Challenge
OLED display manufacturing requires precise temperature uniformity during the plasma CVD (Chemical Vapor Deposition) process. The susceptor heat exchanger must maintain uniform temperature across large glass substrates, but traditional design methods rely on engineer experience and repetitive manual analysis.
Fig. Susceptor system schematic: Galinstan liquid metal circulation via electromagnetic pump
Approach: PPO Reinforcement Learning
Fig. Reinforcement learning environment with General, Record, and FDM spaces
Flow Channel Optimization
- PPO algorithm with discrete action space (up/down/left/right)
- FDM-based thermal simulation for fast iteration (100,000+ episodes)
- 4-way symmetry exploited to minimize computational domain
- Reward function: Temperature uniformity improvement
Electromagnetic Pump Design
- Helical-type EM pump for high-pressure Galinstan delivery (10 bar)
- FEM simulation using COMSOL for MHD analysis
- Sm₂Co₁₇ permanent magnets with magnetic shielding
Results
Susceptor Heat Exchanger
Fig. Flow path evolution during RL training: paths become more complex as learning progresses
- Generated flow-channel layouts targeting a more uniform susceptor temperature in the thermal model
- Investigated reinforcement learning as an alternative to repeated manual channel design
Electromagnetic Pump
Fig. Helical EM pump analysis: magnetic flux density distribution (left) and pressure distribution achieving 10 bar (right)
- Predicted 10 bar discharge pressure at 2 kg/s flow rate through MHD simulation
- Simulated average magnetic flux density: 0.25 T (max 0.4 T)
Publications
Peer-reviewed Journal
- G. Lee, T. Kim, Y. Kim, G. Kim, “Design and Analysis of an OLED Display Susceptor System Using Reinforcement Learning,” Journal of Korean Applied Artificial Intelligence, 1(2), 1-6, 2025. DOI
Conferences
- G. Lee, T. Kim, “Formation of the LCD susceptor heat exchanger path using the deep reinforcement learning,” 2nd Applied Artificial Intelligence Conference, 2023.
- G. Lee, Y. Yu, T. Kim, S. Kim, “Formation of a thermofluid temperature equalization flow path using reinforcement learning,” 1st Applied Artificial Intelligence Conference, 2022.
Software registration
- G. Lee, Y. Yu, T. Kim, “Heat exchanger path generation program through reinforcement learning,” Republic of Korea – Registration No.: C-2022-049451, 2022.
Research Project
(2022) Technical support for the application of artificial intelligence to lower module temperature uniformity on next-generation display manufacturing systems, KAERI.