=================================================================================
Wafer map analysis, on both single pattern and mixed-type pattern maps, is one of the most critical steps for monitoring wafer quality and tracking failures in the semiconductor manufacturing process. Defective dies on wafer bin maps (WBMs) usually cluster into specific spatial patterns, which contain critical information for root cause identification and yield improvement.
Systematic failures are a significant contributor to yield
loss in the modern Interstitial Cystitis (IC)
wafer fabrication
process [16]. As a result, identifying systematic failures
(defect clusters) has emerged as one of the top challenges in
the modern IC industry [17]. In the IC design process, the process of existing wafer test is the main factor of production cost. The test process could take an additional amount of time and
cannot make adjustments of the process immediately. The defect patterns, reflecting failure mechanisms, on wafer bin map are related to yield degradation. Fail bit maps (FBMs) represent the failed cell count during
wafer functional probe tests and have been popularly used as one of the diagnosis tools in semiconductor manufacturing for process monitoring, root cause analysis, and yield improvements. The location of the
failure point is graphically represented in the feature of the wafer maps. Most
companies control the manufacturing processes which occur to any critical defects by identifying the maps so
that it is important to classify the patterns accurately. Therefore, wafer map
defect recognition is an import part of semiconductor production. Lots of information in wafer maps can quickly help
engineers to identify what failure type it is. In this case, the engineers inspect the maps directly. However, it is
difficult to check many wafers one by one because of the increasing demand for semiconductors.
In some cases, WBMs (wafer bin map) exhibit the
signature patterns of specific processes causing defects [29-30]. Therefore, comparing the process histories of similar failures
in a database may provide sufficient information to perform
defect diagnosis and root cause analysis without predetermined
classes [31]. Defect clusters are
consistent and repeated failure patterns on wafers caused by
systematic faults in the manufacturing process [18]. Such faults are
process-related phenomena of different types, which may,
for instance, be attributed to process excursions, handling
errors, faulty equipment and/or insufficient quality materials
[19][20]. Neighbour clustering and image processing are the
common techniques used to obtain features
on each specific defect cluster. [21-23] The wafer map failure patterns can be classified by:
i) Center pattern. It is a block of defect near
the central area of a wafer.
ii) Donut pattern. It is a hollow and block defects located within the wafer.
iii) Edge-loc pattern. It is systematic defects with cluster at the wafer edge.
iv) Edge-ring pattern. It is systematic defects with moon shape at the wafer edge.
v) Loc pattern. It is a cluster
defect within the wafer.
vi) Near-full pattern. The defects cover most of the wafer.
vii) Random pattern. A small number of defective areas are located on a wafer randomly.
viii) Scratch pattern. It is a defect in a straight line or curve.
ix) None pattern. In the no
systematic pattern, the pattern was caused by
random particles falling on a wafer, and defects are randomly
distributed.
Classification of semiconductor defect clusters can be
achieved by applying techniques such as neural networks,
rule induction, decision trees, Hough Transform and
so on. [24-27] Figure 4271a shows the frequency of publications, from year 2001 to 2020, clearly indicating the rise in research interest in this
domain [13]. This graph is extracted from SCOPUS database for keyword “wafer defect classification”
and filtered by selecting the articles published in engineering and computer science domain resulting in
total 167 articles [14].
![Number of “wafer defect classification” documents published in past 20 years in SCOPUS database](images/4271e.png)
Figure 4271a. Number of “wafer defect classification” documents published in past 20 years in SCOPUS
database. [15] |
Wafer map failure pattern recognition (WMFPR) is also called as wafer failure pattern detection (WFPD). The WMFPR plays a key role in preventing yield
loss excursion events for semiconductor manufacturing. The approach has mainly been classified into two branches:
i) Model-based pattern recognition. In this method, a pre-defined probability distribution function is used for each wafer pattern and the best statistical model
is determined by comparing the models by using information criterion.
ii) Feature extraction-based pattern recognition. In this method, a large amount of pioneered suggestions [1-7] have been presented. However, the feature design or extraction has a significant impact on the accuracy of failure pattern classification. These feature extraction-based works can be further divided into three categories:
ii.a) Manually-crafted feature-driven. The domain knowledge of expert engineers is needed when manually designing the feature.
ii.b) Automatic feature extraction-based, e.g. deep learning methods, especially deep convolutional neural networks (CNNs). For instance, Wu et al. [1] develop radon-based and geometry-based features, while Yu et al. [2] define geometrical, gray, texture, and projection features, and merge the features by their proposed dimension reduction and feature extraction methods. [4-7]
ii.c) Combination between manual and automatic methods. For instance, some of the manually-crafted feature-based methods, assigned to experienced engineers, work in an unsupervised fashion. [3]
ii.d) Nonparametric Bayesian clustering based on the Dirichlet process Gaussian mixture model, and hierarchical clustering based on the
symmetric Kullback–Leibler divergence. [28]
![Wafer Bin Map](images/4271f.png)
Figure 4271b. General similarity wafer search and wafer map failure pattern recognition.
Figure 4271c shows a proposed approach used in WMFPR study. In the data processing, with WM-811K dataset, which Jo and Lee had [10], each pixel on the colored images has only three types of value. If the pixel has 0 value, the pixel is out of the wafer. If the pixel has one value, the pixel represents a normal die. If the pixel has 2 values, the pixel represents defect die. These values that pixels have are categorical variables. Therefore, they changed the number of channels in images to 3, each channel represents out of the wafer, normal die, and defect die, like one-hot encoding. To match the size of the images since the images in the dataset
have various sizes. they changed the image size to 64 by 64.
![Proposed approach used in WMFPR study](images/4271c.png)
Figure 4271c. Proposed approach used in WMFPR study. [9]
It is important to note that defect categories have different frequencies of occurrence and many classes lack sufficient associated wafter maps, resulting in a biased classifier where the decision boundary can be extremely altered by the majority classes. Therefore, an inaccurate defect classification result would be reported by the poorly calibrated model.
Figure 4271d shows the flowchart of WMFPR which is based on a two-stage framework. Stage 1 determines whether a wafer map exhibits a failure pattern, while Stage 2 identifies the pattern type. In this flowchart, an SVM (support-vector machines) is used as a classifier. During the training phase, the SVMs determine the support vectors in the
training data, which are applied to predict new wafer maps during the test phase. The main advantage of the two-stage framework is that the parameters can be trimed to optimize the tradeoff between the false-positive rate and the false-negative rate at Stage 1.
![Flowchart of the proposed WMFPR](images/4269a.png)
Figure 4271d. Flowchart of the proposed WMFPR. Stage 1: the SVM determines whether a failure pattern exists. Stage 2: the SVM identifies the wafer map failure pattern. [1] |
In the WMFPR, false-positive (FP) is defined as the rate of misclassifying Nonpattern as Pattern, whereas false-negative (FN) is the rate of misclassifying Pattern as Nonpattern. The detection error tradeoff (DET) curve can be obtained. Figure 4271e (a) shows the combined confusion matrix for the test set (overall accuracy = 94.63%). In the figure, the annotations (ground truth) are shown in the left column, and the predictions by the proposed system are in the top row. The diagonal elements represent the recognition rate of each type. It shows that Local was frequently confused with other failure types. Figure4271e (b) shows several wafer maps that were misclassified as Local. The wafer maps were misclassified, but users generally accept the prediction because these wafer maps seem to saddle across the boundary of two types. This indicates that the users' satisfaction would likely be higher, as indicated by the overall accuracy (94.63%).
Figure 4271e. (a) Combined confusion matrix on the test set.
(b) Wafer maps are easily confused with local, although users generally accept
the predictions because these wafer maps seem to saddle across the boundary
of two types. [1] |
Table 4271. Patterns commonly reflecting specific process
failure information.
Pattern |
Failure information |
Center |
The mechanical polishing is
uneven, or the pressure of the liquid is abnormal [8] |
Edge-Ring |
Abnormal temperature control during
annealing [8] |
Scratch |
Indicates an exception in the moving or cutting processes [8] |
None pattern |
Contains defective grains with random distribution. This is caused by cleaning problems in
cleaning rooms, which are expensive to eliminate completely; therefore, these defective grains
are often considered noise [8] |
Figures 4271f and 4271g show a typical wafer pattern maps from WM811K dataset and STMicroelectronics dataset.
![The structure of the WM-811K dataset](images/4236b.png)
Figure 4271f. Sample wafer examples for different pattern types. [12]
![An example of WDM for each of the class in STMicroelectronics Dataset](images/4271d.png)
Figure 4271g. An example of WDM for each of the class in STMicroelectronics dataset. [11] |
============================================
[1] Ming-Ju Wu, Jyh-Shing R. Jang, and Jui-Long Chen, Wafer Map Failure Pattern Recognition and Similarity Ranking for Large-Scale Data Sets, IEEE
Transactions on Semiconductor Manufacturing (TSM), 28(1), pp. 1–12, 2015.
[2] J. Yu and X. Lu, “Wafer map defect detection and recognition using joint local and nonlocal linear discriminant analysis,” IEEE Transactions on Semiconductor Manufacturing (TSM), vol. 29, no. 1, pp. 33–43, 2016.
[3] M. B. Alawieh, F. Wang, and X. Li, “Identifying wafer-level systematic
failure patterns via unsupervised learning,” IEEE Transactions on
Computer-Aided Design of Integrated Circuits and Systems (TCAD),
vol. 37, no. 4, pp. 832–844, 2017.
[4] T. Nakazawa and D. V. Kulkarni, “Wafer map defect pattern classification and image retrieval using convolutional neural network,” IEEE Transactions on Semiconductor Manufacturing (TSM), vol. 31, no. 2,
pp. 309–314, 2018.
[5] N. Yu, Q. Xu, and H. Wang, “Wafer defect pattern recognition and
analysis based on convolutional neural network,” IEEE Transactions
on Semiconductor Manufacturing (TSM), vol. 32, no. 4, pp. 566–573,
2019.
[6] T. Nakazawa and D. V. Kulkarni, “Anomaly detection and segmentation
for wafer defect patterns using deep convolutional encoder–decoder
neural network architectures in semiconductor manufacturing,” IEEE
Transactions on Semiconductor Manufacturing (TSM), vol. 32, no. 2,
pp. 250–256, 2019.
[7] M. B. Alawieh, D. Boning, and D. Z. Pan, “Wafer map defect patterns
classification using deep selective learning,” in ACM/IEEE Design
Automation Conference (DAC), 2020, pp. 1–6.
[8] Qiao Xu, Naigong Yu and Firdaous Essaf, Improved Wafer Map Inspection Using Attention Mechanism and Cosine Normalization, Machines, https://doi.org/10.3390/machines10020146, 10, 146, 2022.
[9] Ashadullah Shawon, Md Omar Faruk, Masrur Bin Habib, Abdullah Mohammad Khan, Silicon Wafer Map Defect Classification Using Deep Convolutional Neural Network With Data Augmentation, 2019 IEEE 5th International Conference on Computer and Communications (ICCC 2019).
[10] Ha Young Jo and Seong-Whan Lee, One-Class Classification for Wafer Map using Adversarial Autoencoder with DSVDD Prior, https://doi.org/10.48550/arXiv.2107.08823, Machine Learning, 2021.
[11] Roberto di Bella, Diego Carrera, Beatrice Rossi, Pasqualina Fragneto and Giacomo Boracchi, Wafer Defect Map Classification Using Sparse Convolutional Networks, International Conference on Image Analysis and Processing, Image Analysis and Processing – ICIAP 2019 pp 125–136.
[12] Jie Gong and Chen Lin, Wafer Map Failure Pattern Classification Using Deep Learning, 2019.
[13] Park S, Jang J and Kim C O 2021 Discriminative feature learning and cluster-based defect label
reconstruction for reducing uncertainty in wafer bin map labels J. Intell. Manuf. 32 pp 251–263.
[14] Wafer Defect Classification 2020 Available online: http://www.scopus.com
[15] Krishan Kumar Chauhan, Garima Joshi, Manjeet Kaur and Renu Vig, Semiconductor wafer defect classification using convolution neural network: a binary case, Materials Science and Engineering, 1225, 012060, 2022.
[16] J.H. Yeh and Allen Park, "Novel Technique to Identify
Systematic and Random Defects during 65 nm and
45nm Process Development for Faster Yield Learning,"
in IEEE/SEMI Advanced Semiconductor
Manufacturing Conference, 2007, p. 54.
[17] International Technology Roadmap for
Semiconductors, "Yield Enhancement," Semiconductor
Industry Association, 2009.
[18] N.G. Shankar and Z.W. Zhong, "rule-based computing
approach for the segmentation of semiconductor
defects," Microelectronics Journal, vol. 37, p. 500,
2006.
[19] F. Di Palma, G. De Nicolao, G. Miraglia, and O.M. Donzelli, "Process Diagnosis via Electrical-Wafer-Sorting Maps Classification," in Fifth IEEE International Conference on Data Mining, 2005.
[20] M. P.-L. Ooi et al., "Getting More From the
Semiconductor Test: Data Mining With Defect-Cluster
Extraction," IEEE Transactions on Instrumentation and
Measurement, vol. PP, no. 99, p. 1, March 2011.
[21] Young-Seon Jeong, Seong-Jun Kim, and M.K. Jeong,"Automatic Identification of Defect Patterns in
Semiconductor Wafer Maps Using Spatial Correlogram
and Dynamic Time Warping," IEEE Transactions on
Semiconductor Manufacturing, vol. 21, no. 4, p. 625,
Nov 2008.
[22] Chenn-Jung Huang, Chi-Feng Wu, and Chua-Chin
Wang, "Image processing techniques for wafer defect
cluster identification," Design & Test of Computers,
IEEE , vol. 19, no. 2, p. 44, Mar 2002.
[23] Chih-Hsuan Wang, "Recognition of Semiconductor
Defect Patterns Using Spectral Clustering," in IEEE
International Conference on Industrial Engineering
and Engineering Management, Singapore, 2007, pp.
588-591.
[24] Fei-Long Chen and Shu-Fan Liu, "A Neural-Network
Approach To Recognize Defect Spatial Pattern In
Semiconductor Fabrication," IEEE Transactions on
Semiconductor Manufacturing, vol. 13, no. 3, p. 366, Aug 2000.
[25] Chenn-Jung Huang, "Application of neural networks
and filtered back projection to wafer defect cluster
identification," in Proceedings of the 4th International
Symposium on Electronic Materials and Packaging,
2002, p. 99.
[26] Jian Wei Cheng, M.P.-L. Ooi, C. Chan, Ye Chow
Kuang, and S. Demidenko, "Evaluating the
Performance of Different Classification Algorithms for
Fabricated Semiconductor Wafers," in Fifth IEEE
International Symposium on Electronic Design, Test
and Application, Ho Chi Minh City, 2010, p. 360.
[27] K.P. White, B. Kundu, and C.M. Mastrangelo,"Classification of Defect Clusters on Semiconductor
Wafers Via the Hough Transformation," IEEE
Transactions on Semiconductor Manufacturing, vol.
21, no. 2, p. 272, May 2008.
[28] Jea Hoon Lee, Il-Chul Moon and Rosy Oh, Similarity Search on Wafer Bin Map Through Nonparametric and Hierarchical Clustering, IEEE Transactions on Semiconductor Manufacturing, 34(4), DOI: 10.1109/TSM.2021.3102679, (2021).
[29] S.-C. Hsu and C.-F. Chien, “Hybrid data mining approach for pattern
extraction from wafer bin map to improve yield in semiconductor
manufacturing,” International Journal of Production Economics, vol.
107, no. 1, pp. 88–103, 2007.
[30] F.-L. Chen and S.-F. Liu, “A neural-network approach to recognize defect spatial pattern in semiconductor fabrication,” IEEE Trans. Semicond. Manuf., vol. 13, no. 3, pp. 366–373, 2000.
[31] J. Hwang and H. Kim, “Variational deep clustering of wafer map
patterns,” IEEE Transactions on Semiconductor Manufacturing, vol. 33,
no. 3, pp. 466–475, 2020.
|