基于CNN-LSTM的晶圆良率预测

晶圆良率是衡量半导体制造系统加工能力的关键指标，对其精准预测有利于排查晶圆制程工艺缺陷、提高晶圆生产率、控制企业生产成本。基于晶圆允收测试(wafer acceptance test, WAT)大数据，提出了基于卷积神经网络和长短期记忆网络(convolutional neural networks and long short-term memory, CNN-LSTM)的晶圆良率预测方法。该方法对WAT数据进行缺失、异常与归一化预处理；构建CNN模型对复杂WAT参数的关键特征进行识别；考虑相邻晶圆间的时序相关性，设计长短期记忆网络进行回归分析，从而实现晶圆良率的准确预测。以某工厂晶圆允收测试过程中采集的实际生产数据进行实验，并与其他传统晶圆良率预测方法的结果进行对比分析，从而验证所提方法的有效性。     

基于模糊逻辑决策的Interbay系统晶圆卡运输成本模型

为了提高晶圆制造Interbay物料运输系统的多目标调度性能,提出了一种基于模糊逻辑决策的晶圆卡运输成本模型.基于300mm晶圆制造系统数据进行仿真实验并与固定权重的晶圆卡运输成本模型进行比较,结果表明采用该模型能获得更优的Interbay系统多目标调度性能,表明论文提出的基于模糊逻辑决策的晶圆卡运输成本模型是有效的.     

面向紧急订单的晶圆制造AMHS多目标调度方法

为提高晶圆制造整体式自动化物料运输系统中紧急订单的搬运性能,以晶圆卡平均等待时间、平均加工周期、平均搬运时间作为优化目标,建立了整体式自动化物料运输系统多目标调度模型。提出了一种基于循序搜索启发式规则的物料运输系统多目标调度方法,该方法针对晶圆卡搬运请求进行邻域循序搜索以匹配更优的搬运小车,从而提高紧急晶圆卡的搬运效率。仿真实验结果表明,与现有基于启发式优先规则的调度方法相比,该方法在紧急晶圆卡的平均加工周期、平均搬运时间、平均等待时间等性能指标方面具有更优的综合性能,且对普通晶圆卡的搬运性能影响更小,验证了该方法的有效性。     

基于模糊逻辑的晶圆制造整体式AMHS动态调度方法

为提高晶圆制造整体式自动化物料搬运系统和晶圆加工系统的综合性能,提出了一种基于模糊逻辑控制的动态调度方法。该方法实时采集系统中的晶圆卡交货期、晶圆卡等待时间和AMHS搬运负载情况等信息,通过模糊逻辑控制模型动态调整系统当前调度规则,从而实现OHT实时高效搬运晶圆。仿真实验表明:基于模糊逻辑控制的动态调度方法在OHT平均利用率、晶圆平均搬运时间、晶圆平均等待时间和晶圆平均交货期满意度等性能指标方面优于传统单一规则,验证了该方法的有效性。     

基于模糊神经网络的冷连轧机轧制力预测

建立了一种基于数学模型和模糊神经网络共同作用的冷连轧机轧制力预测模型,通过数学模型描述轧制接触面积,模糊网络预测轧制单位压力.提出将Hough变换应用于神经网络的参数确定,从而使最终设计的网络具有最佳结构参数.试验研究证明了所设计模型具有较强的泛化能力和鲁棒性,大大提高了轧制力的预报精度.     

一种模糊神经网络轧辊磨表面粗糙度预测与控制方法

研究了一种模糊神经网络轧辊磨表面粗糙度智能预测及控制的方法,轧辊磨削精度和表面质量指磨削过程中的加工精度、表面粗糙度和物理机械性能,而表面粗糙度是其中最主要的一个因素.提出的基于模糊神经网络的轧辊磨表面粗糙度智能预测方法对于在轧辊磨削工艺中研究基于模糊神经网络的表面粗糙度预测,对于如何在加工过程中辨识表面粗糙度及时作出砂轮动作的调整,保证轧辊磨削质量有重要意义.同时由于可以实现砂轮表面粗糙度的在线控制与调整,提高了轧辊磨削的生产率.     

BP神经网络和数学模型在中厚板板凸度预报中的综合应用

分析中厚钢板板凸度计算模型并给出相应的在线数学模型。为了提高板凸度在线模型预测精度,提出了一种基于模糊聚类BP神经网络的板凸度模型影响系数的优化方法。并采用模糊聚类分析方法,科学选取学习样本,解决了样本多、学习速度慢的问题。通过大量在线数据分析,可知这种方法对中厚板板凸度的预报精度有很大改善,能适应不断变化的工艺过程和设备条件。     

基于模糊自适应变权重算法的采场冒顶函数链神经网络预报

为提高采场声发射事件率预报精度,将采场声发射事件率不同的单个预测模犁的预测值作为函数链神经网络的原始输入值,并将原始输入值按正交的三角函数扩展得到的数值作为函数链神经网络扩展输入值,在分析函数链神经嘲络拟合充要条件的基础上,结合模糊自适应变权重算法计算函数链神经网络权重,对采场声发射事件率进行基于模糊自适应变权重算法的函数链神经网络预测,对其预测结果再进行函数链神经网络算法拟合,然后结合采场冒顶尖点突变模型的判别式对采场冒顶进行预报.某铅锌矿采场冒顶预报结果表明,基于模糊自适应变权重算法的函数链神经网络预测方法的预测误差小于0.3%,可实现采场冒顶精确预报.     

支持向量回归在Zr-2合金晶粒尺寸预测中的应用

根据Zr-2合金的晶粒尺寸在不同热工艺参数(变形温度、变形程度、变形速率)下的12组实测数据,应用基于粒子群算法寻找最优参数的支持向量回归方法,建立了合金晶粒尺寸的预测模型.通过与模糊神经网络模型的结果进行比较,结果表明:基于相同的试验样本,支持向量回归预测模型的平均绝对误差和平均绝对百分误差都比模糊神经网络预测模型的小,而复相关系数大.这说明,支持向量回归预测模型预测精度比模糊神经网络模型要高,是简单而精确的建模方法,可用于优化热加工参数.     

基于动态递归模糊神经网络的轧机辊缝动态辨识

模糊系统和神经网络,由于具有逼近任意连续非线性映射的特性,而广泛应用于系统的辨识与控制。但是传统的模糊神经网络是一种静态映射,不适用于动态系统的辨识,而轧制过程中影响轧机辊缝的因素复杂,外界干扰严重,过程参数难以确定,为提高轧机辊缝动态的辨识精度,提出了一种基于动态递归模糊神经网络的辨识模型。轧制仿真结果表明,该模型具有很高的辨识精度。     

Modelling and prediction of cut shape for plasma arc cutting based on artificial neural network

Abstract

An artificial neural network approach for the modelling of plasma arc cutting processes is introduced. Neural network models have been proposed for predicting the cut shape and estimating the special cutting variables. The implementation of artificial neural networks in the modelling of cutting processes is discussed in detail. The performance of the neural networks in modelling is presented and evaluated using actual cutting data. Moreover, prediction applications of the above neural network models are described for various cutting conditions. It is shown that estimated results based on the proposed models agree well with experimental data; the neural network models yield good prediction results over the entire range of cutting process parameters spanned by the training data. The testing and prediction results show the effectiveness and satisfactory prediction accuracy of the artificial neural network modelling. The developed models are applicable to carbon steel.     

大直径硅片超精密磨削技术的研究与应用现状

随着IC制造技术的飞速发展，为了增加IC芯片产量和降低单元制造成本，硅片趋向大直径化，原始硅片的厚度也相应增大以保证大尺寸硅片的强度；与此相反，为了满足IC封装的要求，芯片的厚度却不断减小，需要对图形硅片进行背面减薄。硅片和芯片尺寸变化所导致的硅片加工量的增加以及对硅片加工精度和表面质量更高的要求，使已有的硅片加工技术面临严峻的挑战。本文详细分析了传统硅片加工工艺的局限性，介绍了几种大直径硅片超精密磨削加工工艺的原理和特点，评述了国内外硅片超精密磨削技术与装备研究和应用的现状及发展方向，强调了我国开展大直径硅片超精密磨削技术和装备研究的必要性。     

压电双晶片直接驱动式伺服阀的研究

提出一种利用压电陶瓷片直接驱动的伺服阀,分析了双压电晶片的静、动态特性,制造了双晶片直接驱动式伺服阀样机.该伺服阀具有高于传统电磁式伺服阀的频宽与分辨率,它的机械结构简单,抗干扰能力强.     

基于BP神经网络的轮胎模具微铣削能耗预测

针对子午线轮胎模具微铣削加工过程中能耗计算问题,以主轴转速、每齿进给量、切削深度3个重要铣削参数作为变量,设计轮胎模具微铣削加工能耗实验.根据实验数据构建基于BP神经网络的微铣削能耗预测模型.通过改进预测模型的激活函数,提高模型的预测精度.结果表明:所提的预测模型有效,可以实现不同铣削参数组合下的能耗预测.     

基于人工神经网络的焊接接头力学性能预测系统

对神经元网络在焊接接头力学性能预测上的应用做了探索,训练了焊接方法包括焊条电弧焊、气体保护焊、埋弧焊和TIG焊的抗拉强度、屈服强度、断后伸长率和断面收缩率模型.并在此基础上设计完成了基于人工神经元网络的焊接接头力学性能预测系统.利用可视化界面编程技术和数据库技术制作了友好的人机用户界面.焊接接头力学性能预测系统包括添加...     

基于多源数据融合的半导体晶片CMP抛光材料去除率预测

目的 对半导体晶片抛光过程中的工艺参数、耗材使用量、抛光垫状态参数等多源数据预处理后进行数据融合,建立材料去除率(MRR)预测模型,为实现半导体晶片抛光加工工艺的决策和处理奠定基础。方法 研究晶片抛光加工中的数据特点及数据融合需求,提取数据集中每个晶片加工过程中的统计特征并生成新数据集,同时引入邻域特征以应对晶片加工过程中动态因素对材料去除率的影响。提出基于深度自动编码器的多源数据融合及材料去除率预测方法。设计深度自动编码器参数,优化深度自动编码器的损失函数从而增强深度自动编码器对强相关性特征变量的重建。基于深度自动编码器进行多源传感器信号融合,降低数据维度。使用超参数搜索算法优化BP神经网络超参数,利用BP神经网络方法将融合后的数据进行半导体晶片抛光过程中的材料去除率预测。结果 采用PHM2016数据集对模型进行验证,均方误差MSE达到7.862,相关性R²达到91.2%。结论 基于多源数据的融合模型能有效预测MRR,可以对半导体晶片CMP工艺过程的智能决策与控制起到良好的辅助作用。     

Mechano-chemical polishing of silicon wafers

Rapid progress in recent IC fabrication industry has increased the demand of tight specification of non-uniformity (NU) and surface polishing in silicon wafer planarization. Chemical–mechanical polishing (CMP) is currently the most popular method for IC wafer planarization. However, the sub-surface damage problem caused by hard abrasives and chemical waste problem of CMP have decreased the throughput and increased the cost of IC fabrication. This study is to investigate the mechano-chemical polishing (MCP) of silicon wafers by slurry of soft abrasives, BaCO₃ and through experiments to verify that the solid phase chemical reaction (SPCR) is the main reaction process involved in MCP. A planarization mechanism with compliance has been designed and tested through MCP experiments. Experimental results of MCP of silicon wafers have achieved the average of surface roughness improvement ratio (SRIR) to 99% and the surface roughness R_a=0.633 nm measured by atomic force microscope (AFM). The material removal rate (MRR) has been calculated and the significant influence of slurry weight percent and polishing pressure have been found. The NU has also been estimated for evaluation of MCP parameters. The sub-surface damage of silicon wafer has not yet been found in experimental results and hence the MCP process of silicon wafers has been verified to become a green or environment-friendly technology of silicon wafer planarization.     

Theoretical analysis on the effects of crystal anisotropy on wiresawing process and application to wafer slicing

Most of the crystals sliced using wiresaw are anisotropic to an extent. The effect of crystal anisotropy on the process of slicing using wiresaw is studied and presented in this paper. A method is proposed to determine the direction of approach (DOA) which will give a better surface finish and reduce deviation from the desired surface normal by maintaining symmetry in material removal rates on the two sides of the wire. The effect of cleavage anisotropy on wiresaw slicing is also studied. If the DOA is perpendicular to a cleavage direction, then the longitudinal direction of the wire aligns with the cleavage direction which increases the tendency of wafer breakage, resulting in lower yield of the wafers. This can be easily avoided by choosing an appropriate DOA. Theoretical analysis is carried out using the proposed methods for slicing silicon wafers. Recommendations are made for three most commonly sliced orientations of silicon: (100), (110) and (111). DOA can be any direction for (100) and (110) wafers from the symmetry point of view but preferred DOAs do exist for these wafers from cleavage point of view. For (111) crystal there are exactly six DOAs with symmetry. However, these six DOAs do not lie in the preferred zones suggested by cleavage criterion. It is suggested that in such situations the symmetry criterion should be given precedence over the cleavage criterion during wiresawing process, as the semiconductor industry has strict tolerances in place for surface normal deviation and flatness.     

Optimization of transmittance characteristic of indium tin oxide film using neural networks

Indium thin oxide (ITO) films were deposited by a DC-powered plasma sputtering system. The transmittance characteristics of ITO film are investigated and optimized by using a neural network model and by a statistical factor analysis. For systematic modeling, 2^5-1 with a resolution V design was utilized for 5 process parameters including wafer temperature, DC power, chamber pressure, Cesium (Cs) canister temperature, and Cs carrier flow rate. A generalized regression neural network was used to build a model of transmittance. The statistical parameter analysis revealed a larger main effect of the power and pressure over the others. The prediction performance of genetic algorithm-optimized model is 2.70 in the root mean square error. The impact of the carrier flow rate or the wafer temperature was insensitive to the variation in the power. An increase in the transmittance was noted as either the Cs temperature or the pressure increased in particular at a lower power. The impact of the wafer temperature and carrier flow rate was the opposite of those for the Cs temperature and pressure. A high transmittance at a low surface roughness was noted as a function of the power and the Cs carrier flow rate.