基于卷积神经网络判定方法的激光微透镜阵列微米级加工工艺

Micron-level processing technology of microlens array(MLA)photolithography based on convolutional neural network

在MLA曝光工艺中,曝光点的数量庞大,通过高倍率显微镜配合人工目检来判定曝光质量耗时耗力,造成工艺成本偏高。为了解决这个问题,设计了一种便于检测的圆环形图案并引入深度学习中的目标检测Yolov5模型,一定程度上能够取代人工目检,完成对曝光质量的快速判定。基于上述方法,分析了不同光刻胶厚度之下,线能量密度的最优区间与光刻胶的剖面倾角。并在同等线能量密度下通过圆度判定曝光图案失真情况。在本研究的MLA曝光工艺中,选取光刻胶厚度、激光曝光功率以及加工平台移动速度作为自变量,评价曝光合格率、光刻胶剖面倾角以及曝光圆度等加工质量参数具有重要的工程意义。

During microlens array(MLA)photolithography exposure process,the number of photolithography points is considerably large,thus,judgement of the photolithography quality by human eyes with a high-magnification microscope is time-consuming and labor-intensive,resulting in high process cost.To solve this problem,an easily detected circular pattern was designed and a Yolov5 model for target detec⁃tion in deep learning was introduced,which can replace manual eye inspection to a certain extent and com⁃plete the rapid judgement of photolithography quality.Based on the proposed method,the optimal interval of the level of energy density during laser scanning and the profile dip angle of the photoresist were ana⁃lyzed under different photoresist thicknesses.At the same level of energy density during laser scanning,the distortion of photolithography pattern was judged considering circularity.Further,the photoresist thickness,laser power,and processing platform moving speed were selected as independent variables in the MLA photolithography process to evaluate processing quality parameters processing quality parame⁃ters,such as photolithography qualification rate,photoresist profile inclination angle,and photolithogra⁃phy circularity,is of great significance for engineering.