A new strategy for defect inspection by the virtual inspection in semiconductor wafer fabrication

The defect of process equipments is a major factor that impairs the yields in the mass production of semiconductor wafer fabrication and it is a main supervision means to use high-resolution defect inspection tools to detect and monitor the defect damage. Due to the high investment costs of these inspection tools and the resulting decrease in the throughput, how to improve the sampling rate is an important issue for the associated inspection strategy. This paper proposes a new concept and implementation of virtual inspection (VI) to enhance the detection and monitoring of defect in semiconductor production process. The underlying theory of the VI concept is that the state variables identifications (SVIDs) of process equipments can reflect the process quality effectively and loyally. The approach of VI is to combine the application of the fault detection and classification (FDC), and the defect library and the re-engineering of inspection procedure to reach the full-scope of strategic objective. VI enables the defect monitoring to enter a new era by promoting the monitoring level of defect inspection from the previous lot-sampling basis to the wafer-sampling level, and hence upgrades the sampling strategy from random-sampling to full and right-sampling. In this study, various typical defect cases are utilized to illustrate how to create VI models and verify the reliability of the proposed approach. Furthermore, a feasible architecture of the VI implementation for mass production in semiconductor factory is presented in the paper.     

晶圆表面微米级缺陷检测

为降低晶圆缺陷对半导体制造的影响,在基于改进的多重中值滤波算法的基础上,以差影法为基本原理,采用归一化互相关的模版匹配方法实现晶圆表面缺陷检测。改进的多重中值滤波算法有效实现噪声点与非噪声点的分辨,归一化模版匹配算法对光照具有很好的鲁棒性。对大量的晶粒进行实验,实验结果表明,该方法可有效检测出晶圆表面的缺陷,精度达到15μm左右,所提检测算法在实际的应用中可代替人工,快速、准确地实现晶圆的缺陷检测。     

An automatic wafer inspection system using pipelined imageprocessing techniques

An automatic wafer pattern inspection system has been developed that can detect defective patterns 6 μm or larger in multilayered wafer patterns at a speed 30 times faster than that of a human inspector. The false-alarm rate is less than 0.5 occurrences/chip. This performance is achieved mainly by the use of a special comparison method between two adjacent patterns obtained through a single optical setup, and also by the use of digital design pattern data (CAD data). The main functions of the design pattern data are to specify the inspection area, to designate optimum parameters for inspection, and to separate defective portions into different layers, thereby facilitating the classification of the defects. All image processing is performed in one pass by a high-speed pipeline-structured image processor that can analyze an input image signal at a 7 MHz video rate     

An intelligent system for wafer bin map defect diagnosis: An empirical study for semiconductor manufacturing

Wafer bin maps (WBMs) that show specific spatial patterns can provide clue to identify process failures in the semiconductor manufacturing. In practice, most companies rely on experienced engineers to visually find the specific WBM patterns. However, as wafer size is enlarged and integrated circuit (IC) feature size is continuously shrinking, WBM patterns become complicated due to the differences of die size, wafer rotation, the density of failed dies and thus human judgments become inconsistent and unreliable. To fill the gaps, this study aims to develop a knowledge-based intelligent system for WBMs defect diagnosis for yield enhancement in wafer fabrication. The proposed system consisted of three parts: graphical user interface, the WBM clustering solution, and the knowledge database. In particular, the developed WBM clustering approach integrates spatial statistics test, cellular neural network (CNN), adaptive resonance theory (ART) neural network, and moment invariant (MI) to cluster different patterns effectively. In addition, an interactive converse interface is developed to present the possible root causes in the order of similarity matching and record the diagnosis know-how from the domain experts into the knowledge database. To validate the proposed WBM clustering solution, twelve different WBM patterns collected in real settings are used to demonstrate the performance of the proposed method in terms of purity, diversity, specificity, and efficiency. The results have shown the validity and practical viability of the proposed system. Indeed, the developed solution has been implemented in a leading semiconductor manufacturing company in Taiwan. The proposed WBM intelligent system can recognize specific failure patterns efficiently and also record the assignable root causes verified by the domain experts to enhance troubleshooting effectively.     

Centroid neural network for unsupervised competitive learning

An unsupervised competitive learning algorithm based on the classical k-means clustering algorithm is proposed. The proposed learning algorithm called the centroid neural network (CNN) estimates centroids of the related cluster groups in training date. This paper also explains algorithmic relationships among the CNN and some of the conventional unsupervised competitive learning algorithms including Kohonen's self-organizing map and Kosko's differential competitive learning algorithm. The CNN algorithm requires neither a predetermined schedule for learning coefficient nor a total number of iterations for clustering. The simulation results on clustering problems and image compression problems show that CNN converges much faster than conventional algorithms with compatible clustering quality while other algorithms may give unstable results depending on the initial values of the learning coefficient and the total number of iterations.     

A multi-criteria approach for scheduling semiconductor wafer fabrication facilities

In this research, we model a semiconductor wafer fabrication process as a complex job shop, and adapt a Modified Shifting Bottleneck Heuristic (MSBH) to facilitate the multi-criteria optimization of makespan, cycle time, and total weighted tardiness using a desirability function. The desirability function is implemented at two different levels of the MSBH: the subproblem solution procedure level (SSP level) and the machine criticality measure level (MCM level). In addition, we suggest two different methods of choosing the critical toolgroup at the MCM level: (1) the Local MCM approach, which chooses the critical toolgroup based on local desirability values from the SSP level and (2) the Global MCM approach, which chooses the critical toolgroup based on its impact on the desirability of the entire disjunctive graph. Results demonstrate the desirability-based approaches’ ability to simultaneously minimize all three objectives.     

An unsupervised artificial neural network versus a rule-based approach for fault detection and identification in an automated assembly machine

Artificial neural networks (ANNs) are suitable for fault detection and identification (FDI) applications because of their pattern recognition abilities. In this study, an unsupervised ANN based on Adaptive Resonance Theory (ART) is tested for FDI on an automated O-ring assembly machine testbed, and its performance and practicality are compared to a conventional rule-based method. Three greyscale sensors and two redundant limit switches are used as cost-effective sensors to monitor the machine’s assembly process. Sensor data are collected while the machine is operated under normal condition, as well as 10 fault conditions. Features are selected from the raw sensor data, and data sets are created for training and testing the ANN. The performance of the ANN for detecting and identifying known, unknown and multiple faults is evaluated; the performance is compared to a conventional rule-based method using the same data sets. Results show that the ART ANN is able to achieve excellent fault detection performance with minimal modeling requirements; however, the performance depends on careful tuning of its vigilance parameter. Although the rule-based system requires more effort to set up, it is judged to be more useful when unknown or multiple faults are present. The ART network creates new outputs for unknown and multiple fault conditions, but it does not give any more information as to what the new fault is. By contrast, the rule-based method is able to generate symptoms that clearly identify the unknown and multiple fault conditions. Thus, the rule-based method is judged to be the most feasible method for FDI applications.     

A genetic approach to automatic neural network architecture optimization

The use of artificial neural networks for various problems has provided many benefits in various fields of research and engineering. Yet, depending on the problem, different architectures need to be developed and most of the time the design decision relies on a trial and error basis as well as on the experience of the developer. Many approaches have been investigated concerning the topology modelling, training algorithms, data processing. This paper proposes a novel automatic method for the search of a neural network architecture given a specific task. When selecting the best topology, our method allows the exploration of a multidimensional space of possible structures, including the choice of the number of neurons, the number of hidden layers, the types of synaptic connections, and the use of transfer functions. Whereas the backpropagation algorithm is being conventionally used in the field of neural networks, one of the known disadvantages of the technique represents the possibility of the method to reach saddle points or local minima, hence overfitting the output data. In this work, we introduce a novel strategy which is capable to generate a network topology with overfitting being avoided in the majority of the cases at affordable computational cost. In order to validate our method, we provide several numerical experiments and discuss the outcomes.

     

基于MapReduce的液晶屏缺陷检测方法

5th液晶屏在生产过程中会产生多种类型的缺陷,通过单一节点进行缺陷检测存在存储资源和计算时间的瓶颈。利用Hadoop集群的分布式计算、存储能力处理海量的高分辨率液晶屏图像是一个新的思路。针对高分辨液晶屏图像缺陷局部性特点,设计基于MapReduce的分布式缺陷检测方法,对高分辨率图像分块,并行完成每块图像的缺陷检测,再将检测结果归并,从而解决高分辨率图像缺陷检测效率低下问题。通过运行在Hadoop平台上的实验表明,该方法在完成缺陷检测的同时具有良好的效率提升。     

An evolutionary neural network approach for module orientationproblems

A novel neural network approach called “Evolutionary Neural Network (ENN)” is presented for the module orientation problem. The goal of this NP-complete problem is to minimize the total wire length by flipping circuit modules with respect to their vertical and/or horizontal axes of symmetry. In order to achieve high quality VLSI systems, it is strongly desired to solve the problem as quickly as possible in the design cycle. Based on the concept of the genetic algorithm, the evolutionary initialization scheme on neuron states is introduced so as to provide a high quality solution within a very short time. The performance of ENN is compared with three heuristic algorithms through simulations on 20 examples with up to 500 modules. The results show that ENN can find the best solutions in the shortest time     

Soldering defect detection in automatic optical inspection

This paper proposes an integrated detection framework of solder joint defects in the context of Automatic Optical Inspection (AOI) of Printed Circuit Boards (PCBs). Both localization and classifications tasks were considered. For the localization part, in contrast to the existing methods that are highly specified for particular PCBs, we used a generic deep learning method which can be easily ported to different configurations of PCBs and soldering technologies and also gives real-time speed and high accuracy. For the classification part, an active learning method was proposed to reduce the labeling workload when a large labeled training database is not easily available because it requires domain-specified knowledge. The experiments show that the localization method is fast and accurate. In addition, high accuracy with only minimal user input was achieved in the classification framework on two different datasets. The results also demonstrated that our method outperforms three other active learning benchmarks.     

A neural network approach for defect identification and classification on leather fabric

In this paper, an automated vision system is presented to detect and classify surface defects on leather fabric. Visual defects in a gray-level image are located through thresholding and morphological processing, and their geometric information is immediately reported. Three input feature sets are proposed and tested to find the best set to characterize five types of defects: lines, holes, stains, wears, and knots. Two multilayered perceptron models with one and two hidden layers are tested for the classification of defects. If multiple line defects are identified on a given image as a result of classification, a line combination test is conducted to check if they are parts of larger line defects. Experimental results on 140 defect samples show that two-layered perceptrons are better than three-layered perceptrons for this problem. The classification results of this neural network approach are compared with those of a decision tree approach. The comparison shows that the neural network classifier provides better classification accuracy despite longer training times.     

Automated defect inspection of light-emitting diode chips using neural network and statistical approaches

This research explores the automated visual inspection of surface blemishes that fall across two different background textures in a light-emitting diode (LED) chip. Water-drop defects, commonly found on chip surface, impair the appearance of LEDs as well as their functionality and security. Automated inspection of a water-drop defect is difficult because the blemish has a semi-opaque appearance and a low intensity contrast with the rough exterior of the LED chip. Moreover, the blemish may fall across two different background textures, which further increases the difficulties of defect detection. The one-level Haar wavelet transform is used to decompose a chip image and extract four wavelet characteristics. Then, wavelet-based neural network (WNN) and wavelet-based multivariate statistical (WMS) approaches are proposed individually to integrate the multiple wavelet characteristics. Finally, the back-propagation algorithm of WNN and T² test of WMS individually judge the existence of water-drop defects. Experimental results show that both of the proposed methods achieve above 95% and 92% detection rates and below 7.5% and 5.8% false alarm rates, respectively.     

An innovative neural network approach for stock market prediction

The Journal of Supercomputing - This paper aims to develop an innovative neural network approach to achieve better stock market predictions. Data were obtained from the live stock market for...     

An unsupervised learning approach based on a Hopfield-like network for assessing posterior capsule opacification

Posterior capsule opacification (PCO) is the most common complication of cataract surgery, occurring in up to 50% of patients by 2–3 years after the operation [Spalton in Eye 13(Pt 3b):489–492, 1999]. This paper proposes a new approach for the assessment of PCO digital images. The approach deploys an unsupervised learning technique for clustering image pixels into different regions based on chromatic attributes. The innovative aspect of this paper lies in proposing the number of regions in a clustered image as a measurement tool for assessing the PCO. Experiments using synthetic data confirmed the plausibility of this approach. A series of experiments conducted on real PCO images demonstrated the robustness and stability of the proposed algorithm. Finally, the comparison of our method’s assessment with medical expert evaluation reveals a very reasonable concordance.     

A two-phase dynamic dispatching approach to semiconductor wafer testing

There have been numerous advancements and rising competition in semiconductor technologies. In light of this, the wafer test plays a more significant role in providing the most prompt yield feedback for quick process improvement. However, the wafer test shop floor is getting more complicated than ever before because of the increasing change-over rate, nonlinear wafer arrival, and preemption by urgent orders. Furthermore, the foundry wafer test is a heterogeneous production with different production cycle times and a large variety of nonidentical testers. Shop floor conditions, including work in process (WIP) pool, tester status, and work order priority, continuously change. There is a need to operate the kind of production line that simultaneously fulfills multiple objectives. Such objectives are maximum confirmed line item performance (CLIP) for normal lots, 100% CLIP for urgent lots, minimum change-over rate, and shortest cycle time. Thus, a reactive dispatching approach is proposed and expected to perform a real-time solution no matter how/what the shop floor would change. The dynamic approach is mainly triggered by two kinds of major events: one is when an urgent lot comes in, and the other is when a tester is idle. In addition, through a two-phase dispatching algorithm, lot ranking, and lot assignment methods, prioritized WIP lots and an appropriate lot assignment are suggested. A better performance measure is obtained by considering the multiple objectives the wafer test operations seek to achieve.     

基于概率的无监督缺陷预测方法

软件缺陷预测能够提高软件开发和测试的效率,保障软件质量。无监督缺陷预测方法具有不需要标签数据的特点,从而能够快速应用于工程实践中。提出了基于概率的无监督缺陷预测方法—PCLA,将度量元值与阈值的差值映射为概率,使用概率评估类存在缺陷的可能性,然后再通过聚类和标记来完成缺陷预测,以解决现有无监督方法直接根据阈值判断时对阈值比较敏感而引起的信息丢失问题。将PCLA方法应用在NetGen和Relink两组数据集,共7个软件项目上,实验结果表明PCLA方法在查全率、查准率、F-measure上相对现有无监督方法分别平均提升4.1%、2.52%、3.14%。     

Image segmentation with the SOLNN unsupervised logic neural network

In this article, an image segmentation method based on the SOLNN self-organising logic neural network is studied. The input image is initially processed using the TCS texture-highlighting technique and is then presented to the SOLNN network which segments it. The SOLNN is characterised by a variable sensitivity which enables it to be fine-tuned to detect different sub-textures within each texture to the desired degree of detail. The experimental results reported here illustrate the fact that the SOLNN indeed clusters accurately the textural information so that each cluster represents a single texture even for images which are objectively very difficult to segment. Thus, it is supported that the proposed approach leads to the design of an effective texture-based image-segmentation system.