Interlayer control of vacuum coating high-voltage power supplies in anti-fog coatings and optical filters
The vacuum coating line runs twin magnetron sputtering at 40 kHz mid-frequency, each target driven by a 5 kW pulsed DC supply. The anti-fog hydrophilic layer deposits as a 120 nm silica film on automotive lens substrates, and the optical filter stack builds up to 45 alternating silica and titania layers, each layer 60 to 90 nm thick. The supply output holds 420 V at 11 A on each channel, and the arc detection circuit trips on a voltage dip below 35 V within 0.8 microseconds, clamping the output for 20 microseconds before restart. Arc frequency on a healthy process runs 2 to 4 per minute on the silica target and 1 to 2 per minute on the titania target. Substrate bias sits at minus 120 V pulsed, and the bias supply shares the rack with the magnetron channels. Substrate heating holds 200 degrees Celsius through the deposition cycle. The target, a 300 mm by 100 mm rectangular plate, bolts to a water-cooled backing plate, and the backing plate flow runs at 4 liters per minute per target.
Target poisoning in reactive sputtering sets the process window. Oxygen partial pressure during titania deposition holds at 2e-2 Pa, and the supply current droops as the target surface oxidizes. A 10 percent rise in oxygen flow drops the target current by 1.5 A within 30 seconds, and the deposition rate falls from 0.4 nm per second to 0.25. The process group runs the oxygen flow in closed loop, sensing the partial pressure and trimming the supply power to hold the rate. When the loop tuning drifted after a software update in 2024, the rate wandered 12 percent on the first shift, and the first 3 layers of the filter stack came out 8 nm thin on average. The stack metrology caught the error at layer 5, and the batch of 120 substrates went back for rework. The loop gains were returned to the 2023 values, and the tuning table was locked behind an engineer password. The oxygen mass flow controller, a 100 sccm unit on the gas rack, gets a weekly flow check, and the controller drift measures 2 percent per quarter.
Layer thickness control depends on supply power stability. Each layer runs on a timed recipe, and the time is derived from the measured rate, which tracks the power delivered. Power ripple on the magnetron output spec sits at 1 percent peak to peak, and the installed units read 0.7 percent on a fresh supply. After 12 months of continuous use, the output capacitor bank ages, and the ripple climbs to 1.4 percent. The extra ripple shows up as a layer thickness spread of plus or minus 3 percent instead of the spec plus or minus 2. The capacitor bank replacement happens at 18 months, taking 6 hours per supply and costing 2800 dollars. The line runs 4 magnetron supplies in parallel, and the bank swaps are staggered so the line never stops for more than one supply at a time. The rate measurement, a quartz crystal monitor near the substrate, samples the deposition rate at 1 Hz, and the crystal gets replaced every 100 hours because the accumulating film shifts the crystal frequency.
Arc suppression tuning trades layer quality against downtime. The arc clamp at 20 microseconds was lengthened from 10 microseconds after a batch showed nodule defects on the titania layers. Nodules form when an unclamped arc pits the target surface, and the pit ejects droplets that land in the growing film. The longer clamp lets the arc quench fully, but each clamp costs 20 microseconds of deposition, and the 3 arcs per minute cost about 6 seconds per layer. On the 45-layer stack, that adds 4.5 minutes per cycle, about 2 percent of throughput. The process group accepted the loss, because nodule scrap ran 8 percent before the change and fell to 1 percent after. Shortening the clamp back to 10 microseconds would save the 2 percent but bring the nodule risk back, and nobody on the floor wants that trade. The arc detector board, a 4-channel unit in the supply chassis, gets a visual inspection at the quarterly visit, and the board feedthrough capacitors get an ESR check at the annual maintenance shutdown.
Anti-fog layer deposition runs a separate recipe on the same line. The hydrophilic silica top layer deposits at 150 degrees substrate temperature, lower than the 200 degrees used for the filter stack, and the oxygen flow runs 30 percent higher. The supply power for the anti-fog recipe sits at 3.2 kW per target, below the 5 kW rating. The higher oxygen flow poisons the target faster, and the target cleaning cycle, a 5-minute pre-sputter on a shutter, runs before every batch. Without the pre-sputter, the first 10 seconds of deposition come out silicon-rich, and the anti-fog contact angle reads 55 degrees instead of the spec below 20. The contact angle check on witness coupons runs every 4 hours, and the coupon rack sits in the chamber during the full cycle. The lens substrates, glass parts 80 mm in diameter, load onto a rotating drum at 5 rpm, and the drum rotation ensures uniform coating around the lens curve.
Substrate bias drift affects film density. The minus 120 V bias accelerates argon ions into the growing film, and the ion energy packs the film tighter. Bias drift of 10 V shows up as a 2 percent rise in film stress, measured on witness wafer curvature. The bias supply monitors the output with a 12-bit ADC, and the drift over a 12-hour cycle measures 6 V on a calibrated unit. A failed bias supply on one channel in 2024 let the bias sit at minus 90 V for 3 layers, and the film stress measurement on those layers ran 15 percent high. The batch of 80 substrates failed the stress spec and went to scrap. The bias supply got a weekly output check against a handheld meter, and the check adds 15 minutes to the weekly maintenance window. The bias feedthrough, a ceramic insulator on the chamber wall, gets a visual inspection at the monthly visit, and the feedthrough insulation resistance measures 1000 megaohm on a clean unit, falling to 200 megaohm after 6 months of buildup.
Pump-down and vent cycles wear the chamber seals. The chamber vent with dry nitrogen takes 15 minutes, and the pump-down to 5e-4 Pa takes 45 minutes. The door seal, a 12-meter silicone ring, flattens over 200 vent cycles, and the leak-up rate rises from 0.01 Pa per minute to 0.08. The seal replacement happens at 250 cycles, taking 3 hours. The cryopump regenerates every 2 weeks, and the regeneration cycle takes 6 hours, which blocks the line on weekends. The supply cabinet sits in the cleanroom corridor next to the chamber, and the ambient temperature holds at 21 degrees. Cabinet internal temperature under load reads 32 degrees, and the intake filters clean monthly. The supply fan runs at 2000 rpm, and the bearing noise started at 14 months on the first unit, prompting a bearing check at 12 months on the rest. The cryopump cold head, a 4 K stage unit, gets a helium charge check at the quarterly visit, and the cold head vibration trend shows a slow rise from 0.2 mm per second to 0.6 mm per second over 18 months.
Optical filter layer stack metrology catches thickness drift. The spectrophotometer, a 300 nm to 800 nm scanning unit, measures the transmission spectrum of witness coupons after each batch, and the spectrum fit extracts the layer thicknesses. A 5 nm thickness error on any layer shifts the reflection peak by 8 nm, and the spec tolerates plus or minus 3 nm per layer. The metrology routine runs 10 minutes per batch, and the data feeds the process log. The spectrophotometer lamp, a deuterium-tungsten source, lasts 2000 hours, and the lamp replacement happens at the quarterly visit. The witness coupons, 50 mm square glass plates, run through the chamber cycle with the production substrates, and the coupon rack position rotates one slot per batch to average out drum position bias.
Shift handover records the process state. The day crew logs power trend, arc count, oxygen flow, and layer count in the chamber logbook. The night crew reads the trend before starting the first batch, and the contact angle samples get collected on the 2-hour mark. The monthly maintenance hour budget per supply sits at 3 hours, split between capacitor checks, bias calibration, and filter cleaning. The line runs 26 days per month, and the scheduled downtime blocks 2 days for pump service and regeneration. Unplanned downtime averaged 14 hours per quarter before the 2024 software fix, and 7 hours after. The supply chain contributes about 2 hours of that, mostly on fan bearings and contactor contacts. The contactor on the mains input chatters on two units, and the contacts get filed clean during the quarterly visit. The spare parts shelf holds two spare arc detector boards, one spare bias supply module, and three spare magnetron feedthrough insulators, and the stock check runs weekly during the shift handover.
The chamber load lock, a 500 mm diameter airlock, transfers substrates between the atmosphere and the chamber, and the lock pump-down takes 10 minutes to 1e-3 Pa. The lock door seal, a 2-meter O-ring, flattens over 300 vent cycles, and the leak-up rate rises from 0.02 Pa per minute to 0.1. The seal replacement happens at 350 cycles, taking 4 hours. The lock gate valve, a pneumatic 200 mm unit, cycles 2 times per batch, and the valve diaphragm gets a replacement at the 5000 cycle mark. The diaphragm costs 180 dollars, and the valve swap takes 2 hours. The lock roughing pump, a 10 cubic meter per minute rotary vane unit, runs at 1400 rpm, and the pump oil gets changed at the quarterly visit. The pump vibration, measured at the quarterly visit, trends up from 1.2 mm per second to 2.0 mm per second over 18 months, and the pump bearing gets replaced at the 2.5 mm per second mark.
The substrate heater, a 3-zone resistive unit on the chamber wall, holds 200 degrees within plus or minus 3 degrees, and the heater zone current reads 8 A per zone at setpoint. The heater thermocouple, a K-type probe near the substrate, gets a quarterly calibration check, and the thermocouple drift measures 2 degrees per year. The heater power supply, a 24 V at 40 A unit on the rack bottom, gets a monthly ripple check, and the ripple spec sits at 100 mV. The spare heater supply costs 1600 dollars, and the spare sits on the shelf in the chamber cabinet. The heater cable, a 2 m high-current lead, gets a quarterly visual check, and the cable connector gets a torque check at the annual shutdown.
The chamber exhaust, a 200 mm pipe to the scrubber, pulls 1000 cubic meters per hour, and the exhaust flow alarm trips at 800 cubic meters per hour. The scrubber catalyst bed, a 20 kg activated carbon unit, gets a replacement at the 6-month mark, and the bed costs 2800 dollars. The exhaust fan, a 300 mm axial unit, runs at 1500 rpm, and the fan blade cleaning happens monthly. The fan vibration, measured at the quarterly visit, trends up from 1.0 mm per second to 1.6 mm per second over 2 years, and the fan bearing gets replaced at the 2 mm per second mark.
The optical monitor window, a 100 mm quartz port on the chamber wall, gets a visual inspection at the quarterly visit, and the window transmission gets a check at the annual shutdown. The window coating, a thin anti-reflection film, wears off over 2 years, and the window replacement happens at the 30-month mark. The spare window costs 950 dollars, and the spare sits on the shelf in the chamber cabinet. The window heater, a 100 W band around the port, keeps the window at 60 degrees to prevent film deposition on the glass, and the heater current reads 0.5 A at setpoint. The heater thermostat, a mechanical bimetal unit, gets a quarterly calibration check, and the thermostat drift measures 3 degrees per year.
The process gas panel, a 6-channel unit on the gas rack, holds the mass flow controllers for argon, oxygen, and sputter gases, and the panel gets a weekly flow check. The gas cylinders, a 6-pack manifold in the gas cabinet, get a pressure check at the weekly visit, and the cylinder exchange happens when the pressure drops below 2 MPa. The cylinder regulator, a dual-stage unit, gets a quarterly leak check, and the regulator diaphragm gets a replacement at the 2-year mark. The gas cabinet exhaust, a 150 mm pipe to the scrubber, pulls 800 cubic meters per hour, and the exhaust flow alarm trips at 600 cubic meters per hour. The gas cabinet leak detector, a semiconductor sensor unit, samples the cabinet air every 60 seconds, and the detector alarm trips at 10 ppm of process gas concentration.
