晶圆表面缺陷模式的在线探测与自适应识别研究

晶圆表面的缺陷通常反映了半导体制造过程存在的异常问题,通过探测与识别晶圆表面缺陷模式,可及时诊断故障源并进行在线调整。提出了一种晶圆表面缺陷模式的在线探测与自适应识别模型。首先该模型对晶圆表面的缺陷模式进行特征提取,基于特征集对每种晶圆模式构建相应的隐马尔科夫模型（Hidden Markov Model,HMM）,并提出基于HMM动态集成的晶圆缺陷在线探测与识别方法。提出的模型成功应用于WM-811K数据库的晶圆缺陷检测与识别中,实验结果充分证明了该模型的有效性与实用性。     

基于GLRGMM的间歇过程在线监控策略

针对间歇过程的非线性和动态性,提出了全局—局部正则化高斯混合模型 （GLRGMM）算法。首先引入邻域保持嵌入算法提取局部流形结构,通过寻求一种低维投影对非线性过程进行全局结构保持,同时最大限度地保留局部流形特征;然后通过对高斯混合模型引入正则项来在线监控更新高斯模型,获取非线性数据流形结构,解决数据动态性问题;最后集成全局—局部监控指标实现在线监控。通过青霉素发酵过程进行了验证,结果表明所提算法比DPCA、GLNPE具有更好的在线监控效果。     

晶圆表面缺陷模式识别的二维主成分分析卷积自编码器

由于半导体制造过程的高度复杂性和动态性,各种过程故障通常导致晶圆表面出现各种缺陷模式.为了有效地识别晶圆表面缺陷模式从而及时地诊断和控制故障源,提出一种深度神经网络模型--二维主成分分析卷积自编码器(two-dimensional principal component analysis-based convolutional autoencoder, PCACAE).首先,提出一种基于改进的二维主成分分析算法(conditional2DPCA,C2DPCA)的图像卷积核,形成PCACAE的第1个卷积层;其次,对卷积输出进行池化操作并卷积编码重构,构建一个卷积编码器,并提取其编码部分作为PCACAE的第2层卷积层的初始化权值,从而形成一个深度网络模型,实现晶圆图像的特征学习;最后, PCACAE网络进行训练微调得到最终网络模型.将PCACAE应用于WM-811K晶圆图像数据库并与其他算法进行对比测试,实验结果表明, PCACAE在晶圆表面缺陷识别上的性能优于其他经典的卷积神经网络模型(如GoogLeNet,DensNet等),从而验证了该方法的有效性与工业可应用性.     

基于声学特征空间非线性流形结构的语音识别声学模型

从语音信号声学特征空间的非线性流形结构特点出发, 利用流形上的压缩感知原理, 构建新的语音识别声学模型. 将特征空间划分为多个局部区域, 对每个局部区域用一个低维的因子分析模型进行近似, 从而得到混合因子分析模型. 将上下文相关状态的观测矢量限定在该非线性低维流形结构上, 推导得到其观测概率模型. 最终, 每个状态由一个服从稀疏约束的权重矢量和若干个服从标准正态分布的低维局部因子矢量所决定. 文中给出了局部区域潜在维数的确定准则及模型参数的迭代估计算法. 基于RM语料库的连续语音识别实验表明, 相比于传统的高斯混合模型(Gaussian mixture model, GMM)和子空间高斯混合模型(Subspace Gaussian mixture model, SGMM), 新声学模型在测试集上的平均词错误率(Word error rate, WER)分别相对下降了33.1%和9.2%.     

一种增量式类内局部保持降维算法

针对在线学习中的算法效率问题,提出了一种增量式类内局部保持降维算法.该算法综合考虑了基于QR分解的降维算法与保类内Fisher判别分析法的优点,根据训练过程中新增的样本进行投影矩阵在线更新,克服了传统的批量式训练方法在线学习时计算量过分冗余的缺陷.同时,通过兼顾输入样本的局部结构和全局分布状态,使得该算法能够有效地应用于多簇、重叠的数据形态.在ORL人脸库和COIL20图像库上的实验表明,该增量式算法不仅在降维效果上基本与批量式算法保持一致,而且具有较大的效率优势.     

一种新的局部空间排列算法

局部切空间排列算法(local tangent space alignment,LTSA)是一种经典的非线性流形学习方法,能够有效地对非线性分布数据进行降维,但它无法学习局部高曲率数据集.针对此问题,给出了描述数据集局部曲率的参数,并提出一种局部最小偏差空间排列(locally minimal deviation spacealignment,LMDSA)算法.该算法考虑到局部切空间低鲁棒性的缺陷,在计算局部最小偏差空间的同时,能够发现数据的局部高曲率现象,通过参数控制及邻域间的连接信息,减少计算局部高曲率空间的可能,进而利用空间排列技术进行降维,手工流形及真实数据集的实验证实了该算法学习局部高曲率数据集的有效性.     

一种基于局部保持的隐变量模型

隐变量模型是一类有效的降维方法,但是由非线性核映射建立的隐变量模型不能保持数据空间的局部结构。为了克服这个缺点,文中提出一种保持数据局部结构的隐变量模型。该算法充分利用局部保持映射的保局性质,将局部保持映射的目标函数作为低维空间中数据的先验信息,对高斯过程隐变量中的低维数据进行约束,建立局部保持的隐变量。实验结果表明,相比原有的高斯过程隐变量,文中算法较好地保持数据局部结构的效果。     

子模式局部保持映射人脸识别

研究表明基于整体思想的人脸识别方法由于忽略图像的局部信息,在识别性能方面不如局部信息特征保持较好的基于子模块思想的识别算法。基于应用流形技术对图像降维后能够较好保持非线性子流形中的局部数据流形结构,提出了一种改进的子模式局部保持映射人脸识别算法。其主要思想是将同类的不同图像一并划分子集,由同位置子图组成子模块,并对子模块运用LPP算法学习其流形结构,与将不同类图像一并划分子集学习流形的方法不同。实验表明,该算法能更好地保持人脸图像的局部流形结构和信息特征,提高了识别率。     

基于GMM的间歇过程故障检测

对间歇过程的多操作阶段进行划分时,往往会被离群点和噪声干扰,影响建模的精确性,针对此问题提出一种新的方法:主元分析--多方向高斯混合模型(Principal component analysis-multiple Gaussian mixture model, PCA-MGMM)建模方法.首先用最短长度法对数据进行等长处理,融合不同展开方法相结合的处理方式消除数据预估问题;利用主元分析方法将数据转换到对故障较为敏感的低维子空间中,得到主元的同时消除了离群点和噪声的干扰;通过改进的高斯混合模型(Gaussian mixture model, GMM)算法对各阶段主元进行聚类,减少了运算量的同时自动得到最佳高斯成分和对应的统计分布参数;最后将局部指标融合为全局概率监控指标,实现了连续的在线监控.通过一个实际的半导体制造过程的仿真研究验证了所提方法的有效性.     

基于流形学习的异常检测算法研究

化探异常识别是成矿预测的重要依据。化探异常识别本质上是一不均衡数据的分类问题。异常识别过程中面临的主要问题是高维数据的处理问题,流形学习通过非线性降维方法实现维数约简。提出了一种基于流形学习的异常识别算法,通过流形学习进行维数约简,结合AdaCost技术,以改善不平衡数据的分类性能。以某锡铜多金属矿床的数据为研究对象进行仿真实验,实验结果表明该算法能够更准确地圈定区域化探异常,为成矿预测与评价提供了新的解决途径。     

Knowledge augmented broad learning system for computer vision based mixed-type defect detection in semiconductor manufacturing

Defect detection is a critical measurement process for intelligent manufacturing systems to provide insights for product quality improvement. For complex products such as integrated circuit wafers, several types of defects are usually coupled in a piece of wafer to form a mixed-type defect, which poses a challenge to current defect detection methods. This paper proposed a knowledge augmented broad learning system with a knowledge module and broad selective sampling module, which provides a multichannel selective sampling network to decouple the mixed-type defects. In this model, each channel is equipped with a pre-trained deformable convolution model to extract the feature of a fixed single-type defect. The knowledge module is designed to activate the candidate network channel by pre-detection of wafer maps. The experiment results indicated that the proposed model outperforms conventional models and other deep learning models, which demonstrated that the knowledge augmented broad selective sampling mechanism is effective for mixed-type defect detection.     

Defect cluster recognition system for fabricated semiconductor wafers

The International Technology Roadmap for Semiconductors (ITRS) identifies production test data as an essential element in improving design and technology in the manufacturing process feedback loop. One of the observations made from the high-volume production test data is that dies that fail due to a systematic failure have a tendency to form certain unique patterns that manifest as defect clusters at the wafer level. Identifying and categorising such clusters is a crucial step towards manufacturing yield improvement and implementation of real-time statistical process control. Addressing the semiconductor industry’s needs, this research proposes an automatic defect cluster recognition system for semiconductor wafers that achieves up to 95% accuracy (depending on the product type).     

A self-growing hidden Markov tree for wafer map inspection

This paper presents an automatic identification of the defect spatial wafer map using a growing wavelet-based hidden Markov tree (gHMT) statistical model. The hierarchical tree-based model, gHMT, utilizes the growing and learning procedure to increase successively the size of the wavelet tree. It can characterize image processing masks from the defect spatial patterns. Like the standard hidden Markov tree, gHMT cannot only capture the statistical behavior of the real-world measurements at multiple scales in space and frequency but also has the ability to accurately identify the locations of the defect regions using the smallest possible size. These regions provide essential information and intrinsic features of each pattern. When all the possible defect patterns are modeled by gHMT, the maximum likelihood classifier is applied to the wavelet energy features extracted from each trained models. Accordingly, defect spatial patterns are identified. The effectiveness of the proposed classifier based on gHMT is illustrated through the experimental data from a wafer foundry plant. It can identify different defect patterns on wafers to help readers delve into the matter.     

Wafer defect pattern recognition by multi-class support vector machines by using a novel defect cluster index

Wafer yield is an important index of efficiency in integrated circuit (IC) production. The number and cluster intensity of wafer defects are two key determinants of wafer yield. As wafer sizes increase, the defect cluster phenomenon becomes more apparent. Cluster indices currently used to describe this phenomenon have major limitations. Causes of process variation can sometimes be identified by analyzing wafer defect patterns. However, human recognition of defect patterns can be time-consuming and inaccurate. This study presents a novel recognition system using multi-class support vector machines with a new defect cluster index to efficiently and accurately recognize wafer defect patterns. A simulated case demonstrates the effectiveness of the proposed model.     

Pattern recognition of manufacturing process signals using Gaussian mixture models-based recognition systems

Unnatural patterns exhibited in manufacturing processes can be associated with certain assignable causes for process variation. Hence, accurate identification of various process patterns (PPs) can significantly narrow down the scope of possible causes that must be investigated, and speed up the troubleshooting process. This paper proposes a Gaussian mixture models (GMM)-based PP recognition (PPR) model, which employs a collection of several GMMs trained for PPR. By using statistical features and wavelet energy features as the input features, the proposed PPR model provides more simple training procedure and better generalization performance than using single recognizer, and hence is easier to be used by quality engineers and operators. Furthermore, the proposed model is capable of adapting novel PPs through using a dynamic modeling scheme. The simulation results indicate that the GMM-based PPR model shows good detection and recognition of current PPs and adapts further novel PPs effectively. Analysis from this study provides guidelines in developing GMM – based SPC recognition systems.     

Defective wafer detection using a denoising autoencoder for semiconductor manufacturing processes

Defective wafer detection is essential to avoid loss of yield due to process abnormalities in semiconductor manufacturing. For most complex processes in semiconductor manufacturing, various sensors are installed on equipment to capture process information and equipment conditions, including pressure, gas flow, temperature, and power. Because defective wafers are rare in current practice, supervised learning methods usually perform poorly as there are not enough defective wafers for fault detection (FD). The existing methods of anomaly detection often rely on linear excursion detection, such as principal component analysis (PCA), k-nearest neighbor (kNN) classifier, or manual inspection of equipment sensor data. However, conventional methods of observing equipment sensor readings directly often cannot identify the critical features or statistics for detection of defective wafers. To bridge the gap between research-based knowledge and semiconductor practice, this paper proposes an anomaly detection method that uses a denoise autoencoder (DAE) to learn a main representation of normal wafers from equipment sensor readings and serve as the one-class classification model. Typically, the maximum reconstruction error (MaxRE) is used as a threshold to differentiate between normal and defective wafers. However, the threshold by MaxRE usually yields a high false positive rate of normal wafers due to the outliers in an imbalanced data set. To resolve this difficulty, the Hampel identifier, a robust method of outlier detection, is adopted to determine a new threshold for detecting defective wafers, called MaxRE without outlier (MaxREwoo). The proposed method is illustrated using an empirical study based on the real data of a wafer fabrication. Based on the experimental results, the proposed DAE shows great promise as a viable solution for on-line FD in semiconductor manufacturing.     

An unsupervised neural network approach for automatic semiconductor wafer defect inspection

Semiconductor wafer defect inspection is an important process before die packaging. The defective regions are usually identified through visual judgment with the aid of a scanning electron microscope. Dozens of people visually check wafers and hand-mark their defective regions. Consequently, potential misjudgment may be introduced due to human fatigue. In addition, the process can incur significant personnel costs. Prior work has proposed automated visual wafer defect inspection that is based on supervised neural networks. Since it requires learned patterns specific to each application, its disadvantage is the lack of product flexibility. Self-organizing neural networks (SONNs) have been proven to have the capabilities of unsupervised auto-clustering. In this paper, an automatic wafer inspection system based on a self-organizing neural network is proposed. Based on real-world data, experimental results show, with good performance, that the proposed method successfully identifies the defective regions on wafers.     

Wavelet-based defect detection in solar wafer images with inhomogeneous texture

Solar power is an attractive alternative source of electricity. Multicrystalline solar cells dominate the market share owing to their lower manufacturing costs. The surface quality of a solar wafer determines the conversion efficiency of the solar cell. A multicrystalline solar wafer surface contains numerous crystal grains of random shapes and sizes in random positions and directions with different illumination reflections, therefore resulting in an inhomogeneous texture in the sensed image. This texture makes the defect detection task extremely difficult. This paper proposes a wavelet-based discriminant measure for defect inspection in multicrystalline solar wafer images.The traditional wavelet transform techniques for texture analysis and surface inspection rely mainly on the discriminant features extracted in individual decomposition levels. However, these techniques cannot be directly applied to solar wafers with inhomogeneous grain patterns. The defects found in a solar wafer surface generally involve scattering and blurred edges with respect to clear and sharp edges of crystal grains in the background. The proposed method uses the wavelet coefficients in individual decomposition levels as features and the difference of the coefficient values between two consecutive resolution levels as the weights to distinguish local defects from the crystal grain background, and generates a better discriminant measure for identifying various defects in the multicrystalline solar wafers. Experimental results have shown the proposed method performs effectively for detecting fingerprint, contaminant, and saw-mark defects in solar wafer surfaces.     

基于神经网络的集成电路生产过程建模与优化

以提高半导体生产线的成品率为目标,利用神经网络对半导体芯片生产过程进行了建 模和优化.首先使用神经网络方法建立模型,确定生产线上工艺参数和成品率之间的映射关系, 构造多维映射函数曲面;随后对多维映射函数曲面进行搜索,搜索成品率最高的最优点,据此确 定工艺参数的规范值;最后,根据优化后的工艺参数规范进行实际生产.采用这种优化建议,半 导体生产线的平均成品率由51.7%提高到了57.5%.