Proceedings of the 2016 International Symposium on Semiconductor Manufacturing Intelligence (ISMI2016)
1
Abstract—Various factors (constraints) should be
considered in semiconductor manufacturing scheduling.
Based on heuristic scheduling rules, this paper designs a
composite rule with comprehensive consideration of various
factors for manufacturing system. The composite rule is
obtained by optimizing its parameters using response surface
methodology for diverse production status. And then,
dynamic scheduling of semiconductor manufacturing is
studied based on Support Vector Regression (SVR). This
approach dynamically obtains composite scheduling rule to
optimize production performance according to real-time
status of semiconductor production line. Finally, the proposed
dynamic scheduling approach is tested on a real
semiconductor manufacturing to verify the effectiveness in
production performance optimization.
Keywords: Dynamic Scheduling, Semiconductor
Manufacturing, Composite Rule, Support Vector
Regression (SVR)
I. INTRODUCTION
Semiconductor manufacturing system is a dynamic
system with various uncertainties (e.g. machine failures,
arrival of urgent jobs and the change of due time, etc.).
Once unpredictable real-time events happened, the
previously “optimal” schedule may lose its feasibility. Such
kind of scheduling considering real-time events may be
defined as “dynamic scheduling” (Mouelhi-Chibani et al,
2010).
Dynamic scheduling of manufacturing systems is
supposed to select appropriate scheduling rules from
scheduling rule sets to meet the needs of production
scheduling (Priore et al, 2014). Now, some researchers
have been studying the dynamic scheduling based on
learning mechanism, that is, to learn scheduling knowledge
from optimized scheduling samples, and to apply the
knowledge learned (scheduling model) to get feasible
scheduling rules according to real-time system status. For
example, Shiue et al (2012) proposed a self-organizing
map-based multiple scheduling rules selection mechanism;
Tsai et al (2007) put forward a RFID-based real-time
scheduling system for an automated semiconductor
manufacturing plant, which selecting features for training
samples and establishing a dynamic scheduling model
based on support vector machine (SVM); Olafsson et al
(2010) suggested a dynamic scheduling strategy selection
method based on genetic algorithm (GA) and decision tree;
Qiao et al (2013) and Ma et al (2014) respectively used
binary particle swarm optimization and support vector
machine (BPSO-SVM) and k-nearest neighbors (KNN)
algorithm to realize dynamic scheduling for semiconductor
manufacturing system. All of above suggested to select
appropriate scheduling rules from simple and effective
heuristic scheduling rule sets to match with manufacturing
system status based on local information of manufacturing
This research is supported in part by National Nature Foundation of
China (No. 61273046, 51475334).
Y.-M. Ma, W.-J. Wu, F. Qiao, Z.-H. Min and L. Li are with the School
of Electronical & Information Engineering, Tongji University, China.
system, such as due time of jobs, equipment load and etc.
However, various factors (constraints) should be
considered during scheduling process in the industry, such
as the urgency of jobs, process constraints, due time of jobs,
etc. that means, global information-based dispatching rules
are needed during implementing dynamic scheduling for
manufacturing system by adjusting their key parameters. Li
et al (2013) used a BP neural network, binary regression
model and particle swarm optimization to study samples,
thereby obtaining a self-adapt scheduling model to meet the
dynamic scheduling needs; Lee et al (2004) used a real-time
dispatching mechanism integrating autonomy and
coordination, in which an advanced dispatching rule was
determined based on global information. Once trigger
events occurred, the parameters of dispatching rules would
be adjusted dynamically. The scheduling structure of this
approach is keeping stable, but the choice of key parameters
is difficult.
Due to the complexity and multiple process constraints
of semiconductor production line, if using advanced
dispatching rules for scheduling, global information is
needed to take into account and results in computationally
demanding; while using simple rules scheduling for
scheduling, the effectiveness of optimization is not satisfied.
Therefore, improved simple rules is suggested for
semiconductor production scheduling (Pickardt et al, 2013;
Bouri et al, 2015; Yu et al, 2012). This paper will propose a
kind of simple and feasible composite scheduling rule and
apply it to the scheduling for semiconductor production line,
then study a dynamic scheduling method based on
composite rules which can generate optimal scheduling
under real-time status of production line.
II. DESIGN AND OPTIMIZATION OF COMPOSITE
SCHEDULING RULE
A. Design of composite scheduling rule
The simple heuristic scheduling rule is supposed to
obtain jobs sequence according to the attributes of jobs (due
time, process time, etc.) to meet the objectives of
manufacturing system scheduling. The composite rule here
is a dynamic integration of several simple scheduling rules
optimizing diverse scheduling objectives, and the
composite priority for jobs sequencing is defined. The
computation process consists of two steps: firstly, obtaining
the priority
of job determined by simple rule
, then, obtaining the integrated priority
of job .
1) Definition of simple rule priority
Suppose job is in machine buffer waiting for
processing. When using rule
to sort jobs, the priority
of job is determined by job attributes related to rule
, and
, where the greater the value of
is,
the earlier the job accepts processing. So there are two
scenarios:
The greater the value of job attributes is, the
earlier the job accepts processing. The priority
Dynamic Scheduling for Semiconductor Production Line Based on
Composite Rule
Yumin Ma, Wenjing Wu, Fei Qiao, Zhihong Min, Li Li