Optimization of Dynamic Response in High-Voltage Power Supplies for Exposure Machines

In microelectronics manufacturing, photovoltaic panel production, and precision display processing, exposure machines are core equipment for lithography processes, where imaging quality directly determines product yield. The high-voltage power supply, as a critical subsystem, provides stable high-voltage input to light sources (e.g., mercury lamps, excimer lasers, or X-ray tubes). During exposure, the light source must pulse on/off within milliseconds, with voltage fluctuations controlled within ±0.1%. Otherwise, uneven exposure energy or line-width distortion may occur. Thus, the dynamic response of high-voltage power supplies—including voltage adjustment speed, ripple suppression, and load transient recovery time—is a key factor affecting process precision. 
I. Core Challenges in Dynamic Response
1. Load Transients and Energy Stability 
   In pulse-mode operation, the load current of exposure machine light sources undergoes periodic step changes (e.g., from no-load to full load instantaneously). If the power supply's response delay exceeds 10 μs or recovery time is prolonged, pulse energy may drop or overshoot, leading to uncontrolled exposure dosage. In step-and-scan lithography, minor dynamic response deviations accumulate into wafer-scale pattern distortions. 
2. Multi-Interference Coupling 
   High-voltage power supplies face three interferences: 
   • Internal noise: Switching device (e.g., IGBT/MOSFET) noise coupled to outputs via parasitic capacitance; 
   • Load instability: Plasma discharge oscillations causing current fluctuations; 
   • Grid disturbances: Voltage sags from workshop equipment start/stop. 
   These interferences superimpose high-frequency ripple (>100 kHz) on the output, requiring peak-to-peak suppression to 0.001% of the output voltage to prevent light source spectral drift. 
II. Key Technical Pathways for Optimization
1. Topology Innovation 
   • Multi-stage regulation: A PFC + LLC resonant conversion + linear regulation architecture enhances input robustness, reduces losses via soft-switching, and enables millisecond-level fine-tuning. 
   • Distributed bus: Locating high-voltage modules (e.g., voltage multiplier circuits) near light sources minimizes transmission impedance effects. Remote low-voltage DC bus (e.g., 400 V) supplies power, reducing capacitive load delays. 
2. Control Algorithm Advancements 
   • Adaptive predictive control: Models the transfer function between load current and output voltage. Real-time feedback sampling dynamically adjusts PID parameters, cutting recovery time to <5 μs and overshoot by 60%. 
   • Feedforward compensation: Directly injects compensating current upon input voltage transients, limiting output deviation to ±0.05% within 100 μs during 20% grid sags. 
3. High-Frequency Devices and Materials 
   • Wide-bandgap semiconductors: SiC MOSFETs or GaN HEMTs enable MHz switching frequencies, reducing filter inductor inertia. High-temperature tolerance relaxes thermal design constraints. 
   • Low-ESR ceramic capacitors: MLCCs (ESR<1 mΩ) parallel to outputs absorb high-frequency ripple, forming π-filters with toroidal inductors to suppress ripple below 10 mVpp. 
III. System-Level Co-Optimization Strategies
1. Thermal-Electronic Integration 
   Thermal simulations optimize heat dissipation (e.g., vapor chambers + micro-heat pipes), limiting junction temperature fluctuations to ±5°C. Temperature compensation coefficients in control loops correct gain errors from conduction resistance drift. 
2. Digital Twin Verification 
   Digital models simulate interference scenarios (e.g., load steps, grid flicker) to pre-validate dynamic response. Industrial cases show <3% deviation between simulation and testing, reducing development cycles by 40%. 
IV. Future Trends
Extreme ultraviolet (EUV) lithography demands three-high power supplies: high voltage (>30 kV), high frequency (>10 kHz), and high precision (<±0.01%). Future optimization will integrate AI for nanosecond-level predictive control, such as deep learning-based plasma instability anticipation. 
Conclusion
Dynamic response optimization in exposure machine high-voltage power supplies is a multidisciplinary challenge. Breakthroughs in topology, control algorithms, and device integration must converge to push the limits of response time, ripple suppression, and stability—enabling the next leap in nano-manufacturing precision.