Semiconductor Manufacturing Systems Research Articles

Multitask learning for virtual metrology in semiconductor manufacturing systems

Abstract Virtual metrology (VM) estimates the real metrology of wafers from process data collected from multiple chambers. In semiconductor manufacturing, independent models for each process chamber are limited because the number of sampled wafers measured at each chamber are too few to build a reliable model. One potential solution to this problem is to pool the data from all chambers to create a model capable of learning and serving as a global predictive model. However, even with chambers that perform the same operation, the condition of their semiconductor tools may vary because of various factors. This study uses, for the first time, various multitask methods to develop VM models. By learning multiple related tasks simultaneously, multitask methods effectively increase the number of observations included in the prediction model. In addition, by identifying the related task, the method can make a prediction using only similar tasks. This property of multitask learning can be useful to account for lack of information in a single chamber and for diversity among the chambers. The experimental results indicate that multitask models consistently outperformed independent and pooled models regardless of the size of the training set used. Among the multitask methods, a multitask tree-based ensemble model outperformed the others in every case. This implies that the problem of wafer quality prediction can be better addressed with a form of multitask learning.

Scheduling Cluster Tools in Semiconductor Manufacturing: Recent Advances and Challenges

Cluster tools are automated robotic manufacturing systems containing multiple computer-controlled process modules. They have been increasingly used for wafer fabrication. This paper reviews the modeling and scheduling methods for cluster tools with both nonrevisiting and revisiting processes. For nonrevisiting processes, we focus on the modeling and scheduling problems of cluster tools with different constraints. Then, their solution methods are reviewed and compared. For revisiting processes, this paper first discusses the scheduling problem of some general manufacturing systems with revisiting. Then, the modeling and scheduling methodologies used to solve the scheduling problems of cluster tools with revisiting processes are reviewed. Future research directions and conclusions are finally discussed. Note to Practitioners —Semiconductor manufacturing systems are among the most advanced and complicated manufacturing systems. Their key equipment is highly automated robot-based cluster tools. With wafer residency time constraints, wafer revisiting, activity time variation, chamber cleaning requirements, and failure-prone process modules (PMs), it is very challenging to schedule and control them. This paper surveys their modeling and scheduling methods. Scheduling them requires one to schedule their robot tasks and processing activities simultaneously. Owing to wafer residency time constraints and the lack of buffers among PMs, it is difficult to conduct their optimal scheduling. This paper presents a thorough review of the state-of-the-art research results about modeling and optimal scheduling of clusters tools and indicates the future research directions.

An Imperialist Competitive Algorithm Incorporating Remaining Cycle Time Prediction for Photolithography Machines Scheduling

Photolithography machines are the common bottleneck in the semiconductor manufacturing system. The operation constraints in photolithography machines are very complicated, including wafers arriving over time, dedicated machine constraints for critical layers, auxiliary resources’ constraints, and dynamic manufacturing environment. In previous studies, the dynamic manufacturing environment has never been considered, which would make remaining cycle time seriously deviate from the expected value, and then result in the deterioration of scheduling performance. In this paper, an imperialist competitive algorithm incorporating remaining cycle prediction is proposed for photolithography machines’ scheduling problem with the objective of total completion time minimization. A deep autoencoder neural network is presented at first to predict remaining cycle time, responding to the environmental changes. Secondly, an imperialist competitive algorithm in the framework of a rolling horizon strategy is proposed to address the scheduling problem, incorporated with the accurately predicted remaining cycle time. Several procedures are designed to improve the performance of the algorithm. To verify the proposed algorithm, a simulation model of a semiconductor manufacturing system is constructed and numerical tests are conducted in the model. Results show that the algorithm proposed can significantly decrease wafers’ average cycle time.

Novel slack-based robust scheduling rule for a semiconductor manufacturing system with uncertain processing time

Novel slack-based robust scheduling rule for a semiconductor manufacturing system with uncertain processing time

An improved imperialist competitive algorithm based photolithography machines scheduling

: Photolithography machine is one of the most expensive equipment in semiconductor manufacturing system, and as such is often the bottleneck for processing wafers. This paper focuses on photolithography machines scheduling with the objective of total completion time minimisation. In contrast to classic parallel machines scheduling, it is characterised by dynamical arrival wafers, re-entrant process flows, dedicated machine constraints and auxiliary resources constraints. We propose an improved imperialist competitive algorithm (ICA) within the framework of a rolling horizon strategy for the problem. We develop a variable time interval-based rolling horizon strategy to decide the scheduling point. We address the global optimisation in every local scheduling by proposing a mixed cost function. Moreover, an adaptive assimilation operator and a sociopolitical competition operator are used to prevent premature convergence of ICA to local optima. A chaotic sequence-based local search method is presented to accelerate the rate of convergence. Computational experiments are carried out comparing the proposed algorithm with ILOG CPLEX, dispatching rules and meta-heuristic algorithms in the literature. It is observed that the algorithm proposed shows an excellent behaviour on cycle time minimisation while with a good on time delivery rate and machine utilisation rate.

Quantile regression metamodeling: Toward improved responsiveness in the high-tech electronics manufacturing industry

Both technology and market demands within the high-tech electronics manufacturing industry change rapidly. Accurate and efficient estimation of cycle-time (CT) distribution remains a critical driver of on-time delivery and associated customer satisfaction metrics in these complex manufacturing systems. Simulation models are often used to emulate these systems in order to estimate parameters of the CT distribution. However, execution time of such simulation models can be excessively long limiting the number of simulation runs that can be executed for quantifying the impact of potential future operational changes. One solution is the use of simulation metamodeling which is to build a closed-form mathematical expression to approximate the input–output relationship implied by the simulation model based on simulation experiments run at selected design points in advance. Metamodels can be easily evaluated in a spreadsheet environment “on demand” to answer what-if questions without needing to run lengthy simulations. The majority of previous simulation metamodeling approaches have focused on estimating mean CT as a function of a single input variable (i.e., throughput). In this paper, we demonstrate the feasibility of a quantile regression based metamodeling approach. This method allows estimation of CT quantiles as a function of multiple input variables (e.g., throughput, product mix, and various distributional parameters of time-between-failures, repair time, setup time, loading and unloading times). Empirical results are provided to demonstrate the efficacy of the approach in a realistic simulation model representative of a semiconductor manufacturing system.

In-line metrology of nanoscale features in semiconductor manufacturing systems

Quality control and defect monitoring are of great importance to the semiconductor industry. This article presents a system to enable inspection of nanoscale features in-line with nanomanufacturing processes. The ultimate goal of this research is to integrate this metrology system into current semiconductor manufacturing processes to enable true in-line wafer inspection and quality control. In the system presented in this paper, AFMs that have been shrunk down on to a single MEMS chip are used to scan the surface of a sample. A flexure-based mechanism allows the MEMS-based AFM to be positioned over a millimeter range of motion, with nanometer level precision when properly actuated. The performance of the system in areas such as positioning repeatability, AFM stability and measurement resolution are evaluated in this study. Owing to the small size of single-chip AFM (1 million times smaller than a conventional AFM instrument), it is a good candidate for multipoint detection. Overall, the system presented in this paper significantly shortens inspection time and dramatically decreases the setup time required for an AFM to load, approach, and measure a sample making in-line inspection of nanoscale features in semiconductor manufacturing systems feasible.

Multiple product admission control in semiconductor manufacturing systems with process queue time (PQT) constraints

In a multi-product manufacture system, the complexity of production control increases. Because multiple products are competing for limited resource, managers need to dynamically adjust production priority based on customer demand and distribution of work-in-processes (WIP). The production control in multi-product system becomes even more complicated when process queue time (PQT) constraints exist. Under PQT constraints, administrators setup upper limits for the waiting time between specific manufacturing steps to ensure quality of products. This upper limit of waiting time makes dynamic production control very challenging.The objective of this paper is to develop production control methods for multi-product systems subject to PQT constraints. However, it is computationally infeasible to solve multiple product problems in a single dynamic optimization model, and we therefore first formulate a single product admission control problems using Markov decision processes (MDP). Based on observations from the single product MDP model, an innovative multiple product admission control (MPAC) heuristic is developed. In simulation study, we compare the performance of MPAC with other popular dispatching methods in literature. Compare to other control methods in literature, the proposed (MPAC) method can reduce production costs by at least 33.8% and reduce scrap count by at least 59.5% in average.

Fast Scheduling of Semiconductor Manufacturing Facilities Using Case-Based Reasoning

This paper presents a scheduling method for semiconductor manufacturing systems through utilizing a case-based reasoning approach that consists of modeling, casebase building, and reasoning steps. Petri nets are employed as a model for representing the behavior of a system and accommodating various types of constraints. Casebase contains the cases, composed of pairs of a state and the decision made at the state, and it is built from the solutions of previously solved problems. Subsequently, a schedule for a new problem with different production requirements, initial setup status, and number of setup change operators is obtained by sequentially retrieving and reusing the cases that are most similar to the states encountered during schedule generation in the reasoning step. Extensive experiments demonstrate the performances of the proposed approach, compared to those of dispatching rule and meta-heuristic methods, and also show the robustness against possible perturbations in terms of production requirements and initial setup status based on real world scale datasets. The proposed method significantly outperformed a meta-heuristic algorithm in terms of computation time while requiring much less sacrifice in a performance metric than the competitive dispatching rules.

A MOSFET Fabrication Using a Maskless Lithography System in Clean-Localized Environment of Minimal Fab

A novel semiconductor manufacturing system minimal fab for customized semiconductor devices and micro electro mechanical systems on a half-inch wafer is developed. A distinct feature of the minimal fab is its clean-localized system eliminating a need of cleanroom environment. The clean level of the system is estimated by measuring separately each clean-localized component of the system including a minimal shipping case, wafer carrier, wafer load-lock system, and machine process chamber using a particle counter. The clean level of the system is analyzed to be in the ISO class 4, which is in the same class of a super clean room. In order to confirm the cleanliness needed for device fabrications, we employ a lithography system of the minimal fab consisting of a minimal coater, a minimal maskless exposure system using digital light processing technique, and a minimal developer hybridizing with conventional fab equipment to fabricate a metaloxide-semiconductor (MOS) field-effect-transistors (FET). The minimum gate-length of the MOSFET is realized at 1 μm with a good transistor characteristic. The measured interface states density of 2.33 × 10 10 cm -2 indicates that the whole processes during the fabrication are in the level of a sufficiently low contamination.

Development of an Agent-Based Collaborative Production System Based on Real-Time Order-Driven Approach

With high variation of ordering in current marketing environment, the collaboration of dynamic production planning and scheduling strategies is becoming critical for manufacturers to make quick and correct decisions. On the basis of the complex and distributed semiconductor manufacturing system environment, it is imperative to consider the external disturbances on production scheduling and solve the related issues. However, current researches mostly focus on the collaboration issues within factory internal hierarchy, and lots of methods are discussed on the job sequence or order fulfillment problems on the work cells. Toward this end, a real-time order-driven (RTOD) approach is introduced to assist manufacturers in responding to real-time orders and reflecting them into collaborative production planning and scheduling. An agent-based heterarchical model is established to modularize the collaboration process of planning and scheduling accordingly. A distributed algorithm is provided to quantify the different scenarios during the collaboration, and three sub-algorithms are discussed in detail. Subsequently, the interaction process of each agent is discussed to identify the solutions paths in accordance with the proposed algorithms specifically. A case study is conducted in S semiconductor factory. A Java Agent Development Frameworks platform is developed to simulate the presented collaborative system, and two numerical experiments testified the effectiveness of RTOD approach accordingly.

Development of an integrated real time dispatching system

Purpose – In a complex semiconductor manufacturing system (SMS) environment, the implementation of dynamic production scheduling and dispatching strategies is critical for SMS distributed collaborative manufacturing events to make quick and correct decisions. The purpose of this paper is to assist manufacturers in achieving the real time dispatching and obtaining integrated optimization for shop floor production scheduling. Design/methodology/approach – In this paper, an integrated model is designed under assemble to order environment and a framework of a real time dispatching (IRTD) system for production scheduling control is presented accordingly. Both of the scheduling and ordering performances are integrated into the days of inventory based dispatching algorithm, which can deal with the multiple indicators of dynamic scheduling and ordering in this system to generate the “optimal” dispatching policies. Subsequently, the platform of IRTD system is realized with four modules function embedded. Findings – The proposed IRTD system is designed to compare the previous constant work in process method in the experiment, which shows the better performance achievement of the IRTD system for shop floor production dynamic scheduling and order control. The presented framework and algorithm can facilitate real time dispatching information integration to obtain performance metrics in terms of reliability, availability, and maintainability. Research limitations/implications – The presented system can be further developed to generic factory manufacturing with the presented logic and architecture proliferation. Originality/value – The IRTD system can integrate the real time customer demand and work in process information, based on which manufacturers can make correct and timely decisions in solving dispatching strategies and ordering selection within an integrated information system.

Simulation-based experimental design and statistical modeling for lead time quotation

In a produce-to-order environment, it is of substantial interest to be able to quote a tight and reliable lead time for a new job (or order) upon its arrival to a manufacturing system. This work developed a simulation-based statistical approach to provide responsive and high-quality prediction of a new job's flow time through the system, which renders the capability of accurately quoting lead times in real time. The approach integrates analytical queueing analysis, design of experiments, and statistical modeling to quantify the dependence of a new job's flow time distribution upon the shop status “seen” by that job. To the best of our knowledge, this is the first attempt to design simulation experiments in the space spanned by a complete set of shop status factors for the thorough investigation of the possible impacts those factors may have on the distribution of a new job's flow time. The method has been applied on a scaled-down semiconductor manufacturing system, and the quality of the quoted lead time has been evaluated based on a large and well-designed validation data set in terms of the six commonly used performance criteria.

Fuzzy neural network-based rescheduling decision mechanism for semiconductor manufacturing

Most semiconductor manufacturing systems (SMS) operate in a highly dynamic and unpredictable environment. The production rescheduling strategy addresses uncertainty and improves SMS performance. The rescheduling framework of SMS is presented as layered scheduling strategies with an optimization rescheduling decision mechanism. A fuzzy neural network (FNN) based rescheduling decision model is implemented which can rapidly choose an optimized rescheduling strategy to schedule the semiconductor wafer fabrication lines according to current system disturbances. The mapping between the input of FNN, such as disturbances, system state parameters, and the output of FNN, optimal rescheduling strategies, is constructed. An example of a semiconductor fabrication line in Shanghai is given. The experimental results demonstrate the effectiveness of proposed FNN-based rescheduling decision mechanism approach over the alternatives such as back-propagation neural network (BPNN) and multivariate regression (MR).

A hybrid scheduling approach for automated flowshops with material handling and time constraints

Flowshop scheduling problems have been extensively studied by several authors using different approaches. A typical flowshop process consists of successive manufacturing stages arranged in a single production line where different jobs have to be processed following a predefined production recipe. In this work, the scheduling of a complex flowshop process involving automated wet-etch station from semiconductor manufacturing systems requires a proper synchronisation of processing and transport operations, due to stringent storage policies and fixed transfer times between stages. Robust hybrid solution strategies based on mixed integer linear programming formulations and heuristic-based approaches, such as aggregation and decomposition methods, are proposed and illustrated on industrial-scale problems. The results show significant improvements in solution quality coupled with a reduced computational effort compared to other existing methodologies.

Film deposition using 1-inch-sized HIPIMS system — Toward minimal fabrication semiconductor production system

A copper film was coated using a high-power impulse magnetron sputtering (HIPIMS) source with a target diameter of 25.4mm. This process is a coating process for a half-inch-scaled semiconductor manufacturing system. To realize the stable plasma production, the following issues were considered: (1) a water-cooled plasma generation system, (2) plasma generation using an external magnet near the plasma source, and (3) facilitation of the expansion of the sputtered metallic species (atoms and ions) to the substrate. A high magnetic flux density of 0.3T at the target surface is used to produce plasma at a source voltage less than 600V with negative polarity, where the glow is not transited to an arc discharge during the pulse. To enhance the deposition rate of the prepared films, a long pulse duration of 600μs at a repetition frequency of 120Hz is used. The argon and sputtered copper ions are in a glow plasma region, and a high deposition rate of 2.5μm/min is realized on a 0.5-inch silicon-wafer.

Exponentially weighted moving average-based procedure with adaptive thresholding for monitoring nonlinear profiles: Monitoring of plasma etch process in semiconductor manufacturing

Recently, there are many situations where the quality of a process is characterized by a relationship of functional data (or profiles) such as time series and image data. Such data have been used for detecting out-of-process and quality improvement in many engineering applications such as semiconductor manufacturing, automobile manufacturing, and nano-machining systems. The functional data contain high dimensionality, high feature correlation, non-stationality, and large amount of noise. Due to such characteristics, most classic statistical process control (SPC) may not perform on-line monitoring satisfactorily on functional data. In addition, local shift monitoring with functional data is more significant than the detection of global shifting patterns. In this paper, wavelet-based exponentially weighted moving average (EWMA) test statistic with adaptive thresholding method, which extracts several significant coefficients from original functional data in the wavelet domain and monitors out-of-control events, is proposed. Instead of monitoring global shifting, the local shifting in functional data is of major significance in our study. Throughout this study, we use a spectroscopy in monitoring of plasma etching process from semiconductor manufacturing to illustrate the implementation of the proposed approach. Experiment studies show that the proposed approach quickly detects smaller local shifts compared with the well-known methods.

Dynamic scheduling of photolithography process based on Kohonen neural network

Semiconductor enterprises should continue to provide more capacity to meet the demands of the highly competitive semiconductor markets. Photolithography is usually the bottleneck process with the most expensive equipment in a semiconductor wafer fabrication system. Usually, photolithography area controls the performance of whole semiconductor manufacturing system. To improve the performances of the photolithography area with dynamic dispatching combination rules, a dynamic scheduling method based on Kohonen neural network (KNN) was proposed in this paper. A dynamic scheduling framework is also proposed. The method has been integrated into the dynamic scheduling framework. A KNN-based sample learning algorithm for selecting best combination rules is presented. Finally, the results of simulation experiments indicate that the proposed method is effective and feasible in real-time scheduling of semiconductor fabrication system under both closed-loop release policy and open-loop release policy.

Friction Effect Analysis of a DC Motor

Precision positioning systems are fundamental components in industrial machines such as machine tools, measuring machines and semiconductor manufacturing systems. The performance of the machines depends on the positioning systems. For precise positioning are often used DC motors. The paper deals with experimental method of DC motor friction identification. The aim of the paper is determining of suitable way of DC motor friction modeling. At first the basic static friction models are introduced and linear mathematical model of DC motor is established. DC motor friction identification is by means of experimental method done using Newton's mechanics. For experiment PITTMAN 9413 motor with encoder is used. For measuring of friction torque the measuring I/O card is used which cooperates with Matlab / Simulink by Real Time Toolbox. Then the mathematical model is simulated and compared with real model of DC motor. In the conclusion the effects of particular friction models affecting DC motor are described and the best way of friction modeling for DC motor is determined.

Bottleneck Prediction Method Based on Improved Adaptive Network-based Fuzzy Inference System (ANFIS) in Semiconductor Manufacturing System

Semiconductor manufacturing (SM) system is one of the most complicated hybrid processes involved continuously variable dynamical systems and discrete event dynamical systems. The optimization and scheduling of semiconductor fabrication has long been a hot research direction in automation. Bottleneck is the key factor to a SM system, which seriously influences the throughput rate, cycle time, time-delivery rate, etc. Efficient prediction for the bottleneck of a SM system provides the best support for the consequent scheduling. Because categorical data (product types, releasing strategies) and numerical data (work in process, processing time, utilization rate, buffer length, etc.) have significant effect on bottleneck, an improved adaptive network-based fuzzy inference system (ANFIS) was adopted in this study to predict bottleneck since conventional neural network-based methods accommodate only numerical inputs. In this improved ANFIS, the contribution of categorical inputs to firing strength is reflected through a transformation matrix. In order to tackle high-dimensional inputs, reduce the number of fuzzy rules and obtain high prediction accuracy, a fuzzy c-means method combining binary tree linear division method was applied to identify the initial structure of fuzzy inference system. According to the experimental results, the main-bottleneck and sub-bottleneck of SM system can be predicted accurately with the proposed method.