Rapid Localization of High-Voltage Power Supply Failures in Lithography Machines

As a core equipment in semiconductor manufacturing, the stability of the high-voltage power supply system (typically outputting tens of thousands to millions of volts) in lithography machines directly affects wafer exposure accuracy. This power supply operates under high-load, high-electric-field conditions, with complex failure causes. Rapid fault localization requires a systematic approach integrating layered diagnostic strategies, multidimensional signal analysis, and intelligent algorithms. 
I. Common Failure Types in High-Voltage Power Supplies
1. Electrical Characteristic Failures 
   • Output Abnormalities: Sudden voltage drops or failure to boost, often caused by power device (e.g., IGBT/MOSFET) breakdown, feedback circuit failure, or switching power faults. 
   • Protective Tripping: Accounts for >40% of failures, triggered by short circuits or insulation aging leading to leakage current surges. 
   • Current Feedback Loss: Operational amplifier circuit damage or monitoring line disconnection, causing beam loss of control. 
2. Mechanical and Thermal Management Failures 
   Transformer or high-voltage stack thermal stress cracks due to poor heat dissipation, manifesting as arcing or abnormal noises; cooling system blockages (e.g., pure water pipe scaling) induce overheating protection and accelerate device aging. 
3. System-Level Failures 
   Control board logic errors caused by electromagnetic interference (EMI) or component aging, e.g., low-level control signal failure. 
II. Rapid Localization Methodology
1. Layered Diagnostic Process 
   • Primary Tier: Power-off inspection of fuses, power connections, and base voltage to eliminate basic circuit issues. 
   • Secondary Tier: Oscillator testing under no-load conditions, using oscilloscopes to measure high-frequency signals (e.g., 30kHz blocking oscillator), locating root causes like filter capacitor leakage. 
   • Tertiary Tier: Disassembly inspection of target wear, seal ring aging, and coolant status, combined with infrared thermal imaging to identify localized overheating. 
2. Multidimensional Signal Monitoring 
   • Electrical Signals: Real-time voltage ripple and current harmonic collection, analyzed via Fourier transform for frequency-domain features. For example, proportional voltage decay indicates feedback loop failure. 
   • Mechanical Signals: Vibration sensors capture abnormal spectra, with wavelet transforms identifying resonance from bearing wear or structural loosening. 
   • Thermal Signals: Non-contact infrared probes monitor temperature gradients; ±5% differentials trigger alerts. 
3. Intelligent Diagnostic Technologies 
   • Deep Learning Models: CNN/RNN networks identify fault features. E.g., pre-charge voltage curves failing to reach 90% threshold (V3 < 0.9V1) indicate contactor adhesion or pre-charge resistor burnout. 
   • Multi-Sensor Fusion: Integrated voltage, vibration, and temperature data processed via decision-level fusion algorithms (e.g., Kalman filtering), improving localization accuracy by 40%. 
III. Systematic Handling Workflow
1. Emergency Response: Immediately cut power upon overvoltage/overcurrent triggers to prevent failure propagation. 
2. Closed-Loop Verification: 
   • Post-component replacement, simulate operation under low voltage, comparing virtual and actual output curves via digital twin models. 
   • Calibrate voltage dividers and monitoring instruments to ensure ±1% feedback precision. 
3. Predictive Maintenance: 
   • Develop fault tree models (FTA) for regular insulation testing (e.g., ceramic sleeves) to prevent dust-induced creepage. 
   • Use LSTM networks to predict component lifespan, e.g., replacing electrolytic capacitors when capacitance drops to 80% of nominal value. 
IV. Challenges and Future Trends
Current difficulties include electromagnetic shielding in high-power-density designs (e.g., liquid cooling + soft magnetic materials) and the complexity of silicon carbide device protection. Future efforts will focus on embedded diagnostic chips for failure prediction (e.g., thermal imaging pre-judging transformer failure) and enhancing fault-tolerant control to ensure single-point failures do not interrupt processes. 
Conclusion
Rapid localization of high-voltage power supply failures in lithography machines requires a closed-loop monitoring-analysis-decision system integrating hardware detection and intelligent algorithms. Layered diagnostics reduce localization time, while predictive maintenance lowers failure rates, ultimately ensuring semiconductor manufacturing continuity and yield.