基于强化学习的准分子激光器能量控制算法研究

Energy Control of Excimer Laser Based on Reinforcement Learning

光刻用准分子激光器的能量特性在集成电路的光刻过程中至关重要,直接影响光刻机曝光线条的精度。为了实现对于衡量能量特性的能量稳定性和剂量精度的精确控制,从放电电压调节的角度对激光脉冲的能量特性控制进行了研究。为了设计能量特性控制算法,首先对准分子激光器的放电特性建立了仿真模型,并验证了模型的有效性。然后,设计了基于强化学习的准分子激光器能量特性控制算法。最后在仿真模型上,分别采用Z-N(Ziegler-Nichol)参数整定的比例积分(PI)算法、粒子群优化(PSO)整定的PI算法和基于强化学习的算法对出光脉冲进行了控制,将最终的结果进行对比。实验结果证明,在基于强化学习的能量控制算法的控制下,激光器的能量稳定性小于4%,剂量精度小于0.3%,并且动态性能要优于Z-N参数整定的PI算法、PSO整定的PI算法。证明了算法的优越性,提高了光刻用准分子激光器的鲁棒性和实用性,满足了半导体光刻需求。

The energy characteristics of lithography excimer lasers are critical in the lithography process of integrated circuits and directly affect the accuracy of the exposure lines of the lithography machine.In order to design a laser energy control algorithm,a simulation model is built for the discharge characteristics of the excimer laser,and the validity of the model is verified.Then,design an energy control algorithm for excimer laser based on reinforcement learning.Finally,on the simulation model,the Z-N(Ziegler-Nichol)parameter tuning proportion integral(PI)algorithm,particle swarm optimization(PSO)tuning PI algorithm and reinforcement learning-based algorithm are used to control the pulse of laser output,and compare the final results.The experimental results show that under the control of the energy control algorithm based on reinforcement learning,the laser energy stability is less than 4%,the dose accuracy of is less than 0.3%,and the dynamic performance is better than the Z-N parameter tuning PI algorithm and PSO tuning PI algorithm.Prove the superiority of the algorithm,improve the robustness and practicability of lithography excimer laser,and meet the needs of photolithography.