基于遗传算法的光刻机光源掩模优化技术研究
文献类型：学位论文
| 作者 | 杨朝兴 |
| 学位类别 | 博士 |
| 答辩日期 | 2016 |
| 授予单位 | 中国科学院上海光学精密机械研究所 |
| 导师 | 王向朝 |
| 关键词 | 光刻 分辨率增强技术 光源掩模优化 遗传算法 |
| 其他题名 | Study on Source Mask Optimization Based on Genetic Algorithm for Lithograhpy |
| 中文摘要 | 集成电路产业是信息技术产业的核心。光刻机是大规模集成电路制造的关键设备。光刻分辨率决定了集成电路的特征尺寸。为了制造更小特征尺寸的集成电路，业内提出了一系列光刻分辨率增强技术。与传统光刻分辨率增强技术如光学邻近效应校正技术相比，光源掩模优化技术同时优化照明光源和掩模图形，具有更高的优化自由度，是进一步提高光刻分辨率和工艺窗口的关键技术之一。基于遗传算法的光刻机光源掩模优化（Source Mask Optimization based on Genetic Algorithm, GA-SMO）技术不需要掌握光刻成像模型等先验知识，可以使用复杂光刻成像模型和目标函数。此外GA-SMO技术具有全局寻优能力。但是GA-SMO技术模仿自然进化过程进行优化搜索，需要大量光刻仿真计算，导致其优化速度慢，从而限制了该技术的实际应用。针对该问题，本文对GA-SMO技术进行了研究，主要内容包含以下几个方面： 1.提出了一种基于实数编码的GA-SMO技术。使用实数字符串代替二进制字符串描述照明光源和掩模图形，使用实数编码遗传算法的选择、交叉和变异算子代替二进制编码遗传进化算子。仿真实验表明，通过使用实数编码遗传算法提高了GA-SMO的收敛速度。以交错接触孔图形的仿真实验为例，使用实数编码遗传算法后GA-SMO的适应度值收敛到6.0所需的平均代数比二进制编码GA-SMO减少65.5%（约100代）。 2.提出了一种基于多极光源描述方法的GA-SMO技术。多极光源描述方法使用一组圆形子光源描述小光瞳填充比例的自由形式照明光源。相比像素光源描述方法，多极光源描述方法可以用更少的变量描述小光瞳填充比例自由形式照明光源，从而提高遗传算法收敛速度和光刻成像仿真速度。仿真实验表明，在优化成像质量相近的情况下基于多极光源描述方法的GA-SMO技术的优化速度比基于极坐标像素光源描述方法的GA-SMO技术提高7倍。 3.提出了一种基于动态适应度函数的GA-SMO技术。在遗传算法优化过程中采用动态适应度函数模拟实际光刻工艺条件误差对光刻结果的影响，得到对光刻工艺条件误差不敏感的优化照明光源和掩模图形。该技术无需优化权重系数，即可获得与权重优化后的加权适应度函数技术相近的工艺宽容度，减少了优化时间。典型逻辑图形的仿真实验表明，曝光剂量误差为15%时，动态适应度函数技术得到的最优光源和最优掩模的可用焦深达到200nm，与加权适应度函数技术的优化效果相当。动态适应度函数技术也可用于降低最优光源和最优掩模对其它工艺条件误差如彗差的敏感度。 4.提出了一种基于多染色体的GA-SMO技术。使用多染色体遗传算法，实现了像素光源和像素掩模的联合优化。与采用矩形掩模优化的单染色体技术相比，多染色体技术具有更高的优化自由度，获得了更优的光刻成像质量和更快的收敛速度。典型逻辑图形的仿真实验表明，多染色体技术得到的最优光源和最优掩模的适应度值比单染色体技术减少7.6%，多染色体技术仅需132代进化即可得到适应度值为5200的最优解，比单染色体技术减少127代。 |
| 英文摘要 | The integrated circuit industry is the core of the information technology industry. Lithography tool is the key equipment for the very-large-scale integrated circuit fabrication. The resolution of the lithography determines the critical dimension of the integrated circuit. Resolution enhancement technology is used to fabricate integrated circuits with smaller critical dimension. In comparison with traditional resolution enhancement technology such as optical proximity correction, source mask optimization (SMO) technology optimizes the illumination source and mask pattern simultaneously and has larger degree of freedom for optimization. SMO is the key technology to increase the resolution and process window of lithography further. Source mask optimization based on genetic algorithm (GA-SMO) does not require prior knowledge about lithographic process, so it can use more complex lithography model and objective function. In addition, GA-SMO has the capability of global search. However, as GA-SMO imitates the principle of natural evolution to optimize the source and mask, it requires the lithography simulation to be performed many times. This slows down the optimization process and restricts the practical application of GA-SMO. In this dissertation, GA-SMO is studied to improve the convergence speed. The main contents are as follows. 1. A GA-SMO method using real-coded genetic algorithm is proposed. The source and mask solutions are represented by floating point strings instead of bit strings. The selection, crossover, and mutation operators are replaced by corresponding floating-point versions as well. Simulation results show that the convergence speed of the real-coded GA is improved in comparison with the binary-coded version. For example, the simulation result for the staggered contact hole pattern shows that the real-coded method needs about 53.7 generations on average to converge to a solution with the fitness value of 6.0, while the binary-coded method needs about 65.5% (100) more generations to get similar result. 2. A source representation method which increases the speed of GA-SMO is proposed. The proposed representation method uses a group of circular poles to describe the freeform illumination source whose pupil filling ratio (PFR) is small. Compared with conventional pixelated source representation methods, the proposed multi-pole source representation method can represent the low-PFR freeform illumination source with fewer variables. The multi-pole source representation method increases the speed of the GA convergence and the lithography simulation. Simulations are conducted to prove the improvement of GA-SMO speed. On the premise that the simulation conditions are the same and optimization qualities are comparable, GA-SMO using the proposed multi-pole source representation method is about 7 times faster than that using the polar-grid pixelated source representation method. 3. A dynamic GA-SMO method is developed. The dynamic method uses dynamic fitness function in genetic algorithm to simulate the impact of real-world process variations on the lithographic results. In this way, the imaging quality of the optimized source and mask is not sensitive to process variations. The dynamic method can get similar result as the conventional weighting SMO method and does not need to optimize the weighting coefficient. The dynamic method saves the time of weighting factor optimization process and increases the speed of the GA-SMO process. Simulation result shows that the dynamic method can get a usable defocus of 200nm when the dose error is 15%. This result is comparable with the optimization result of the weighting method. The dynamic GA-SMO method can be also used to make the optimal source and mask less sensitive to other process errors, such as coma. 4. A pixelated SMO method based on multi-chromosome genetic algorithm is proposed. The multi-chromosome GA-SMO method uses two chromosomes to optimize the pixelated source and pixelated mask simultaneously. By contrast, the single-chromosome GA-SMO method uses one chromosome to optimize the pixelated source and rectilinear mask. So the multi-chromosome method has larger degree of freedom for optimization than the single-chromosome method. As a result, the multi-chromosome method achieves better imaging quality and faster convergence speed. Simulation results show that the optimal fitness value of multi-chromosome method is 7.6% smaller than that of the single-chromosome method. Besides that, the multi-chromosome method needs only 132 generations to converge to an optimal solution with the fitness value of 5200, while the single-chromosome method needs 259 generations to get similar result. |
| 语种 | 中文 |
| 源URL | [http://ir.siom.ac.cn/handle/181231/15963]  |
| 专题 | 上海光学精密机械研究所_学位论文 |
| 推荐引用方式 GB/T 7714 | 杨朝兴. 基于遗传算法的光刻机光源掩模优化技术研究[D]. 中国科学院上海光学精密机械研究所. 2016. |
入库方式： OAI收割
来源：上海光学精密机械研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。