Synchronization of excimer laser high-voltage power supplies in laser marking and mask repair systems
Excimer laser marking stations run the tube at 400 Hz repetition rate, and the high-voltage chain sits between the mains feed and the laser discharge tube as a charge supply, a pulse forming network, and a thyratron switch. The charge supply pumps the PFN capacitors to 650 V between shots, with a recharge window of 2.3 ms at 400 Hz. The thyratron fires the network into the tube at 24 kV, and the discharge lasts about 20 nanoseconds. Pulse energy on a fresh gas fill reads 300 mJ, and the energy stability spec holds within plus or minus 2.5 percent over a burst. The charge supply output is regulated to 650 V within 2 V, because a 5 V drift in charge voltage moves the shot energy by 1.8 percent, enough to push marking depth out of spec on certain polymer substrates. Energy feedback from a photodiode tap on the output beam corrects the charge setpoint every 10 shots, and the loop settles after 30 shots into a burst. The PFN capacitor bank, a stack of 16 pulse capacitors, sees 24 kV peak and 400 Hz stress, and the capacitance drifts 2 percent per year.
Long bursts expose an energy droop that the feedback loop fights but never fully wins. Over a 60-second burst at 400 Hz, the tube warms 8 degrees Celsius and the gas mixture shifts, so the shot energy droops 4 percent even with a flat charge voltage. The feedback loop pushes charge voltage up from 650 V to 672 V to compensate, and the end-of-burst energy returns within 1 percent of the start. The cost shows up on the thyratron: higher charge voltage raises the tube current peak by 6 percent, and the thyratron anode dissipation climbs 9 watts per pulse train. Thyratron lifetime on the station averages 8e8 shots at 650 V setpoint, and the droop compensation pushes the measured lifetime down to 6.5e8 shots. The maintenance planner schedules thyratron swaps at 6e8 shots regardless, to keep the margin wide. Marking depth consistency across the burst improved enough to justify the tube wear, and the process group refused the alternative of slowing the burst to 300 Hz. The burst length, 60 seconds, was set by the product cycle time, and a shorter burst would waste fixture index time.
Synchronization between the charge supply, the trigger chain, and the host motion table decides whether a mark lands on the right spot. The station marks moving parts on a rotary fixture, and each trigger pulse must follow the encoder position within 1.5 microseconds. The HV trigger board receives a TTL sync from the motion controller, delays the signal by a calibrated 80 nanoseconds for cable propagation, and fires the thyratron grid. Jitter on the trigger chain measures 1.2 nanoseconds peak to peak on an oscilloscope capture, well under the 1.5 microsecond budget. When the encoder cable shield broke on one station, the sync reference drifted 3 microseconds, and marks shifted 0.4 mm on the product. Scrap ran 120 parts before the fault was traced to a cold solder joint on the encoder connector. The cable was rerouted away from the motion cable bundle, because the moving chain had abraded the shield over 6 months of cycles. The trigger board power supply, a 15 V rail on the control rack, gets a monthly ripple check, and the ripple sits at 20 mV peak to peak. A noisy rail would add jitter to the thyratron grid, and the 15 V rail cap bank was replaced at 24 months after the ripple crept to 80 mV.
Mask repair cells run at lower repetition rate but tighter fluence control. The repair station opens opaque defects on quartz mask blanks, and each repair pulse must deliver 120 mJ per square centimeter on a 5 by 5 micrometer spot. The charge supply holds 610 V, the tube runs at 50 Hz, and the burst length stays at 5 to 20 pulses per repair site. Fluence tolerance is plus or minus 3 percent, because under-dose leaves residual chrome and over-dose digs into the quartz substrate. The energy tap on the repair cell uses a calibrated photodiode with a 16-bit digitizer, and the setpoint adjusts after every single shot rather than every 10. The loop gain is deliberately high on this cell, which causes occasional oscillation on first shot after a long idle. A pre-fire of 3 dummy pulses at 20 percent energy warms the tube and damps the oscillation, and the operator routine includes the pre-fire automatically since 2023. The repair cell beam delivery optics focus the spot through a 100 mm objective, and the focus position drifts 5 micrometers per 24 hours as the optical bench settles. The focus check, done daily on a reference reticle, takes 10 minutes and happens before the first repair site.
Tube gas life sets the consumable budget. A fresh gas fill lasts about 5e6 shots on the marking station, after which the output energy droops 12 percent and the tube begins showing red glow at the discharge edges. Gas refill costs 900 dollars per fill and takes 4 hours of station time. The refill window was widened from 5e6 to 6e6 shots on two stations, and the extra lifetime came with a 3 percent rise in arc frequency. Arcs inside the tube dump extra energy into the electrodes, and electrode erosion accelerates. The tube, a 45000 dollar component, lasts about 12 fills at the conservative schedule; at the widened schedule, the lifetime estimate drops to 9 fills. The maintenance group returned to the 5e6 shot window after one tube failed early with pitted electrodes. The math on refill cost versus tube cost tipped back to the conservative schedule. The tube electrodes, copper-tungsten bars 15 mm wide, show 0.3 mm erosion after 12 fills, and the erosion rate doubles when arc frequency rises. The vacuum gauge on the tube head reads 5e-3 Pa on a healthy fill, and a slow rise to 8e-3 Pa signals gas leak through the electrode seals.
Preionization degradation shows as a slow rise in arc count. The preionizer sparks 4 times per shot across the tube, and the spark gaps erode over 3e6 shots. Eroded gaps widen from 1.2 mm to 1.6 mm, the preionization energy drops, and the main discharge arcs more often. Arc count on a healthy station runs 3 to 5 per million shots; after 3e6 shots, the count climbs to 40 per million, and each arc costs one marked part because the pulse energy goes nowhere useful. Preionizer gap adjustment takes 2 hours and requires opening the tube end flange, which means a gas purge and refill. The adjustment interval was set at 2.5e6 shots after data from six stations showed a sharp arc rise after that point. The adjustment tooling is a custom spacer gauge, and the two technicians on the floor keep the gauge in the station cabinet rather than the tool room, because access delay once cost a missed production window. The preionizer supply, a 15 kV pulse unit on the tube head, draws 30 W average, and the supply capacitor bank gets replaced at 3e6 shots along with the gap adjustment.
Cooling water quality causes a slow drift nobody expected. The tube sits in a closed loop with 8 liters per minute of deionized water at 22 degrees inlet. Water resistivity in the loop spec sits at 2 megaohm-centimeter, and a monthly check on the marking station reads 1.8 to 2.5. On one station the resistivity fell to 0.4 megaohm-centimeter, and the tube wall began showing micro-leaks along the ceramic seam. The leak developed over three weeks, and the station started losing gas 15 percent faster than the fill schedule predicted. Root cause traced to a failed deionizer cartridge in the loop, which let dissolved ions build up and conduct across the ceramic-to-metal seal. The seal was replaced, the loop recharged with fresh DI water, and the deionizer cartridge went onto a 6-month replacement schedule instead of the yearly one. The station has run clean for 14 months since. The water flow alarm, set at 6 liters per minute, has never tripped, but the flow gauge reading trends down 0.2 liters per minute per year as the filter cartridges clog, and the cartridges get changed on the same 6-month cycle.
High-voltage cable between the charge supply and the PFN runs 3 meters in shielded conduit, and the cable insulation checks out at 1000 megaohm on a fresh install. After 18 months, the insulation resistance falls to 200 megaohm, measured with a 500 V megohmmeter at the cable end. The leakage current, 3 microamps at 650 V, shows up on the charge supply current reading as a slow upward drift. The cable replacement happens at 24 months, taking 2 hours, and the spare cable costs 450 dollars. The cable bend radius at the PFN end, 150 mm, satisfies the manufacturer spec, and the flexing during burst cycling is limited to thermal expansion rather than motion. The conduit ground strap, a 16 mm braid, gets a torque check at the quarterly visit, because a loose strap couples switching noise into the energy feedback photodiode signal.
Shift handover shows the operational reality. The day shift logs shot count, energy trend, arc count, and gas fill history in the station logbook, and the night shift reads the trend before starting the first burst. The charge supply needs little attention; the only recurring fault is a fan bearing rattle at 18 months, caught by the night shift on acoustic inspection. Cabinet temperature under load reads 36 degrees, and the intake filter gets cleaned monthly in the sub-fab. Marking depth samples, checked every 2 hours on witness coupons, catch process drift before scrap does. The whole HV chain, from charge supply to thyratron to trigger board, draws 2.2 kW from the mains during bursts and 120 W at idle. Nobody on the floor treats the chain as a problem child, and the monthly maintenance hour budget for the station sits at 4 hours, most of the budget spent on gas refill rather than on the electrical parts. The spare parts cabinet holds two spare trigger boards, one spare photodiode module, and three preionizer capacitor packs, and the stock check runs weekly during the shift handover.
The marking depth measurement on witness coupons uses a confocal microscope, and the scan step size is 0.5 micrometer. The microscope lamp, a halogen source, lasts 1500 hours, and the lamp replacement happens at the monthly visit. The coupon rack, a 10-position holder inside the station chamber, rotates one position per burst, and the rotation mechanism gets a visual inspection at the quarterly visit. The rotation motor, a 12 V geared unit, draws 0.3 A, and the motor current trend shows no drift over 2 years. The station exhaust fan, a 200 mm axial unit, runs at 1800 rpm, and the fan blade cleaning happens monthly to remove dust buildup. The fan vibration, measured at the quarterly visit, trends up from 0.8 mm per second to 1.5 mm per second over 18 months, and the fan bearing gets replaced at the 2 mm per second mark.
The high-voltage interlock chain on the station covers the door switch, the water flow switch, and the gas pressure switch. The door switch trips the HV enable within 20 milliseconds, and the switch contact resistance reads under 100 milliohm on a clean unit. The water flow switch trips at 6 liters per minute, and the switch has never fired falsely. The gas pressure switch trips at 0.3 MPa, and the switch log shows one trip in 3 years, traced to a regulator fault on the gas rack. The interlock logic board, a 16-channel unit on the control rack, gets a visual inspection at the quarterly visit, and the board capacitors get an ESR check at the annual shutdown. The spare interlock board costs 950 dollars, and the spare sits on the shelf in the station cabinet.
The station air filter, a 400 by 400 mm panel, gets replaced at the quarterly visit, and the filter pressure drop reads 80 Pa on a clean unit. The station exhaust duct, a 200 mm spiral pipe, gets a visual inspection at the quarterly visit, and the duct joints get a torque check at the annual shutdown. The station work light, a 36 W fluorescent fixture, gets a tube replacement at the annual shutdown, and the fixture ballast gets an ESR check at the same visit. The station door interlock, a magnetic contact switch, trips the HV enable within 15 milliseconds when the door opens, and the switch contact resistance reads under 80 milliohm on a clean unit. The door interlock log shows one spurious trip in 2 years, traced to a loose cable shield on the switch sensor.
