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.     

Evaluating the reliability level of virtual metrology results for flexible process control: a novelty detection-based approach

The purpose of virtual metrology (VM) in semiconductor manufacturing is to support process monitoring and quality control by predicting the metrological values of every wafer without an actual metrology process, based on process sensor data collected during the operation. Most VM-based quality control schemes assume that the VM predictions are always accurate, which in fact may not be true due to some unexpected variations that can occur during the process. In this paper, therefore, we propose a means of evaluating the reliability level of VM prediction results based on novelty detection techniques, which would allow flexible utilization of the VM results. Our models generate a high-reliability score for a wafer’s VM prediction only when its process sensor values are found to be consistent with those of the majority of wafers that are used in model building; otherwise, a low-reliability score is returned. Thus, process engineers can selectively utilize VM results based on their reliability level. Experimental results show that our reliability generation models are effective; the VM results for wafers with a high level of reliability were found to be much more accurate than those with a low level.     

Multi-step virtual metrology for semiconductor manufacturing: A multilevel and regularization methods-based approach

In semiconductor manufacturing, wafer quality control strongly relies on product monitoring and physical metrology. However, the involved metrology operations, generally performed by means of scanning electron microscopes, are particularly cost-intensive and time-consuming. For this reason, in common practice a small subset only of a productive lot is measured at the metrology stations and it is devoted to represent the entire lot. Virtual Metrology (VM) methodologies are used to obtain reliable predictions of metrology results at process time, without actually performing physical measurements. This goal is usually achieved by means of statistical models and by linking process data and context information to target measurements. Since semiconductor manufacturing processes involve a high number of sequential operations, it is reasonable to assume that the quality features of a given wafer (such as layer thickness and critical dimensions) depend on the whole processing and not on the last step before measurement only. In this paper, we investigate the possibilities to enhance VM prediction accuracy by exploiting the knowledge collected in the previous process steps. We present two different schemes of multi-step VM, along with dataset preparation indications. Special emphasis is placed on regression techniques capable of handling high-dimensional input spaces. The proposed multi-step approaches are tested on industrial production data.     

Virtual metrology and feedback control for semiconductor manufacturing processes using recursive partial least squares

Virtual metrology (VM) is the prediction of metrology variables (either measurable or non-measurable) using process state and product information. In the past few years VM has been proposed as a method to augment existing metrology and has the potential to be used in control schemes for improved process control in terms of both accuracy and speed. In this paper, we propose a VM based approach for process control of semiconductor manufacturing processes on a wafer-to-wafer (W2W) basis. VM is realized by utilizing the pre-process metrology data and more importantly the process data from the underlying tools that is generally collected in real-time for fault detection (FD) purposes. The approach is developed for a multi-input multi-output (MIMO) process that may experience metrology delays, consistent process drifts, and sudden shifts in process drifts. The partial least squares (PLS) modeling technique is applied in a novel way to derive a linear regression model for the underlying process, suitable for VM purposes. A recursive moving-window approach is developed to update the VM module whenever metrology data is available. The VM data is then utilized to develop a W2W process control capability using a common run-to-run control technique. The proposed approach is applied to a simulated MIMO process and the results show considerable improvement in wafer quality as compared to other control solutions that only use lot-to-lot metrology information.     

基于动态MANCOVA的混合制程虚拟测量算法

在晶圆/液晶面板等批次加工过程中,产品质量的及时估计与品质管制是提高产能和降低成本的有效途径.针对"少量多样"的混合制程,利用逐步回归算法挑选该制程的关毽变量,引入产品的效益因子,建立混合制程的虚拟测量模型;为克服系统扰动对模型精度的影响,以产品效益因子为状态量建立该制程的状态方程,利用Kalman滤波器递归估计模型参数得到动态的MANCOVA模型;最后通过某湿式蚀刻制程的工程应用验证了该算法的有效性.     

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.     

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).     

Knowledge acquisition from batch semiconductor manufacturing data

Semiconductor manufacturing data consist of the processes and the machines involved in the production of batches of semiconductor circuit wafers. Wafer quality depends on the manufacturing line status and it is measured at the end of the line. We have developed a knowledge discovery system that is intended to help the yield analysis expert by learning the tentative causes of low quality wafers from an exhaustive amount of manufacturing data. The yield analysis expert, by using the knowledge discovered, will decide on which corrective actions to perform on the manufacturing process. This paper discusses the transformations carried out within the data from raw data to discovered knowledge, and also the two main tasks performed by the system. The features of the inductive algorithm performing those tasks are also described. Yield analysis experts at Lucent Technologies, Bell Labs Innovations in Spain are currently using this knowledge discovery application.     

测量时延在线估计与批间控制器协同设计

批间控制是半导体批次生产过程中常用算法,其关键问题在于能够及时获取上一批次的制程输出,受测量手段及其成本限制,实际的生产制程很难满足这一要求.为此,本文提出一种基于贝叶斯统计分析的测量时延估计算法.在分析晶圆质量与实测时延、估计时延、以及制程漂移之间的逻辑关系的基础上,并将晶圆的质量信息按加工时间顺序划分两个相邻的滚动时间窗口.基于贝叶斯后验概率函数,及时捕获后一个滚动时间窗口内过程输出发生漂移的概率,从而判断是否有测量时延发生,并估算该时延大小.在此基础上,给出批间控制器的测量时延补偿策略,及时调整制程的控制量,提高晶圆的加工品质.仿真结果验证所提出算法的有效性.     

A neural network-based approach for dynamic quality prediction in a plastic injection molding process

This paper presents an innovative neural network-based quality prediction system for a plastic injection molding process. A self-organizing map plus a back-propagation neural network (SOM-BPNN) model is proposed for creating a dynamic quality predictor. Three SOM-based dynamic extraction parameters with six manufacturing process parameters and one level of product quality were dedicated to training and testing the proposed system. In addition, Taguchi’s parameter design method was also applied to enhance the neural network performance. For comparison, an additional back-propagation neural network (BPNN) model was constructed for which six process parameters were used for training and testing. The training and testing data for the two models respectively consisted of 120 and 40 samples. Experimental results showed that such a SOM-BPNN-based model can accurately predict the product quality (weight) and can likely be used for various practical applications.     

An integrated advanced process control framework using run-to-run control,virtual metrology and fault detection

This paper develops a new advanced process control (APC) system for the multiple-input multiple-output (MIMO) semiconductor processes using the partial least squares (PLS) technique to provide the run-to-run control with the virtual metrology data, via the gradual mode or the rapid mode depending on the current system status, in order to deal with metrology delays and compensate for different types of system disturbances. First, we present a controller called the PLS-MIMO double exponentially weighted moving average (PLS-MIMO DEWMA) controller. It employs the PLS method as the model building/estimation technique to help the DEWMA controller generate more consistent and robust control outputs than purely using the conventional DEWMA controller. To cope with metrology delays, the proposed APC system uses the pre-processing metrology data to build up the virtual metrology (VM) system that can provide the estimated process outputs for the PLS-MIMO DEWMA controller. Lastly, the Fault Detection (FD) system is added based upon the principal components of the PLS modeling outcomes, which supplies the process status for the VM mechanism and the PLS-MIMO DEWMA controller as to how the process faults are responded. Two scenarios of the simulation study are conducted to illustrate the APC system proposed in this paper.     

Intelligent monitoring and diagnosis of manufacturing processes using an integrated approach of KBANN and GA

In many manufacturing processes, some key process parameters (i.e., system inputs) have very strong relationship with the categories (e.g., normal or various faulty products) of finished products (i.e., system outputs). The abnormal changes of these process parameters could result in various categories of faulty products. In this paper, a hybrid learning-based model is developed for on-line intelligent monitoring and diagnosis of the manufacturing processes. In the proposed model, a knowledge-based artificial neural network (KBANN) is developed for monitoring the manufacturing process and recognizing faulty quality categories of the products being produced. In addition, a genetic algorithm (GA)-based rule extraction approach named GARule is developed to discover the causal relationship between manufacturing parameters and product quality. These extracted rules are applied for diagnosis of the manufacturing process, provide guidelines on improving the product quality, and are used to construct KBANN. Therefore, the seamless integration of GARule and KBANN provides abnormal warnings, reveals assignable cause(s), and helps operators optimally set the process parameters. The proposed model is successfully applied to a japing-line, which improves the product quality and saves manufacturing cost.     

Optimum process mean and manufacturing quantity settings for serial production system under the quality loss and rectifying inspection plan

In 1994, Al-Sultan presented a single sampling plan applied in determining the optimum process mean for two machines in a serial production system. However, Al-Sultan did not consider the quality cost for the product within the specification limits, pointed out that the non-conforming items in the sample of accepted lot is replaced or eliminated from the lot, and proposed an integrated model with production and quality. In this study, the author considers the problem of quality loss for the modified Al-Sultan’s model with k machines in a serial production system based on a single sampling rectifying inspection plan. Taguchi’s symmetric quadratic quality loss function is applied in evaluating the product quality. Then, the author further proposes a modified and integrated economic manufacturing quantity (EMQ) model based on the application of the modified Al-Sultan’s model for obtaining the maximum expected total profit of product per unit of time. The numerical results show that the price of an accepted products sold has the most important effect on both the process means and the expected total profit per unit of time.     

Semiconductor manufacturing process control and monitoring: A fab-wide framework

The semiconductor industry has started the technology transition from 200 mm to 300 mm wafers to improve manufacturing efficiency and reduce manufacturing cost. These technological changes present a unique opportunity to optimally design the process control systems for the next generation fabs. In this paper we first propose a hierarchical fab-wide control framework with the integration of 300 mm equipment and metrology tools and highly automated material handling system. Relevant existing run-to-run technology is reviewed and analyzed in the fab-wide control context. Process and metrology data monitoring are discussed with an example. Missing components are pointed out as opportunities for future research and development. Concluding remarks are given at the end of the paper.     

Cycle forecasting EWMA (CF-EWMA) approach for drift and fault in mixed-product run-to-run process

In semiconductor manufacturing processes, mixed-products are usually fabricated on the same set of process tool with different recipes. Run-to-run controllers which based on the exponential weighted moving average (EWMA) statistic are probably the most frequently used in industry for the quality control of certain semiconductor manufacturing process steps. However, for mixed-product drifted process, if the break length of a product is large, then the process output at the beginning runs of each cycle will far deviate from the target value which will lead to a possible high rework rate and lots of waste wafers. Therefore, this study aims to develop a new approach named cycle forecasting EWMA (CF-EWMA) approach to deal with the problem of large deviations in the first few runs of each cycle. Furthermore, a common fault, i.e., the step fault, is also considered in this paper, and fault tolerant cycle forecasting EWMA (FTCF-EWMA) approach is proposed. Simulation study shows that the proposed approaches are effective.     

Intelligent remote monitoring and diagnosis of manufacturing processes using an integrated approach of neural networks and rough sets

This research develops a methodology for the intelligent remote monitoring and diagnosis of manufacturing processes. A back propagation neural network monitors a manufacturing process and identifies faulty quality categories of the products being produced. For diagnosis of the process, rough set is used to extract the causal relationship between manufacturing parameters and product quality measures. Therefore, an integration of neural networks and a rough set approach not only provides information about what is expected to happen, but also reveals why this has occurred and how to recover from the abnormal condition with specific guidelines on process parameter settings. The methodology is successfully implemented in an Ethernet network environment with sensors and PLC connected to the manufacturing processes and control computers. In an application to a manufacturing system that makes conveyor belts, the back propagation neural network accurately classified quality faults, such as wrinkles and uneven thickness. The rough set also determined the causal relationships between manufacturing parameters, e.g., process temperature, and output quality measures. In addition, rough set provided operating guidelines on specific settings of process parameters to the operators to correct the detected quality problems. The successful implementation of the developed methodology also lays a solid foundation for the development of Internet-based e-manufacturing.     

Virtual metrology modeling of time-dependent spectroscopic signals by a fused lasso algorithm

This paper proposes a fused lasso model to identify significant features in the spectroscopic signals obtained from a semiconductor manufacturing process, and to construct a reliable virtual metrology (VM) model. Analysis of spectroscopic signals involves combinations of multiple samples collected over time, each with a vast number of highly correlated features. This leads to enormous amounts of data, which is a challenge even for modern-day computers to handle. To simplify such complex spectroscopic signals, dimension reduction is critical. The fused lasso is a regularized regression method that performs automatic variable selection for the predictive modeling of highly correlated datasets such as those of spectroscopic signals. Furthermore, the fused lasso is especially useful for analyzing high-dimensional data in which the features exhibit a natural order, as is the case in spectroscopic signals. In this paper, we conducted an experimental study to demonstrate the usefulness of a fused lasso-based VM model and compared it with other VM models based on the lasso and elastic-net models. The results showed that the VM model constructed with features selected by the fused lasso algorithm yields more accurate and robust predictions than the lasso- and elastic net-based VM models. To the best of our knowledge, ours is the first attempt to apply a fused lasso to VM modeling.     

Towards a generic distributed and collaborative digital manufacturing

A framework for distributed manufacturing is proposed to facilitate collaborative product development and production among geographically distributed functional agents using digitalized information. Considering the complexity of products created in a distributed manufacturing scenario, it often requires close collaborations among a number of facilities. In this research work, various functional agents, such as the manufacturability evaluation agent (MEA), manufacturing resource agent (MRA), process-planning agent (PPA), manufacturing scheduling agent (MSA), shop floor agent (SFA), fault diagnosis agent (FDA), etc., can interact coherently for distributed manufacturing. With specific agents having unique functionalities, a manufacturing managing agent (MMA) acts as the centre of this distributed manufacturing system. The MMA agent assists the specific agents’ to work seamlessly and also to collaborate closely with the participating agents. In this way, the production cycle of a part can be optimized from product design to final manufacturing since all the production procedures are considered logically and every procedure is correlated. The agent language based on the knowledge query manipulation language (KQML) includes many pre-defined performatives that ease the participating agents to carry out their tasks intelligently by interpreting commands from one another. Additionally, to ensure the adaptiveness and upgradeability of the system, the internal structure of each functional agent that is based on JATLite is modularized into several components, including a communication interface, central work engine, knowledge base pool, and input/output modifier for possible future methodology enhancements.     

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.     

Continuous prediction of manufacturing performance throughout the production lifecycle

We describe methods for continual prediction of manufactured product quality prior to final testing. In our most expansive modeling approach, an estimated final characteristic of a product is updated after each manufacturing operation. Our initial application is for the manufacture of microprocessors, and we predict final microprocessor speed. Using these predictions, early corrective manufacturing actions may be taken to increase the speed of expected slow wafers (a collection of microprocessors) or reduce the speed of fast wafers. Such predictions may also be used to initiate corrective supply chain management actions. Developing statistical learning models for this task has many complicating factors: (a) a temporally unstable population (b) missing data that is a result of sparsely sampled measurements and (c) relatively few available measurements prior to corrective action opportunities. In a real manufacturing pilot application, our automated models selected 125 fast wafers in real-time. As predicted, those wafers were significantly faster than average. During manufacture, downstream corrective processing restored 25 nominally unacceptable wafers to normal operation.