Elsevier

Computers in Industry

Volume 65, Issue 8, October 2014, Pages 1115-1125
Computers in Industry

Fuzzy neural network-based rescheduling decision mechanism for semiconductor manufacturing

https://doi.org/10.1016/j.compind.2014.06.002Get rights and content

Highlights

  • The rescheduling framework of SMS is presented as layered scheduling strategies with an optimized decision mechanism.

  • The system's states and disturbances including machine failures and rush orders are mathematically characterized.

  • An adaptable and robust fuzzy neural networks (FNN) based optimal rescheduling decision mechanism is developed.

  • Numerical results demonstrate that the proposed method outperforms in terms of daily movement and machine utilization.

Abstract

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).

Introduction

Semiconductor manufacturing systems (SMS) are more complicated than conventional manufacturing systems in terms of technologies and manufacturing procedures. A mix of different process types (batch processes and single wafer processes), sequence dependent setup times, expensive equipment, and reentrant flows are typical for SMS [1]. Between 250 and 500 process steps on 50–120 different types of equipment are required to produce a chip of average complexity. Since the 1990s, the market of semiconductor fabrication has become increasingly global, dynamic and customer driven. An organization's competitive advantage depends more and more on its responsiveness in meeting market changes and opportunities, and in coping with unforeseen circumstances (i.e., machine breakdowns, rush orders, etc.) Thus, it is important to reduce inventories, decrease cycle time, and improve resource utilization. These goals call for optimization and scheduling approaches that optimize the allocation of scarce and expensive resources among competing activities [2]. Uzsoy et al. [3] provided an excellent review of scheduling research for SMS. SMS operates in uncertain dynamic environments, where disturbances include: machine failure, lot rework, and rush orders. Production rescheduling is an effective response to uncertainty created by the exterior business environment and the interior production conditions. When disturbances happen, the rescheduling strategy needs to be selected and carried out.

This paper is motivated by the problem of rescheduling semiconductor wafer fabrication lines, where the schedule solutions are easy to be infeasible due to dynamic and uncertain production environments. Existing approaches to rescheduling the semiconductor wafer fabrication system generally consist of a single strategy for special situations, which is not enough for the real-time and dynamic manufacturing environments. Therefore this paper focuses on an optimal rescheduling method and proposes a fuzzy neural network (FNN)-based rescheduling decision mechanism for SMS.

The paper is organized as follows. The next section gives a brief introduction to SMS and reviews literature about rescheduling. Section 3 describes the layered rescheduling framework of SMS. Section 4 constructs an optimization rescheduling decision mechanism based on FNN. Section 5 reports numerical experiments based on the data of a 6 inch SMS in Shanghai. The paper concludes in Section 6 with a discussion of the application of FNN-based rescheduling decision mechanism to SMS.

Section snippets

Literature review

The semiconductor manufacturing process can be broadly classified into three major stages: material preparation, wafer fabrication, and assembly and testing [4]. Wafer fabrication is the most complex stage. Multiple layers of integrated circuits are imprinted on the wafer surface. Wafer fabrication requires an enormous investment in facilities. The necessary machines are not duplicated in a fabrication line but are grouped by type. So, wafers repeatedly revisit the same machine type for

Layered rescheduling framework of SMS

SMS operates in uncertain dynamic environments, where the main disturbances include machine failure, lot rework, rush orders and so on. When such unpredicted disturbances occur, rescheduling is required. However, considering the complexity of SMS, it is necessary to adopt different rescheduling strategy on basis of current system status and the impact of the disturbances to reduce the computing cost, and at the same time to ensure the stability and effectiveness of scheduling. Therefore in

FNN-based optimal rescheduling decision mechanism

Fuzzy neural networks (FNN) are an ingenious combination of fuzzy logic and neural networks. It combines the advantages of fuzzy systems (including the ability to process fuzzy information using fuzzy algorithms) and neural networks (including a learning ability and a high-speed parallel structure.) The FNN approach is adaptable and robust, and is well suited for the SMS rescheduling problem.

The FNN-based rescheduling decision model includes an input layer, an output layer and several hidden

Numerical experiments

This section reports numerical experiments and verifies the effectiveness of the FNN-based rescheduling decision mechanism. A simulation model based on a 6-inch SMS in Shanghai is used to produce the experiment data. This particular SMS is made of 11 key machine groups, which add up to 34 key machines described by mean time to failure rate (MTTF) and mean time to repair rate (MTTR) parameters. The three different types of lot products are denoted A, B and C. The process of each lot product is

Conclusion

SMS is characterized by reentrant flows, a mix of different process types, sequence dependent setup times and very expensive equipment. Further, an organization's competitive advantage depends more and more on its responsiveness and adaptability to changing circumstances (such as machine breakdowns, material shortages, etc.) The rescheduling framework of SMS is presented as layered scheduling strategies with an optimal rescheduling decision mechanism. This paper suggests a novel approach using

Acknowledgement

This work was supported by the National Natural Science Foundation of China under Grant Nos. 60934008 and 51375307.

J. Zhang is currently a professor at School of Mechanical Engineering, Shanghai Jiao Tong University in China. She received her PhD in mechanical engineering from Nanjing University of Aeronautics & Astronautics, China in 1997, her BS and MS degrees from Jiangsu University of Science & Technology, China in 1984 and 1991, respectively. From 1998 to 1999, she was a postdoctoral fellow in Department of Industrial and Manufacturing Systems Engineering, Huazhong University of Science & Technology,

References (31)

  • T. Chitchachornvanich

    Optimizing Semiconductor Fabrication Scheduling in the Face of Uncertainties

    (1998)
  • P. András et al.

    Stability-oriented evaluation of rescheduling strategies, by using simulation

    Computers in Industry

    (2007)
  • M. Cheng et al.

    A fast rescheduling method in semiconductor manufacturing allowing for tardiness and scheduling stability

  • M. Cheng et al.

    Online rescheduling in semiconductor manufacturing

  • H.P. Huang et al.

    A new approach to on-line rescheduling for a semiconductor foundry Fab

  • Cited by (28)

    • Knowledge augmented broad learning system for computer vision based mixed-type defect detection in semiconductor manufacturing

      2023, Robotics and Computer-Integrated Manufacturing
      Citation Excerpt :

      Compared to common pattern recognition tasks, mixed-typed defect detection remains to be tough due to the diverse graphic characteristics and the mix of different defective patterns [11]. During wafer manufacturing, layered circuits [12] are built during the re-entrant process. The circuits in different layers are connected together [13] to form an integrated circuit.

    • A new power system active rescheduling method considering the dispatchable plug-in electric vehicles and intermittent renewable energies

      2022, Applied Energy
      Citation Excerpt :

      On this basis, to reschedule the system under featured circumstances become a crucial solution. The concept of rescheduling comes from the scheduling theory and rescheduling methods have been widely adopted in many engineering fields such as manufacturing [44–47] and transportation industry [48–50]. Carlos et al. [44] proposed a rescheduling mechanism that satisfied distributed constraints and contract network protocols to improve the adaptability of production systems to unpredictable order requirements.

    • An adaptive ensemble deep forest based dynamic scheduling strategy for low carbon flexible job shop under recessive disturbance

      2022, Journal of Cleaner Production
      Citation Excerpt :

      Liu et al. (2014) quantified the impact of disturbances by establishing a profit-loss cloud model concerning time and cost, which were regarded as the trigger condition of rescheduling, and an example was then given to demonstrate the rationality and feasibility of the model. Zhang et al. (2014) took disturbances and system status as decision basis and presented a fuzzy neural network (FNN) based layered rescheduling mechanism, which illustrated its effectiveness by the comparison with back-propagation neural network (BPNN) and regression approach. Recently, the deep forest algorithm has been preferred in applying to dynamic scheduling, because of its few super parameters, no parameter adjustment and high parallel computing efficiency.

    • A hybrid fuzzy-neural-based dynamic scheduling method for part feeding of mixed-model assembly lines

      2022, Computers and Industrial Engineering
      Citation Excerpt :

      For example, a scheduling policy based on the back propagation networks (BPNs) was introduced by (Wang & Choi, 2014), aiming to minimize the makespan of a flexible flow shop (FFS) with stochastic processing times. Zhang, Qin, Wu, and Zhai (2014) implemented a fuzzy neural network (FNN) based rescheduling decision model to schedule the semiconductor wafer fabrication lines. However, as far as the author knows, researches on applying the ANNs to solving the part feeding scheduling problems are extremely limited.

    • Dynamic dispatching for interbay automated material handling with lot targeting using improved parallel multiple-objective genetic algorithm

      2021, Computers and Operations Research
      Citation Excerpt :

      Compared with intrabay AMHS, interbay AMHS involves a larger number of automated vehicles, more complex transportation tasks, and more stochastic transportation demands with a higher risk of transportation deadlocks and vehicle blockages (Wu et al., 2011). Therefore, developing effective and efficient scheduling strategies for interbay AMHS has received considerable attention over the last few years (Zhang et al., 2014; Zhou and Zhou, 2018). As shown in Fig. 1, an interbay AMHS usually consists of three subsystems: transportation rails, stockers, and automated vehicles.

    View all citing articles on Scopus

    J. Zhang is currently a professor at School of Mechanical Engineering, Shanghai Jiao Tong University in China. She received her PhD in mechanical engineering from Nanjing University of Aeronautics & Astronautics, China in 1997, her BS and MS degrees from Jiangsu University of Science & Technology, China in 1984 and 1991, respectively. From 1998 to 1999, she was a postdoctoral fellow in Department of Industrial and Manufacturing Systems Engineering, Huazhong University of Science & Technology, China. From 2000 to 2001, she was a research assistant in Department of Industrial and Manufacturing Systems Engineering, University of Hong Kong, China. In 2002, she joined the School of Mechanical Engineering, Shanghai Jiao Tong University in China as an associate professor. She has published more than 30 articles at international journals. Her current research interests are modeling, scheduling and planning in manufacturing and intelligent manufacturing system.

    W. Qin is Mechanical Engineering currently a postdoctoral fellow at School of Shanghai Jiao Tong University in China. He received the BS degree from Shanghai Jiao Tong University, China, in 2004 and the MS degree from Tsinghua University, China, in 2006. In 2011, he received his PhD degree from The University of Hong Kong. His current research interests are production scheduling and control, intelligent manufacturing system and supply chain management.

    L.H. Wu received the BS degree in mechanical engineering from Zhengzhou University of Light Industry, China, in 2002 and the MS degree in Mechanical Engineering from Wuhan University, China, in 2006. In 2011, he received his PhD degree from Shanghai Jiao Tong University, China. His current research interests are complex manufacturing system modeling.

    W.B. Zhai received his PhD degree from Shanghai Jiao Tong University, China in 2005. His current research interests are scheduling and planning in manufacturing.

    View full text