Virtual Metrology Research Articles

Confidence of Reported Information for real time in Diagnosis of Complex Discrete Events Systems: A Semiconductor application

Nowadays, Semiconductor Manufacturing is operating in an intense competitive environment. These companies are developing manufacturing solutions to improve the quality of the process which improves the production equipment effectiveness. We can cite many different directions such as virtual metrology, dynamic control plan, and maintenance, etc. In this paper, we propose to contribute to an online diagnosis method helping human operator to identify equipments at the origin of a propagated failure. Our contribution consists in providing an online confidence calculating module of the data issued from the manufacturing equipments. Bayesian networks theory and naïve algorithm are at the base of the proposed approach. We suggest its application to a semiconductor manufacturing industry.

Real-time virtual metrology and control for plasma etch

Plasma etch is a semiconductor manufacturing process during which material is removed from the surface of semiconducting wafers, typically made of silicon, using gases in plasma form. A host of chemical and electrical complexities make the etch process notoriously difficult to model and troublesome to control. This work demonstrates the use of a real-time model predictive control scheme to control plasma electron density and plasma etch rate in the presence of disturbances to the ground path of the chamber. Virtual metrology (VM) models, using plasma impedance measurements, are used to estimate the plasma electron density and plasma etch rate in real time for control, eliminating the requirement for invasive measurements. The virtual metrology and control schemes exhibit fast set-point tracking and disturbance rejection capabilities. Etch rate can be controlled to within 1% of the desired value. Such control represents a significant improvement over open-loop operation of etch tools, where variances in etch rate of up to 5% can be observed during production processes due to disturbances in tool state and material properties.

이상치 탐지 방법론을 활용한 반도체 가상 계측 결과의 신뢰도 추정

Samsung SDSThe purpose of virtual metrology (VM) in semiconductor manufacturing is to predict every wafer’s metrological values based on its process equipment data without an actual metrology. In this paper, we propose novelty detection-based reliability estimation models for VM in order to support flexible utilization of VM results. Because the proposed model can not only estimate the reliability of VM, but also identify suspicious process variables lowering the reliability, quality control actions can be taken selectively based on the reliance level and its causes. Based on the preliminary experimental results with actual semiconductor manufacturing process data, our models can successfully give a high reliance level to the wafers with small prediction errors and a low reliance level to the wafers with large prediction errors. In addition, our proposed model can give more detailed information by identifying the critical process variables and their relative impacts on the low reliability.

A novel automatic virtual metrology system architecture for TFT-LCD industry based on main memory database

The semiconductor and thin-film-transistor-liquid-crystal-display (TFT-LCD) industries widely value Automatic Virtual Metrology System (AVMS). AVMS needs to handle a large volume of VM-related data, which may cause poor internal database performance. In general, AVMS adopts efficient but expensive commercial database management systems (DBMSs) to yield good AVMS performance. This usually makes the AVMS construction cost very high. Therefore, the industries require a novel AVMS architecture with lower cost and greater efficiency in database. This paper proposes a novel AVMS architecture based on Main Memory Database (MMDB) technology. Specifically, the MMDB is used to improve the performance bottlenecks of the current Disk Resident Database (DRDB). Also, we design automatic data-backup and automatic data-query sources integration mechanisms to effectively relieve rapidly increased data volume in the original AVMS architecture. In addition, the novel AVMS architecture adopts a free commercial MMDB to significantly reduce total system cost. Integrated testing results show that the proposed AVMS architecture and developed technologies can enable the AVMS to have better data-storage efficiency, superior data-query performance, and lower database cost. The proposed AVMS architecture and research results in this paper can be a useful reference for TFT-LCD manufacturing companies in constructing their own AVM systems. The proposed AVMS architecture can also be applied in the semiconductor and solar-cell industries.

Global and Local Virtual Metrology Models for a Plasma Etch Process

Virtual metrology (VM) is the estimation of metrology variables that may be expensive or difficult to measure using readily available process information. This paper investigates the application of global and local VM schemes to a data set recorded from an industrial plasma etch chamber. Windowed VM models are shown to be the most accurate local VM scheme, capable of producing useful estimates of plasma etch rates over multiple chamber maintenance events and many thousands of wafers. Partial least-squares regression, artificial neural networks, and Gaussian process regression are investigated as candidate modeling techniques, with windowed Gaussian process regression models providing the most accurate results for the data set investigated.

Developing an Automatic Virtual Metrology System

Virtual Metrology (VM) is a method to conjecture manufacturing quality of a process tool based on data sensed from the process tool and without physical metrology operations. VM has now been designated by the International SEMATECH Manufacturing Initiative and International Technology Roadmap for Semiconductors as one of the focus areas for the next-generation factory realization roadmap of the semiconductor industry. This paper defines the VM automation levels, proposes the concept of automatic virtual metrology (AVM), and develops an AVM system for automatic and fab-wide VM deployment. The example of automatic VM model refreshing for chemical vapor deposition (CVD) tools is also illustrated in this paper. The AVM system has been successfully deployed in a fifth-generation thin-film-transistor-liquid-crystal-display (TFT-LCD) factory in Chi Mei Optoelectronics (CMO), Taiwan.

Machine learning-based novelty detection for faulty wafer detection in semiconductor manufacturing

Since semiconductor manufacturing consists of hundreds of processes, a faulty wafer detection system, which allows for earlier detection of faulty wafers, is required. statistical process control (SPC) and virtual metrology (VM) have been used to detect faulty wafers. However, there are some limitations in that SPC requires linear, unimodal and single variable data and VM underestimates the deviations of predictors. In this paper, seven different machine learning-based novelty detection methods were employed to detect faulty wafers. The models were trained with Fault Detection and Classification (FDC) data to detect wafers having faulty metrology values. The real world semiconductor manufacturing data collected from a semiconductor fab were tested. Since the real world data have more than 150 input variables, we employed three different dimensionality reduction methods. The experimental results showed a high True Positive Rate (TPR). These results are promising enough to warrant further study.

Automatic Data Quality Evaluation for the AVM System

This paper proposes the schemes of automatic process and metrology data-quality evaluations for the automatic virtual metrology (AVM) system. Firstly, principal component analysis is applied to extract data features of all the collected equipment process data; then Euclidean distance is utilized to unify all the principal components into a single index denoted by process data quality index (DQI <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">X</sub> ) for evaluating the quality of process data. Second, adaptive resonance theory 2 (ART2) and normalized variability are applied to define the metrology data quality index (DQI <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">y</sub> ) for appraising the quality of metrology data. The thresholds of both DQI <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">X</sub> and DQI <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">y</sub> are also defined and can be adaptively calculated. The DQI <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">X</sub> and DQI <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">y</sub> data quality evaluation schemes are well suited for the AVM systems of the semiconductor and thin film transistor-liquid crystal display industries to online, real-time, and automatically evaluate the quality of all the collected process and metrology data. As such, abnormal data will not be adopted for VM model training or tuning and VM conjecture accuracy can be maintained.

A virtual metrology model based on recursive canonical variate analysis with applications to sputtering process

In data driven process monitoring, soft-sensor, or virtual metrology (VM) model is often employed to predict product's quality variables using sensor variables of the manufacturing process. Partial least squares (PLS) are commonly used to achieve this purpose. However, PLS seeks the direction of maximum co-variation between process variables and quality variables. Hence, a PLS model may include the directions representing variations in the process sensor variables that are irrelevant to predicting quality variables. In this case, when direction of sensor variables’ variations most influential to quality variables is nearly orthogonal to direction of largest process variations, a PLS model will lack generalization capability. In contrast to PLS, canonical variate analysis (CVA) identifies a set of basis vector pairs which would maximize the correlation between input and output. Thus, it may uncover complex relationships that reflect the structure between quality variables and process sensor variables. In this work, an adaptive VM based on recursive CVA (RCVA) is proposed. Case study on a numerical example demonstrates the capability of CVA-based VM model compared to PLS-based VM model. Superiority of the proposed model is also presented when it applied to an industrial sputtering process.

Method for improving the accuracy of quality prediction in the manufacturing process of semiconductors

This paper describes the high accurate Virtual Metrology (VM) technique we propose, which may be applied to a mass production process of multiple products, e.g. semiconductors.VM technology is necessary in a production process where sufficient amount of inspection data is not available. By applying VM technology, quality of the product and process error can be predicted based on the process data and relatively small amount of inspection data. Most common VM technology applied to these fields of manufacturing process for quality prediction uses a multivariate statistical model. However, the accuracy of prediction models using linear multiple regression drops when, during the generation of the model, a collinearity is found between terms in the process data that forms the explanatory variables. In addition, when updating the prediction model in order to increase its prediction accuracy, a collinearity among explanatory variables causes the regression coefficient to become unstable, resulting in a degradation of the prediction accuracy.The method proposed in this paper enables us to create a quality prediction model composed of mutually uncorrelated variables and to perform prediction while updating the coefficient of each variable in the prediction model formula.Our proposed technology will solve two problems: degradation of the quality prediction accuracy due to mutual interference between multiple process data and overfitting upon update of the metrology model.

Benefit Model of Virtual Metrology and Integrating AVM Into MES

Frequent monitoring in both tool and process is required to detect the quality issue early so as to improve the process stability of semiconductor manufacturing. However, more tool and process monitoring means more metrology operation cost and longer cycle time. Recently, a promising technology denoted virtual metrology (VM) has bloomed. VM can convert sampling inspection with metrology delay into real-time and on-line total inspection. Therefore, VM has now been designated by International SEMATECH Manufacturing Initiative and International Technology Road Map for Semiconductors as one of the focus areas for the next generation factory realization road map of the semiconductor industry. The authors have developed the so-called automatic virtual metrology (AVM) system to implement and deploy the VM operations automatically. The purpose of this paper is to develop a business model to measure the profitability of VM based on in-depth manufacturing practices and metrology operations required for semiconductor manufacturing. This paper also proposes a novel manufacturing system that integrates AVM into the manufacturing execution system (MES). The interfaces among AVM, other MES components, and run-to-run (R2R) modules in the novel manufacturing system are also defined such that the total quality inspection system can be achieved and the R2R capability can be migrated from lot-to-lot control to wafer-to-wafer control.

A Virtual Metrology System for Predicting End-of-Line Electrical Properties Using a MANCOVA Model With Tools Clustering

The ability to predict end-of-line electrical properties of wafer in semiconductor manufacturing processes is critical to developing and maintaining a high yield. However, this is difficult because an advanced wafer manufacturing process consists of 300-400 steps, and in-line metrology data is only available for a few steps and for infrequently sampled wafers. Although a large amount of equipment sensor outputs are readily available for most wafers, most of the sensor variables may not be related to the end-of-line properties. Further, differences in end-of-line properties of wafers processed by tools of the same stage do not imply differences in the values of sensor variables between these tools. Thus, it is important to develop a reliable screening and model building procedure to construct a robust virtual metrology model with good generalization capability. Despite its simplicity, this approach is found to have significantly better generalization capability than nonlinear models, as well as substantial improvement in modeling and prediction capabilities of linear models that use only in-line metrology. The proposed method is also evaluated by an industrial application in a local fabrication unit.

Selection Schemes of Dual Virtual-Metrology Outputs for Enhancing Prediction Accuracy

Selection schemes between neural-network (NN) and multiple-regression (MR) outputs of a virtual metrology system (VMS) are studied in this paper. Both NN and MR are applicable algorithms for implementing virtual-metrology (VM) conjecture models. A MR algorithm may achieve better accuracy only with a stable process, whereas a NN algorithm may have superior accuracy when equipment property drift or shift occurs. To take advantage of both MR and NN algorithms, the simple-selection scheme (SS-scheme) is first proposed to enhance the VM conjecture accuracy. This SS-scheme simply selects either NN or MR output according to the smaller Mahalanobis distance between the input process data set and the NN/MR-group historical process data sets. Furthermore, a weighted-selection scheme (WS-scheme), which computes the VM output with a weighted sum of NN and MR results, is also developed. This WS-scheme generates a well-behaved system with continuity between the NN and MR outputs. Both the CVD and photo processes of a fifth-generation TFT-LCD factory are adopted in this paper to test and compare the conjecture accuracy among the solo-NN, solo-MR, SS-scheme, and WS-scheme algorithms. One-hidden-layered back-propagation neural network (BPNN-I) is applied to establish the NN conjecture model. Test results show that the conjecture accuracy of the WS-scheme is the best among those solo-NN, solo-MR, SS-scheme, and WS-scheme algorithms.

A Virtual Metrology Model Based on Recursive Canonical Variate Analysis with Applications to Sputtering Process

A Virtual Metrology Model Based on Recursive Canonical Variate Analysis with Applications to Sputtering Process

Multilevel Kernel Methods for Virtual Metrology in Semiconductor Manufacturing

AbstractIn semiconductor manufacturing, Virtual Metrology (VM) methodologies aim to obtain reliable estimates of process results without actually performing measurement operations, that are cost-intensive and time-consuming. This goal is usually achieved by means of statistical models, linking (easily collectible) process data to target measurements. In this paper, we tackle two of the most important issues in VM: function regression in high-dimensional spaces and data heterogeneity caused by inhomogeneous production and equipment logistics. We propose a hierarchical framework based on kernel methods and solved by means of multitask learning strategies and mixed-effects statistical models to improve the quality of estimates. The proposed methodology is validated on actual process and measurement data from the semiconductor manufacturing.

Real-time Virtual Metrology and Control of Plasma Electron Density in an Industrial Plasma Etch Chamber

AbstractPlasma etching is a semiconductor manufacturing process during which material is removed from the surface of silicon wafers using gases in plasma form. A host of chemical and electrical complexities make the etch process notoriously difficult to model and troublesome to control. This work demonstrates the use of a real-time model predictive control scheme to maintain a consistent plasma electron density in the presence of disturbances to the ground path of the chamber. The electron density is estimated in real time using a virtual metrology model based on plasma impedance measurements. Recursive least squares is used to update the controller model parameters in real time to achieve satisfactory control of electron density over a wide operating space.

Development of a virtual metrology for high-mix TFT-LCD manufacturing processes

Nowadays, TFT-LCD manufacturing has become a very complex process, in which many different products being manufactured with many different tools. The ability to predict the quality of product in such a high-mix system is critical to developing and maintaining a high yield. In this paper, a statistical method is proposed for building a virtual metrology model from a number of products using a high-mix manufacturing process. Stepwise regression is used to select “key variables" that really affect the quality of the products. Multivariate analysis of covariance is also proposed for simultaneously applying the selected variables and product effect. This framework provides a systematic method of building a processing quality prediction system for a high-mix manufacturing process. The experimental results show that the proposed quality prognostic system can not only estimate the critical dimension accurately but also detect potentially faulty glasses.

Virtual metrology for run-to-run control in semiconductor manufacturing

In semiconductor manufacturing processes, run-to-run (R2R) control is used to improve productivity by adjusting process inputs run by run. A process will be controlled based on information obtained during or after a process, including metrology values of wafers. Those metrology values, however, are usually available for only a small fraction of sampled wafers. In order to overcome the limitation, one can use virtual metrology (VM) to predict metrology values of all wafers, based on sensor data from production equipments and actual metrology values of sampled wafers. In this paper, we develop VM prediction models using various data mining techniques. We also develop a VM embedded R2R control system using the exponentially weighted moving average (EWMA) scheme. The experiments consist of two parts: (1) verifying VM prediction models with actual production equipments data, and (2) conducting simulations of the R2R control system. Our VM prediction models are found to be accurate enough to be directly implemented in actual manufacturing processes. The simulation results show that the VM embedded R2R control system improves productivity.

Yield Management Enhanced Advanced Process Control System (YMeAPC)—Part I: Description and Case Study of Feedback for Optimized Multiprocess Control

In today's competitive semiconductor manufacturing environment, improving fab productivity and reducing cost requires that all systems work collaboratively towards production, quality and cost targets. Equipment engineering systems (EES), including advanced process control have risen to the top as key enablers for maximizing fab productivity, however these systems have been hindered as they focus on equipment and process, rather than fab-wide, metrics. Extending EES into the yield management space allows for leveraging yield prediction information as feedback into closing the loop around the fab to better achieve productivity goals. With this added capability of yield prediction, EES can: 1) predict yield excursions as well as excursion sources (leveraging virtual metrology technology), thereby avoiding costly post mortem yield recovery activities; 2) improve EES capabilities such as maintenance management and scheduling/dispatch; and 3) utilize yield prediction information as feedback to all levels of control in the fab, from individual processes up through factory-level controllers so that processes can be continuously tuned to meet yield and device performance targets without resorting to design changes to slower products. Case studies of application of this feedback information illustrate that benefits can be achieved from straight-forward applications, however these benefits can be expanded (especially) as more complex scheduling and control solutions are implemented. Realizing this type of yield-enhanced EES solution (YMeAPC) in a cost-effective manner requires fab-wide adherence to standards and best practices for component integration, event-based control system operation, and user interface management.

Integrating support vector machine and genetic algorithm to implement dynamic wafer quality prediction system

The semiconductor and thin-film transistor liquid crystal display (TFT-LCD) industries are currently two of the most important high-tech industries in Taiwan and occupy over 40% of the global market share. Moreover, these two industries need huge investments in manufacture production equipments (PE) and own a large production scale of the global market. Therefore, how to increase the processing quality of PE to raise the production capacity has become an important issue. The statistical process control technique is usually adopted to monitor the important process parameters in the current Fabs. Furthermore, a routine check for machine or a predictive maintenance policy is generally applied to enhance the stability of process and improve yields. However, manufacturing system cannot be obtained online quality measurements during the manufacturing process. When the abnormal conditions occur, it will cause a large number of scrapped substrates and the costs will be seriously raised. In this research, a virtual metrology (VM) system is proposed to overcome those mentioned problems. It not only fulfills real-time quality measurement of each wafer, but also detects the performance degradation of the corresponding machines from the information of manufacturing processes. This paper makes four critical contributions: (1) The more principal component analysis (PCA) we used, the higher the accuracy we obtained by the kernel function approach; (2) the more support vector data description (SVDD) we used, the higher the accuracy we obtain in novelty detection module; (3) our empirical results show that genetic algorithm (GA) and incremental learning methods increase the training/learning of support vector machine (SVM) model; and (4) by developing a wafer quality prediction model, the SVM approach obtains better prediction accuracy than the radial basis function neural network (RBFN) and back-propagation neural network (BPNN) approaches. Therefore, this paper proposes that the artificial intelligent (AI) approach could be a more suitable methodology than traditional statistics for predicting the potential scrapped substrates risk of semiconductor manufacturing companies.