Auswahl von Hochspannungsfilterkondensatoren für Lithographie-Lichtquellen

Auswahl von Hochspannungsfilterkondensatoren für Lithographie-Lichtquellen

Abstract

Alignment lasers, interferometers and calibration lamps in a lithography tool run from 5–15 kV auxiliary rails inside a pi-network filter. This article gives the selection arithmetic for those filters: the attenuation of a two-capacitor pi stage, the 3.2 V ripple of a 1 nF Class I part at 50 kHz, the 30% effective capacitance of a biased Class II part, and the partial-discharge screen that holds a decade of service.

1. Einleitung

Lithography tools carry a family of auxiliary light sources beside the main exposure beam. Alignment lasers, interferometer reference sources, calibration lamps and UV inspection illuminators all feed on clean optical output.

The supplies behind those sources convert 48–400 VDC into a 5–15 kV DC rail at 0.1–5 mA, filtered by a pi-network of high-voltage capacitors. The quality of that rail sets the optical noise, the wavelength stability and the usable life of the lamp.

Output quality is second-order for throughput and first-order for metrology. An alignment lamp that flickers by 0.2% corrupts the wafer-stage measurement used to position the next die, and a calibration source whose spectrum drifts invalidates the dose calibration of the whole tool.

The filter capacitor that shapes this rail therefore carries a requirement that industrial practice rarely demands: capacitance stable to parts per hundred across temperature and bias, partial discharge below the detection floor, and ten years of service without drift.

2. Kernmechanismus

2.1 Attenuation of the pi-network

The high-voltage filter is conventionally a pi-network: input capacitor C1, a series inductor L, output capacitor C2, then the lamp. The switching ripple at frequency f is attenuated by the LC section.

A = 1 / ((2π × f)² × L × C_eff)

where A is the ratio of output ripple to input ripple, f the switching frequency, L the series inductance, and C_eff the effective shunt capacitance.

C_eff = C1 × C2 / (C1 + C2)

where C_eff is the effective shunt capacitance, C1 the input capacitor, and C2 the output capacitor. Attenuation scales with the square of frequency and linearly with C_eff, so doubling the capacitance adds 6 dB while doubling the filter frequency adds 12 dB.

2.2 Ripple at the lamp electrode

The ripple that reaches the lamp comes from the capacitive division of the ripple current plus the loss term of the capacitor.

ΔV_ripple = I_ripple / (2π × f × C2) + I_ripple × ESR

where ΔV_ripple is the peak ripple at the lamp electrode, I_ripple the ripple current through C2, f the switching frequency, C2 the output capacitance, and ESR the equivalent series resistance of C2. At 50 kHz with 1 mA of ripple current, a 1 nF capacitor gives 3.2 V. Doubling to 2 nF halves that to 1.6 V, because the capacitive term dominates the loss term.

A 10 kV rail with 0.1% ripple carries 10 V at the electrode. The usable limit is one third of that, so a design target of 3.3 V is the number the capacitor has to meet.

2.3 Why Class II dielectric loses the margin

A Class II part loses capacitance under DC bias. At 80% of rated voltage an X7R capacitor may hold only 30% of its nominal value. The computed 1.6 V ripple then becomes 5 V, and the 3.2 V of a nominal 1 nF part becomes 10.6 V.

The filter corner for DC smoothing sits far below the switching frequency, and the corner moves with capacitance.

f_c = 1 / (2π × R × C)

where f_c is the corner frequency, R the load resistance, and C the filter capacitance. A 1 MΩ load with a 1 nF filter gives 160 Hz. A Class II capacitor that loses 40% of its value raises the 50 kHz ripple at the lamp by 4.4 dB, because the attenuation term rises in proportion to the lost capacitance.

2.4 Partial discharge sets the service life

Under sustained DC bias the dominant degradation mechanism is partial discharge at internal voids. The inception voltage must sit above the operating stress with margin.

V_operating ≤ V_PDIV / k_PD

where V_operating is the DC stress in service, V_PDIV the partial-discharge inception voltage, and k_PD the margin factor, taken as 1.2. Each discharge erodes a few picograms of dielectric, and over years of continuous operation the erosion channel bridges the electrodes. The part then fails as a short circuit that takes the supply down.

Fig. 1 — Class I N4700 holds 100% of nominal from 0 to 10 kV bias, while an X7R part drops to 30% and turns a 3.2 V ripple into 10.6 V.

3. Schwachstellen im Engineering

Pain Point 1: Bias-driven capacitance loss passes the switching ripple

Auxiliary sources used for interferometry or alignment cannot tolerate intensity modulation above a fraction of a percent. A filter capacitor whose capacitance collapses under DC bias passes the switching ripple.

The resulting optical flicker is diagnosed only after weeks of optical-side troubleshooting, although the electrical fix was one component swap away.

Pain Point 2: Partial-discharge erosion that no inspection finds

A filter capacitor with a PD inception voltage only slightly above the operating voltage passes the factory test and then erodes over 2–5 years. The failure arrives as a short circuit in the high-voltage section of the supply.

The capacitor looks undamaged in a visual inspection, so the root cause is missed and the replacement part repeats the same weakness.

Pain Point 3: Temperature drift of the filter corner

A 10 kV supply with a 1 MΩ load and a 1 nF filter smooths DC with a 160 Hz corner. A Class II capacitor that loses 40% of its value at high temperature raises the 50 kHz ripple at the lamp by 4.4 dB.

A supply that was quiet at 25 °C becomes noisy at 65 °C, with no change to any other part of the circuit.

Pain Point 4: Mounting fragility inside an optical frame

High-voltage filter capacitors are physically large. A leaded part cantilevered from the board inside a vibrating optical frame develops micro-cracks in the solder joint or the termination.

The result is intermittent contact that is nearly impossible to reproduce in the service bay.

4. Entwurfsschritte

Strategy 1: Derive the ripple allowance from the optical requirement

Determine the maximum tolerable intensity modulation, typically 0.1% or less, and translate it to electrode voltage ripple through the lamp V-I sensitivity. On a 10 kV rail that gives 10 V.

Size C2 so the ripple current term of §2.2 stays under one third of that figure, including the ESR term, which puts the design target at 3.3 V.

Strategy 2: Specify a voltage-stable Class I dielectric

Wählen N4700 (T3M) strontium titanate parts, whose capacitance is independent of DC bias and drifts only ±30 ppm/°C. The attenuation computed on paper is then the attenuation delivered at every temperature and load.

A Folienkondensator with a low reversal tolerance is the weaker choice where the rail also carries ripple current from the lamp arc.

Strategy 3: Screen partial discharge at 1.2× rated voltage

Require PD below 5 pC at 1.2× rated voltage as a per-part acceptance test. That single specification removes the slow-erosion failure mode from the reliability analysis.

Verify the operating point against the inception voltage, not only the rated value, because the inception voltage falls with temperature and with the number of discharge events already absorbed.

Strategy 4: Add a second pi-stage and support the parts

A two-stage pi-filter costs one extra capacitor and one inductor, and adds 20–40 dB of attenuation margin. The supply then tolerates capacitor spread and aging without a redesign.

Above a certain body height, use the HVCT8G screw-terminal family or a bolted mount, add a clamp, and apply conformal coating. UV-degraded dust in an optical bay attacks the creepage surface of a coated part far more slowly than a bare one.

Strategy 5: Monitor rail ripple in firmware

Add a ripple-detection ADC on the high-voltage rail through a divider built from high-value Bemessung von Widerständen and a low-value Abtastwiderstand.

Log ripple amplitude over time. A slow rise is the earliest observable signature of filter capacitor aging, and it lets the replacement happen in a scheduled maintenance window.

Komponentenparameter	Typischer Wert	Testbedingung
Nennspannung (HVCT8G-Serie)	10 kV DC	Hi-Pot 1.5× für 60 s
Kapazität	4700 pF ±20% (472M)	1 MHz, 1 V<sub>eff</sub>
ESR bei 1 MHz	unter 50 mΩ	-
Schleifeninduktivität (E40-Sehne)	unter 8 nH	40 mm Anschlussabstand
Isolationswiderstand	über 10 GΩ bei 500 V Gleichstrom	25 °C, 60 % relative Luftfeuchtigkeit
Temperaturbereich	−40 °C bis +125 °C	IEC 60068-2-Reihe
Impuls-dv/dt-Fähigkeit	über 50 kV/µs	bei Nennspannung

5. HVC capacitors for lithography auxiliary filters

The rail has to hold its capacitance across bias and temperature, survive a decade without maintenance, and sit still inside a vibrating optical frame. HVC builds N4700 parts from a 6,000 m² owned factory in Dongguan across a 1 kV to 100 kV range.

Sechs Parameter, die hier von Bedeutung sind

• Class I N4700 (T3M) on strontium titanate gives bias-independent capacitance and ±30 ppm/°C, so the 1 nF filter stays 1 nF at 10 kV and 65 °C.
• tanδ below 0.1% at 1 kHz keeps the filter parts cool, so self-heating does not disturb the temperature-sensitive optical frame around them.
• Voltage ratings from 1 kV to 100 kV cover 5 kV calibration sources through 30 kV excimer pre-ionization stages from one qualified family.
• Every part is PD-screened below 5 pC at 1.2× rated voltage, which supports 10-year service targets without PD erosion.
• Robust 441 screw-type and doorknob capacitors packages with vibration-qualified terminations suit a lithography frame.
• Field pedigree: an HVC 40 kV / 441 screw-type capacitor replaced a first-tier brand part in high-voltage field-service programs, and domestic OEMs are qualifying the same part. HVC parts match the electrical parameters and footprints of Murata DHS/DHR and TDK UHV/FHV ranges.

Lieferung und Support

Standard parts ship in 4–8 weeks and matched filter sets in 6–10 weeks. HVC applications engineering supports attenuation calculation, capacitance matching across a pi-network, and thermal review for optical-frame mounting.

Send the rail voltage, ripple current, attenuation target and ambient temperature. HVC returns a filter capacitor selection with measured C-versus-DC-bias curves, PD test data and thermal validation results, with samples in two weeks.

6. Häufig gestellte Fragen

Q1: Why does the dielectric class come before the capacitance value?

Because a Class II part does not deliver the capacitance it is ordered at. An X7R capacitor at 80% of rated bias may hold 30% of its nominal value, so a computed 1.6 V ripple becomes 5 V and a 3.2 V design becomes 10.6 V. Class I N4700 holds its value across bias and temperature, so the design calculation and the bench measurement agree.

Q2: What ripple does a 1 nF filter give at 50 kHz?

ΔV_ripple = I_ripple / (2π × f × C2) plus the ESR term. At 50 kHz with 1 mA of ripple current, a 1 nF Class I capacitor gives 3.2 V. That is inside the 3.3 V target derived from the 10 V / 0.1% optical limit on a 10 kV rail.

Q3: How do I set the partial-discharge requirement?

Specify below 5 pC at 1.2× rated voltage, and check the inception voltage against the real operating stress. The margin factor is 1.2, because inception voltage falls with temperature. Per-part PD screening is what removes the slow-erosion mode that otherwise takes 2–5 years to surface as a short circuit.

Q4: Why does capacitor temperature drift matter in an optical bay?

Because the corner frequency moves with capacitance. A 10 kV supply with a 1 MΩ load and a 1 nF filter smooths DC at 160 Hz. A Class II part that loses 40% of its value raises the 50 kHz ripple at the lamp by 4.4 dB, so a supply that is quiet at 25 °C becomes noisy at 65 °C.

Q5: What does a second pi-stage buy?

20–40 dB of attenuation margin for one extra capacitor and one inductor. That margin absorbs capacitor tolerance, temperature drift and aging, so the supply does not depend on a single part holding its value for ten years.

Q6: How do I catch filter aging before it becomes an optical fault?

Monitor the rail ripple through a divider with an ADC and log the amplitude over time. A slow rise is the earliest observable signature, and it schedules the replacement into a maintenance window. Combine it with tall-part mechanical support and conformal coating, because a vibrating frame cracks unsupported leaded terminations.

7. Referenzen

1. E. Kuffel, W. S. Zaengl, J. Kuffel, Grundlagen der Hochspannungstechnik, 2. Aufl., Newnes, 2000.
2. N. Mohan, TM Undeland, WP Robbins, Leistungselektronik: Wandler, Anwendungen und Design, 3. Auflage, Wiley, 2003.
3. MIL-HDBK-217F Mitteilung 2, Zuverlässigkeitsvorhersage elektronischer Geräte 1995.
4. IEC 60384-1:2018, Festkondensatoren für den Einsatz in elektronischen Geräten – Teil 1: Allgemeine Spezifikation.
5. IEC 60270:2025, Hochspannungsprüfverfahren – Ladungsbasierte Messung von Teilentladungen, 4. Aufl. Genf: IEC, 2025.
6. Entwicklung von BME X7R Mehrschicht-Keramikkondensatoren mit hoher Kapazität und hoher Spannung, KEMET Electronics (CARTS USA 2011 Proceedings), https://www.mouser.jp/pdfdocs/2011-03CARTS-DevelopmentofHighCapacitance-HighVoltageBMEX7RMLCCapacitors.pdf .
7. DN Donahoe, M. Pecht, IK Lloyd und S. Ganesan, Feuchtigkeitsbedingte Degradation von Mehrschicht-Keramikkondensatoren, Microelectronics Reliability, Bd. 46, Nr. 2-4, S. 400-408, 2006. https://doi.org/10.1016/j.microrel.2005.05.008.
8. Verwendung von Keramikkondensatoren 3 kV (Stabilitätsdiskussion C0G/NP0 vs X7R/Y5S), Forum Retrotechnique, https://forum.retrotechnique.org/t/utilisation-des-condensateurs-ceramiques-3kv/115858 .
9. GB/T 7354-2018, 高电压试验技术 局部放电测量. Titel: 2018.
10. HVC-Anwendungsentwicklungsteam Hochspannungs-Keramikkondensator Technische Ressourcen, https://www.hv-caps.com