Chapter

Semiconductor manufacturing scheduling and dispatching: State of the art and survey of needs

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

This chapter discusses scheduling and dispatching in one of the most complex manufacturing environments-wafer fabrication facilities. These facilities represent the most costly and time-consuming portion of the semiconductor manufacturing process. After a brief introduction to wafer fabrication operations, the results of a survey of semiconductor manufacturers that focused on the current state of the practice and future needs are presented. Then the chapter presents a review of some recent dispatching approaches and finally an overview of a recent deterministic scheduling approach is provided.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... arrivals of new jobs, machine breakdowns, and order cancellations. Unfortunately, it is not the case in real-world scheduling applications as disruptions are an natural feature in practice [8], [9], [10]. Moreover, the real-world scheduling problems can have a large number of jobs which cause computational difficulty for optimisation methods (both exact or meta-heuristics) [11], [12]. ...
... Inputs: simulation model of DJSS, simplified model S Output: the best evolved rule ∆ * 1: randomly initialise the population P ← {∆ 1 , . . . , ∆ n } 2: set ∆ * ← null and the fitness f * ← +∞ 3: setup a set of replications R for full evaluation 4: generation ← 0, select a replication π 5: while generation ≤ maxGeneration do 6: for all ∆ i ∈ P do 7: evaluate f g (∆ i ) by applying a ∆ i to π 8: f * g ← +∞ and ∆ * g ← null 9: ...
Article
Full-text available
Automated design of dispatching rules for production systems has been an interesting research topic over the last several years. Machine learning, especially genetic programming (GP), has been a powerful approach to dealing with this design problem. However, intensive computational requirements, accuracy and interpretability are still its limitations. This paper aims at developing new surrogate assisted GP to help improving the quality of the evolved rules without significant computational costs. The experiments have verified the effectiveness and efficiency of the proposed algorithms as compared to those in the literature. Furthermore, new simplification and visualisation approaches have also been developed to improve the interpretability of the evolved rules. These approaches have shown great potentials and proved to be a critical part of the automated design system.
... Because of the increasing automation pressure caused by Automated Material Handling Systems (AMHS), job processing and transportation have to be synchronized, and new requirements on production control to deal with a larger product diversity, it seems that scheduling approaches are both promising and ultimately necessary in the semiconductor manufacturing domain (cf. Pfund et al. 2006). There is a vast literature related to scheduling in wafer fabs. ...
... Many dispatching systems are in place in wafer fabs (cf. the results of a survey related to semiconductor wafer fab scheduling described by Pfund et al. 2006). Often the Applied Real Time Dispatcher product is used (cf. ...
Chapter
In this chapter, we discuss production planning approaches for semiconductor manufacturing. Planning is on the highest level of the PPC hierarchy. Planning approaches provide important input for the order release schemes discussed in Chap. 6. We start by describing short-term planning approaches. Spreadsheet modeling and simulation are used in this situation.
... The goal of this paper is to develop a new online GP (OGP) method for dynamic production scheduling problems, in which multiple scheduling decisions, flexible operations, and dynamic changes are considered. We are interested in these problems as they feature key characteristics of a wide range of production systems in practice such as flexible job shops [13], semiconductor fabrication plants [14], [15], and smart factories [16]. There are three key aspects that make OGP distinct from other methods proposed in previous studies. ...
... The limitation of test beds is outside the scope of this paper and is almost always outside the control of the researcher. The reader is referred to Mönch et al. 2013;Mönch et al. 2011;and Pfund et al., 2006 for a complete summary of current state of the art and challenges of dispatch scheduling. ...
Managing the supply chain of a semiconductor based package goods enterprise-including planning, scheduling, and equipment configurations-is a complicated undertaking, particularly in a manner that is responsive to changes throughout the demand supply network. Typically, management responds to the complexity and scope by partitioning responsibility that narrows the focus of most of the groups in an organization-though the myriad of decisions are tightly integrated. Improving system responsiveness is best addressed by an advanced industrial engineering (AIE) team that is typically the only group with the ability to see the forest and the trees. These teams integrate information and decision technology (analytics) into an application which improves some aspect of planning, scheduling, and equipment configuration. This paper illustrates the need for AIE teams to serve as agents of change, touches on three success stories, highlights the sporadic progress and incubation process in applying analytics to support responsiveness where forward progress by early adopters is often followed with stagnation or reversal as subsequent adopters require a natural incubation period. This paper and its companion paper (Part II. Fab Capability Assessment) identify modeling challenges and opportunities within these critical components of responsiveness: semiconductor fabrication facility/factory capability assessment, moderate length process time windows, moving beyond opportunistic scheduling, and plan repairs to modify unacceptable results. Although aspects of this paper have the feel of a review paper, this paper is different in nature-a view from the trenches which draws from the collective clinical experience of a team of agents of change within the IBM Microelectronics Division (MD) from 1978 to 2012. During much of this period MD was a fortune 100 size firm by itself with a diverse set of products and manufacturing facilities around the world. During this time frame, the team developed and institutionalized applications to support responsiveness within IBM and by IBM clients, while staying aware of what others are doing within the literature and industry. The paper provides insights from the trenches to shed light on the past but more importantly to identify opportunities for improvement and the critical role of advanced industrial engineers as agents of change to meet these challenges.
... Today dispatching rules are commonly utilized in semiconductor manufacturing [6]. However, deterministic scheduling is a promising approach and ultimately may be necessary to further improve the effectiveness of wafer fabrication [7]. This paper will discuss the optimization of schedules for tool groups in a wafer fab. ...
Conference Paper
Full-text available
In semiconductor manufacturing, we face many complex scheduling problems, which are difficult to tackle with conventional optimization techniques. Evolutionary algorithms are a reasonable method to approach these problems. However, their quality depends on the adequateness of the employed problem encoding and determining the best problem representation is a big challenge. Therefore we propose to use a number of different representations (species), competing with each other in an evolutionary algorithm. We developed an extensible evolutionary algorithm, which adheres to a modular design philosophy enabling customization in a multitude of ways. It supports multiple species and races, different selection strategies and further extensions like aging. A typical problem we investigate is offline scheduling for parallel cluster tools, which effectively is a combination of partitioning and sequencing of jobs with additional constraints. In this paper we try to maximize the throughput of a cluster tool group by minimizing the makespan of the scheduled jobs. The multi-species approach we employed generated promising results for this particular problem.
... Scheduling semiconductor factories is a very challenging task due to the high complexity of the manufacturing process as well as the importance of this task in a competitive and capital-intensive production environment. This high complexity, and the requirement to be able to react quickly to disruptions on the shop floor (e.g. machine failures), have led to the prevalent use of dispatching rules (Pfund, Mason, and Fowler 2006). Dispatching rules can be used to take decisions in real-time. ...
Conference Paper
Full-text available
Developing dispatching rules for complex production systems such as semiconductor manufacturing is an involved task usually performed manually. In a tedious trial-and-error process, a human expert attempts to improve existing rules, which are evaluated using discrete-event simulation. A significant improvement in this task can be achieved by coupling a discrete-event simulator with heuristic optimization algorithms. In this paper we show that this approach is feasible for large manufacturing scenarios as well, and it is also useful to quantify the value of information for the scheduling process. Using the objective of minimizing the mean cycle time of lots, we show that rules created automatically using Genetic Programming (GP) can clearly outperform standard rules. We compare their performance to manually developed rules from the literature.
... Moreover , because each scheduling decision is made at the latest possible moment, i.e. immediately before its implementation, dispatching rules naturally possess the ability to quickly react to unexpected changes, which makes them particularly suited for stochastic and dynamic scheduling problems (for a list of papers explicitly addressing stochastic dynamic environments see Table I). These properties, together with their simple and intuitive nature, their ease of implementation and their flexibility to incorporate domain knowledge and expertise [69] explain the wide usage of dispatching rules in practice [70] and the ongoing research on the development of new, more effective dispatching rules (see, e.g. [71], [72], [73]). ...
Article
Full-text available
Hyper-heuristics have recently emerged as a powerful approach to automate the design of heuristics for a number of different problems. Production scheduling is a particularly popular application area for which a number of different hyperheuristics have been developed and shown to be effective, efficient, easy to implement, and reusable in different shop conditions. In particular, they seem a promising way to tackle highly dynamic and stochastic scheduling problems, an aspect that is specifically emphasised in this survey. Despite their success and the substantial number of papers in this area, there is currently no systematic discussion of the design choices and critical issues involved in the process of developing such approaches. This review strives to fill this gap by summarising the state of the art, suggesting a taxonomy, and providing the interested researchers and practitioners with guidelines for the design of hyper-heuristics in production scheduling. This paper also identifies challenges and open questions and highlights various directions for future work.
Article
The purpose of this paper is to statistically compare six dispatch rules for semiconductor assembly & test operations with respect to two hierarchical objectives: minimise the weighted sum of key device (chip) shortages and maximise the weighted throughput of lots processed. Three factors affecting system performances are studied: the different rules, whether the machines are set up at time zero, and whether the maximum number of changeovers is restricted over the planning horizon. System performance is determined with both a discrete event AutoSched AP simulation model and a greedy randomised adaptive search procedure (GRASP) for 30 real and randomly generated data sets. A full factorial design with common random numbers was conducted to concurrently investigate the interaction among the three factors. The statistical analysis showed that the two rules designed to process as many hot lots (i.e., key device lots) as possible performed the best in minimising the weighted shortage. We also found that initial setups and limiting the number of setups over a 2- to 3-day planning horizon can significantly weaken the rules’ performance but in some cases the interaction of these factors mitigated their negative impacts. Due to the general nature of the models, the results should be applicable to most reentrant flow shops. The GRASP is now being used by the sponsoring company and the most promising dispatch rules are being studied in an experimental environment.
Article
Model predictive control (MPC)-based approach to fab-wide scheduling has been suggested to solve constraint-aware production optimization and in-process inventory level control simultaneously at each scheduling instance. However, application of this approach to real fab suffers from computational difficulties brought by the need to solve a huge optimization problem on-line as real fab scheduling problems are characterized by long cycle times, multiple product types, hundreds of machines/processing steps and re-entrant product flows. This study explores the use of an offset-blocking strategy combined with a modified recursive least square (RLS) estimation in the fab-wide scheduler, in order to alleviate the difficulty. The strategy is tested on a modified version of published case study called Intel Mini-Fab (IMF) problem. Despite its simplicity, the blocking strategy showed excellent performance in the face of realistic demand changes and plant/model mismatch.
Article
Full-text available
Abstract One way to accelerate evolutionary algorithms with expensive fitness evaluations is to combine them with surrogate models. Surrogate models are efficiently computable approximations of the fitness function, derived by means of statistical or machine learning techniques from samples of fully evaluated solutions. But these models usually require a numerical representation, and therefore can not be used with the tree representation of Genetic Programming (GP). In this paper, we present a new way to use surrogate models with GP. Rather than using the genotype directly as input to the surrogate model, we propose using a phenotypic characterization. This phenotypic characterization can be computed efficiently and allows us to define approximate measures of equivalence and similarity. Using a stochastic, dynamic job shop scenario as an example of simulation-based GP with an expensive fitness evaluation, we show how these ideas can be used to construct surrogate models and improve the convergence speed and solution quality of GP.
Conference Paper
In this paper, we study the performance of an extension of the shifting bottleneck heuristic (SBH) for complex manufacturing systems that makes in addition to machine-related decisions also material handling-related decisions. We use the heuristic within a rolling horizon approach. We are interested in assessing the impact of the capacity of the material handling system, measured in number of available vehicles, on the performance of the proposed scheduling algorithm. Simulation experiments with models of manufacturing systems with processing conditions similar to those found in semiconductor wafer fabrication facilities (wafer fabs) are performed. It turns out that the capacity of the material handling system has an impact on the performance of the integrated scheduling heuristic.
Article
We propose a two-stage hyper-heuristic for the generation of a set of work centre-specific dispatching rules. The approach combines a genetic programming (GP) algorithm that evolves a composite rule from basic job attributes with an evolutionary algorithm (EA) that searches for a good assignment of rules to work centres. The hyper-heuristic is tested against its two components and rules from the literature on a complex dynamic job shop problem from semiconductor manufacturing. Results show that all three hyper-heuristics are able to generate (sets of) rules that achieve a significantly lower mean weighted tardiness than any of the benckmark rules. Moreover, the two-stage approach proves to outperform the GP and EA hyper-heuristic as it optimises on two different heuristic search spaces that appear to tap different optimisation potentials. The resulting rule sets are also robust to most changes in the operating conditions.
Article
In this paper, we describe a hierarchically organised multi-agent-system prototype that is used for production control of semiconductor wafer fabrication facilities (wafer fabs). Starting from requirements for modern production control systems and advanced production control algorithms, we design the hierarchical system. We suggest a three-layer hierarchy. We describe production control algorithms for each layer of the hierarchy. We consider the performance measure total weighted tardiness. After the design step, we discuss the implementation of the system prototype as a multi-agent-system (MAS). Our implementation is based on the PROSA reference architecture for holonic manufacturing systems. The MAS approach especially allows for a distribution of the system on a computer cluster. We present the results of computational experiments with the system prototype. Furthermore, we suggest future research directions.
Article
Wafer fabrication is a capital-intensive and highly complex manufacturing process. In the wafer fabrication facility (fab), wafers are grouped as a lot to go through repeated sequences of operations to build circuitry. Lot scheduling is an important task for manufacturers to improve production efficiency and meet customers’ requirements of on-time delivery. In this research we propose a dispatching rule for lot scheduling in wafer fabs, focusing on three due date-based objectives: on-time delivery rate, mean tardiness, and maximum tardiness. Although many dispatching rules have been proposed in the literature, they usually perform well in some objectives and bad in others. Our rule implements good principles in existing rules by means of (1) an urgency function for a single lot, (2) a priority index function considering total urgency of multiple waiting lots, (3) a due date extension procedure for dealing with tardy lots, and (4) a lot filtering procedure for selecting urgent lots. Simulation experiments are conducted using nine data sets of fabs. Six scenarios formed by two levels of load and three levels of due date tightness are tested for each fab. Performance verification of the proposed rule is achieved by comparing with nine benchmark rules. The experimental results show that the proposed rule outperforms the benchmark rules in terms of all concerned objective functions.
Article
Full-text available
In this paper, we discuss scheduling problems in semiconductor manufacturing. Starting from describing the manufacturing process, we identify typical scheduling problems found in semiconductor manufacturing systems. We describe batch scheduling problems, parallel machine scheduling problems, job shop scheduling problems, scheduling problems with auxiliary resources, multiple orders per job scheduling problems, and scheduling problems related to cluster tools. We also present important solution techniques that are used to solve these scheduling problems by means of specific examples, and report on known implementations. Finally, we summarize some of the challenges in scheduling semiconductor manufacturing operations.
Article
Full-text available
This paper addresses an extension of the resource-constrained project scheduling problem that takes into account storage resources which may be produced or consumed by activities. To solve this problem, we propose the generalization of two existing mixed integer linear programming models for the classical resource-constrained project scheduling problem, as well as one novel formulation based on the concept of event. Computational results are reported to compare these formulations with each other, as well as with a reference method from the literature. Conclusions are drawn on the merits and drawbacks of each model according to the instance characteristics.
Conference Paper
In semiconductor foundry fab, it is not always true that the traditional bottleneck resource allocation assumes a balanced loading for downstream equipment. Thus, a Dynamic Dispatching Model for bottleneck resource allocation is proposed in order to reduce the downstream machine lost time and increase the downstream machine total moves. The model considers not only the WIP status but also the machine process time. Results indicate that by setting up the model, the instruction table will be generated to indicate the allocation results for the downstream tools simultaneously. From the improvement ratio, the model was useful for foundry Fab in minimizing lost time and maximizing total machine moves.