Control of Lithography Light Source High-Voltage Power Supplies in Immersion and High-NA Lithography Systems

An excimer laser source for lithography discharges a pulse of a few millijoules in tens of nanoseconds at several kilohertz, and the pulse energy must be stable to a fraction of a percent because the dose delivered to the resist is the integral of many pulses. The high-voltage supply charges the pulse power module's capacitor between discharges, and the accuracy of that charge determines the pulse energy. In immersion and high-numerical-aperture systems, where the depth of focus is a few tens of nanometres, the dose error becomes a critical dimension error directly.

Charging accuracy is the first requirement and it is not a voltage accuracy requirement. The energy in the capacitor is proportional to the square of the voltage, so a voltage error of a tenth of a percent produces an energy error of two tenths. The supply's regulation must therefore be specified in energy terms, and the calibration must be done against pulse energy rather than against the voltage reading.
Repetition rate sets the charging time available. At six kilohertz, the interval between pulses is a hundred and sixty-seven microseconds, and the charging circuit must deliver the full charge in perhaps a hundred and twenty of those. A supply that charges in a hundred and fifty microseconds cannot support six kilohertz, regardless of its steady-state specifications.
Charging in a resonant circuit is the usual approach and the supply's job is to control the resonant pulse's energy. A series resonant converter charges the capacitor through a resonant inductor, and the charge terminates when the current crosses zero; controlling the initial voltage of the resonant capacitor sets the charge. A supply with a precise initial voltage and a fast termination produces repeatable energy without measuring the pulse.
Pulse-to-pulse energy stability is affected by the discharge's residue. After each pulse, a small amount of energy remains in the capacitor and in the parasitic elements, and the next charge starts from that residue rather than from zero. A supply that does not account for the residue produces a pattern of alternating energies, visible as a two-pulse periodicity in the dose. Measuring the residue voltage at the start of each charge and correcting the target removes the pattern.
Gas mixture and pressure affect the discharge voltage for a given energy, and the supply must follow. As the gas ages, the discharge voltage rises, and a supply that holds the charge voltage constant produces a falling pulse energy; a supply that closes the loop on pulse energy adapts. The measurement of pulse energy, usually from a photodiode or an energy sensor in the beam path, becomes the reference for the charging loop.
Timing jitter is a source-level requirement and the supply contributes through its trigger. The trigger must fire the switch within a few nanoseconds of the command, and the charging must be complete before the trigger arrives with margin. A supply whose charging complete signal is late forces the trigger to be delayed, which increases the jitter in the pulse's arrival at the wafer.
Immersion systems add a thermal constraint that couples to the supply. The projection lens and the immersion liquid are temperature-stable to millikelvin, and heat from the source must not reach them. The supply's dissipation, even at a distance, contributes to the enclosure's thermal load; a supply with high efficiency reduces the burden on the environmental control system.
High-numerical-aperture systems require tighter dose control because the process window is smaller. The same dose error that produces a two-nanometre critical dimension shift at a lower numerical aperture produces a larger shift at a higher one, and the supply's stability specification must scale accordingly. Specifying the supply against the process window, rather than against a generic standard, is what makes the numbers meaningful.
Electromagnetic interference from the pulsed discharge is severe and the supply must be designed for it. The discharge radiates broadband energy, and the charging converter's control electronics sit within a metre of it. Shielded enclosures, fibre control links, and filtering on every cable entry are baseline; a supply that is not designed for this environment will produce erratic charging that is attributed to the laser.
Component ageing in the charging circuit appears as a slow change in efficiency. The resonant inductor's core loss rises, the capacitor's capacitance falls, and the charge delivered for a given initial voltage changes. Tracking the initial voltage required to reach a target energy over time gives an early indication of ageing, well before the energy control loop runs out of range.
Redundancy and hot swap are practical requirements in a fab. A source that stops for a supply failure costs production hours, and a supply that can be replaced in twenty minutes with a spare limits the loss. Designing the interface so that a replacement does not require recalibration is a packaging decision that is worth more than a marginal improvement in efficiency.
Diagnostics should be at the pulse level. Recording charge voltage, residue voltage, pulse energy, and the time between charging complete and trigger for each pulse gives a data set that identifies drift, jitter and periodicity separately. Averaged data hides exactly the effects that matter.
Cost is dominated by the accuracy and the reliability requirements, not by the power level. A supply that delivers a few hundred watts with the stability required for a lithography source costs far more than an industrial unit of ten times the power, because the accuracy and the pulse-level control are the expensive parts. Comparing on power rating alone misses the point.
Limits should be acknowledged. The pulse energy is set by the discharge physics and the gas, and the supply sets the initial conditions; no supply can produce a stable energy from a discharge that is itself unstable. Similarly, the source's bandwidth for dose correction is limited by the repetition rate, and the supply cannot accelerate that.
An acceptance test that mirrors the application is the right one. Run at the specified repetition rate for a million pulses, recording energy, charge voltage, residue and timing; compute the dose stability over windows of a hundred pulses, which is how the scanner integrates. A million pulses is a few minutes of running and reveals everything the steady-state tests miss.
Gas ageing is a slow drift that the supply must follow without running out of range. As the mixture degrades, the discharge voltage rises and the required charge voltage rises with it; a supply whose maximum voltage is set with a small margin will reach its limit before the gas is exhausted, and the source will be replaced early. Specifying the voltage range with twenty percent headroom allows the gas to be used to its end.
Trigger jitter and charging margin are linked, and the link should be documented. The margin between charging complete and trigger arrival is what absorbs jitter; reducing the margin to gain repetition rate removes the absorption. Recording the margin, and refusing to reduce it below a stated value, keeps the timing stable as settings are optimised.
Records per pulse rather than per wafer are what the source needs. A wafer's dose is the sum of thousands of pulses, and a defect caused by a single anomalous pulse is invisible in the sum. Pulse-level logs with an anomaly threshold surface the individual events, and correlating them with die-level inspection results shows whether a rare pulse error is costing yield.
Maintenance is dominated by the store and the switch, and both should be replaced on charge-count rather than on failure. A store that fails in service takes the switch with it, doubling the cost and the downtime. Replacing on count, with the old unit returned for analysis, keeps the failure mode visible without paying for it in production.
Limits should be acknowledged plainly. The supply sets the initial conditions for each discharge, and the discharge's stability depends on the gas, the electrodes and the optical cavity. No amount of charging accuracy can compensate for a cavity that is misaligned, and the effort spent improving the supply beyond a certain point yields nothing at wafer level.
Thermal stability of the charging circuit is part of the dose budget. The resonant inductor's loss varies with temperature, and the charge delivered for a given initial voltage changes by a fraction of a percent as the enclosure warms from a cold start. A warm-up sequence that brings the supply to thermal equilibrium before production begins removes a drift that would otherwise appear in the first wafers of a lot.
A closing observation: the pulse-level data set is the most valuable asset the source produces, because it captures behaviour that wafer-level metrology cannot resolve. Keeping it, with a consistent schema, and reviewing it when yield moves, turns a maintenance record into a diagnostic instrument.
Finally, the supply's enclosure and the source's thermal environment should be reviewed together. A charging supply that dissipates heat into the same air path that cools the optical module raises the module's temperature and shifts the cavity's alignment; separating the air paths, or moving the supply to a remote bay, is a packaging decision with optical consequences.