摘要
在正交试验的基础上,将BP神经网应用于负性光刻胶(SU-8)加工高分辨率和高深宽比微结构的工艺研究,建立了光刻图形质量与前烘时间、前烘温度、曝光量、后烘时间之间的预测模型,该模型采用三层前向网络,学习算法为梯度下降法。通过实验,得出:前烘温度与前烘时间对光刻质量影响最大,对120~340μm厚的光刻胶,前烘温度取90℃,前烘时间50~100min时,图形的相对线宽差最小;最佳后烘时间(85~95℃)为30min;超声搅拌能缩短显影时间,改善图形质量。试验结果与网络预测结果吻合。结果表明,将BP神经网络应用于UV—LIGA技术中,可以优化光刻工艺。
The process of photolithographic was studied in this paper by using an Artificial Neural Network (ANN) based on orthogonal experimental design. From experiment it could be concluded that the soft bake temperature and time were the key factor of the structure quality. When the photoresist thickness ranged from 120 to 340μm, the soft bake temperature and time was 90℃ and 50-120 minutes, the perfect image could be obtained. The best post exposure bake temperature was 85-95℃ with less 30 minutes bake. In order to obtain the suitable parameters of the various thickness photoresist, an Back Propagation (BP) network with 3 layers was built based on orthogonality experiment. The ANN was trained with BP algorithm. The prediction was according to the experiment results, which proved that the lithographic process of the fabrication of high resolute micro structure could be ootimized with ANN.
出处
《光电工程》
EI
CAS
CSCD
北大核心
2006年第11期36-39,共4页
Opto-Electronic Engineering
关键词
光刻工艺
UVLIGA
预测模型
优化
Lithographic process
UV LIGA
Prediction model
Optimization