uniformity across the substrate is a separate problem. The plasma density varies across a large target, and the supply cannot correct that; the geometry does. What the supply can do is hold the same power at every point in the process so that the geometry's correction remains valid.
  Solar thin films add a throughput dimension. A coater that runs a web of substrate at metres per minute needs a rate that is stable for hours, and a supply that drifts with temperature will produce a film whose thickness varies along the roll.
  Temperature drift in the supply is dominated by the current sensor. A shunt with a temperature coefficient of a hundred parts per million per degree will change the power by a tenth of a percent per ten degrees, and the rate follows. A compensated shunt or a Hall sensor reduces that by an order of magnitude.
  A coater case illustrates the chain. An optical filter produced a colour shift at one end of a run, and the crystal monitor read correctly. The cause was the target voltage, which rose by two percent over the run as the target eroded, and the supply was in voltage mode, so the power fell and the rate fell with it.
  Switching the supply to power mode removed the shift, because the supply raised the voltage to compensate for the erosion. The lesson is that the process needs a controlled quantity that stays meaningful as the hardware changes, and power is that quantity.
  A second case involved arcs. A reactive oxide process produced particles in the coating at a rate of a few per run. The supply's arc detection threshold was set for a larger arc, and the microarcs passed undetected. Lowering the threshold and shortening the blanking removed the particles and did not disturb the plasma.
  A third case involved a web coater whose film was thinner at the start of each roll. The supply's output had a slow start because its soft-start ramp was set for a larger machine, and the first few metres of the roll were coated at a rising rate. A shorter ramp, matched to the web's acceleration, fixed the start.
  Instrumentation for a coater is modest: the supply's own voltage and current, the crystal monitor's rate, and the substrate temperature. The three are logged together, and a discrepancy between the supply's power and the monitor's rate is the earliest sign of a target that has changed.
  Maintenance follows the measurements. The current sensor is calibrated quarterly, the arc detection is tested monthly with a simulated arc, and the target is inspected for oxide growth at each maintenance window.
  Spares are few: one arc detection board, one current sensor, one set of output capacitors. The supply's power stage is rarely spared, since a failure there is usually a failure of cooling or of the mains, and both are diagnosable.
  Cost should be judged against the layer count. A supply with power regulation and arc handling costs perhaps half again as much as a voltage-mode unit, and a single rejected stack of a hundred layers costs more than the difference, which makes the decision easy in a production environment and difficult in a job shop that runs one stack a week.
  Limits need stating. No supply can compensate for a target whose erosion has changed the geometry, and no arc detection can prevent particles that come from the chamber walls rather than from the target. The supply holds the electrical state, and the chamber holds the rest.
  The design intent is a single sentence. Deliver a power that is constant to a fraction of a percent for hours, cut it within a microsecond when an arc starts, and let the process control the thickness through a monitor that the supply does not disturb.
  Shorter: hold the rate, kill the arc, and leave the rest to the geometry.
  Vacuum compatibility of the supply's output path is not obvious and matters more than it seems. The connector at the chamber wall sees the full target voltage and a plasma that is a few centimetres away, so the connector's surface must be shielded against deposition; a conductive film on the connector's ceramic is a leakage path that grows with every run.
  A shield that is easy to remove and clean monthly is worth more than a connector with a longer creepage path, because the deposit arrives whatever the geometry.
  Magnetron targets erode unevenly, and the erosion changes the magnetic field that confines the plasma. The supply cannot see that change directly, and it shows up as a shift in the voltage needed for a given power. A log of that voltage per run is the earliest warning of a target that needs replacement.
  Sputtering supplies in this role often run at frequencies above the mains, and the choice of frequency interacts with the plasma. Frequencies in the tens of kilohertz are common for applications with pulsed output, and the pulse pattern is often used to discharge the target surface and prevent arc growth.
  Pulse parameters are process parameters. The pulse width, the pulse frequency and the reverse voltage during the pulse all affect the deposition, and the supply must expose them as set points rather than bury them in firmware.
  A coater that changes materials needs presets. A preset per material, holding power, pulse pattern, arc threshold and ramp, prevents the operator from re-tuning the supply at every material change and shortens the setup from hours to minutes.
  Nothing in this domain is expensive individually. The regulation, the arc handling, the sensor accuracy and the presets are each modest additions, and together they define whether a coater produces a hundred good stacks or a hundred doubtful ones.
  The last point is cultural. A coater's yield is usually attributed to the chamber, the targets and the operator, and the supply is noticed only when it fails. A measurement discipline that logs the supply alongside the monitor moves the attribution to where it belongs.