You are on page 1of 15

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224342982

Levels of Capacity and Material Handling System Modeling for Factory


Integration Decision Making in Semiconductor Wafer Fabs

Article  in  IEEE Transactions on Semiconductor Manufacturing · December 2008


DOI: 10.1109/TSM.2008.2005368 · Source: IEEE Xplore

CITATIONS READS
37 663

3 authors:

Jesus Jimenez Gerald T. Mackulak


Texas State University Arizona State University
47 PUBLICATIONS   332 CITATIONS    98 PUBLICATIONS   1,235 CITATIONS   

SEE PROFILE SEE PROFILE

J.W. Fowler
Arizona State University
288 PUBLICATIONS   7,001 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Master's Thesis View project

Optimization in Manufacturing View project

All content following this page was uploaded by J.W. Fowler on 27 May 2014.

The user has requested enhancement of the downloaded file.


600 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 21, NO. 4, NOVEMBER 2008

Levels of Capacity and Material Handling System


Modeling for Factory Integration Decision Making in
Semiconductor Wafer Fabs
Jesus A. Jimenez, Gerald T. Mackulak, and John W. Fowler

Abstract—As the costs of building a new wafer fab increase, a the automated material handling systems (AMHS), and the
detailed simulation model representing the production operations, equipment–AMHS interactions [2].
the tools, the automated material handling systems (AMHS), and Simulation has been actively used by fab managers for con-
the tool–AMHS interactions is needed for accurately planning the
capacity of these facilities. The problem is that it currently takes ducting capacity analysis and AMHS design; it is probably the
too long to build, experiment, and analyze a sufficiently detailed only available tool capable of carrying out capacity analysis of
model of a fab. The key for building accurate and computation- the entire fab effectively. However, it currently takes too long to
ally efficient fab models is to decide on the right amount of model design, build, experiment, and analyze a sufficiently detailed ca-
details, specifically those details representing the equipment ca- pacity model of a wafer fab using available simulation tools. The
pacity and the AMHS. This paper identifies a method for classi-
fying a fab model by the level of capacity detail, the level of AMHS
reason is that these models need to represent complex system
detail, or the level of capacity/AMHS detail. Within the capacity/ configurations, consisting of a large number of tools, multiple
AMHS modeling level, our method further differentiates between product types, several processing routes (each consisting of hun-
detailed integrated capacity/AMHS models and abstract coupled dreds of operations), very complex process characteristics, etc.
capacity/AMHS models. The proposed classification method serves Furthermore, model complexity grows significantly in propor-
as the basis of a framework that helps users select the system com-
tion to how accurately the AMHS is represented. In practice,
ponents to be modeled within a desired level of detail. This re-
search also provides a review of past-published literature summa- most simulations currently model the AMHS separately from
rizing the work done at each of the proposed fab modeling levels. A the fab capacity model so that simulations can be relatively man-
case study comparing the performance between an integrated ca- ageable and efficient.
pacity/AMHS model and a coupled capacity/AMHS model is pre- The choice of the level of capacity and AMHS modeling de-
sented. The study demonstrates that the coupled model generates termines the speed and the level of predictive accuracy of the
cycle time estimates that are not statistically different than those
generated by the integrated model. This paper also shows that the simulations. If a model contains the details necessary to make
coupled model can improve CPU time by approximately 98% in it an accurate representation of the wafer fab, it may require too
relation to the integrated model. much execution time to allow the desired level of experimenta-
Index Terms—Automated material handling systems (AMHS), tion and scenario testing in the time allocated for this activity.
capacity planning, discrete-event simulation, modeling, wafer On the other hand, if a model contains enough details to make
fabrication. it execute fast, it may lack the capability to generate predictions
that are accurate enough for effective decision-making. Model
builders are tasked with determining the right amounts of model
I. INTRODUCTION detail within the capacity and AMHS constructs to produce rea-
sonable answers within a given time bound.

T HE manufacture of 300-mm semiconductor wafers is


complex and increasingly fully automated. Thus, the
overall cost of a wafer fabrication facility (fab) is currently
The current available work in modeling does not provide suf-
ficient support for selecting the right levels of capacity/AMHS
detail. In a study outside the context of wafer fab modeling,
about $3 billion dollars [1]. One of the key strategies used Lee and Fishwick [3] formulated an integer programming (IP)
for maximizing profit in such cost-intensive environments is model that seeks to find the optimal abstraction level for a
to design the fab capacity accurately in order to enable high steam-powered propulsion ship model. In their IP formulation,
equipment utilizations and efficient production operations. This the resulting optimal abstraction level must satisfy constraints
is the reason why capacity design studies must include every that specify expected project deadlines and solution accuracy
possible detail about the factory operations, the equipment, requirements. Therefore, the proposed IP model requires pre-
dicted values of simulation CPU times and solution accuracy
Manuscript received September 30, 2005; revised June 23, 2008. Current ver- at each abstraction level under consideration prior to the opti-
sion published November 05, 2008.
J. A. Jimenez is with the Ingram School of Engineering, Texas State
mization effort. Unfortunately, the application of this IP-based
University-San Marcos, San Marcos, TX 78666-4616 USA (e-mail: approach for the optimization of the level of fab modeling is
Jesus.Jimenez@txstate.edu). currently impractical because there is a lack of tools that can
G. T. Mackulak and J. W. Fowler are with the Industrial Engineering
Department, Arizona State University, Tempe, AZ 85287–5906 USA (e-mail:
make fair predictions of CPU times and solution accuracy. This
Mackulak@asu.edu; John.Fowler@asu.edu). is why practitioners have traditionally used empirical methods
Digital Object Identifier 10.1109/TSM.2008.2005368 to determine adequate levels of Fab detail.
0894-6507/$25.00 © 2008 IEEE
JIMENEZ et al.: LEVELS OF CAPACITY AND MATERIAL HANDLING SYSTEM MODELING 601

The empirical methods for selecting the level of Fab detail Quadrant 1: This represents a very crude modeling level
consist primarily of modeling guidelines and generic simula- that produces fast rough-cut capacity analysis, but without any
tion models. These methods often use model builder’s exper- AMHS details. These models are often spreadsheet modeled.
tise to identify a reduced amount of model detail that can meet As Fig. 1 explains, spreadsheet models are widely popular in
the study requirements given by the user. The problem is that capacity planning due to its ease of use and fast analysis time.
those empirical methods that allow the selection of the level of However, this method does not typically yield high predictive
capacity detail do not give sufficient information to be able to accuracy.
choose the level of AMHS detail and vice versa. Refer to [4] and The modeling level described in Quadrant 2 gives a suffi-
[5] for studies supporting the selection of the level of capacity ciently detailed representation of the AMHS, almost to the point
detail and [6] and [7] for studies supporting the selection of the of emulating the actual flow of products throughout the material
level of AMHS details. Because the production of a 300-mm handling system. The problem is that these models offer little in-
wafer is now strongly dependent on automation, there is an in- formation about the production of wafers. As it is shown in the
creasing need for incorporating the AMHS into current model diagram in Fig. 2(a), the details of the production operations are
development methodologies. replaced by “From–To” material flow tables, which define the
This paper describes a method for classifying fab models rates at which materials are moved between two locations in the
by the level of capacity detail, the level of AMHS detail, or fab. The AMHS simulations have the following sequence: 1) a
the combined level of capacity/AMHS detail. This classifica- wafer lot moving between processes arrives to a “source” lo-
cation (e.g., stocker, tool port); 2) the lot requests a vehicle and
tion method is used ultimately as a framework to identify rec-
waits to be picked up at the “source” location; 3) an assigned ve-
ommended levels of detail for a variety of problems found in
hicle travels to the “source” location to pick up the lot; 4) upon
capacity/AMHS design. The development of this framework is
vehicle arrival, the lot is transferred into the vehicle, which then
based primarily on an extensive literature review of past pub-
delivers the material to a “destination” location; and 5) the lot
lished simulation studies for each of the proposed levels of de-
is transferred into the “destination” location, where it is termi-
tail. We analyze the data gathered from the literature review as
nated from the simulation.
to cluster studies with similar model characteristics in terms of The modeling level in Quadrant 3 mimics the activities
the factors that affect the level of fab detail (i.e., the planning involved in the fabrication of wafers, i.e.: 1) wafers arrive to a
horizon of the problem, the performance measures, model ar- process; 2) wait in queue for a free machine, auxiliary resource,
chitecture, etc.). For each level of capacity/AMHS detail that etc.; 3) start the processing step; and 4) move to the next
we propose, we then specify which fab components should be processing step in sequence. In the model, the details of these
modeled so that the user can accomplish desired study goals manufacturing activities are grouped by category (i.e., product
efficiently. description, stations, process description, storages, “From–To”
Among the levels of capacity/AMHS that we identify in this AMHS delivery times, etc.). As it can be seen in the diagram
paper, we further differentiate between an integrated capacity/ in Fig. 2(b), these simulations provide little or no information
AMHS model (i.e., a detail model) and a coupled capacity/ about the AMHS.
AMHS model (i.e., an abstract model). Practitioners use these Traditionally, the modeling level in Quadrant 2 has been
models to represent the entire wafer fab and the AMHS. This widely used in AMHS design, whereas the modeling level in
paper presents a case study to illustrate the tradeoffs between Quadrant 3 has been popular in fab capacity planning. Most
solution accuracy and computational efficiency that can occur capacity planning simulations are currently run separately from
by using an abstract modeling approach over a detailed mod- AMHS design simulations. Simulating the AMHS and capacity
eling approach. separately often works well because the AMHS is designed to
The paper is organized as follows. Section II describes our not be the bottleneck of the fab [8]. Although this approach
method for classifying fab models. Section III provides a re- helps maintain relatively manageable models, it is more dif-
view of the current published literature. Section IV presents ficult to accurately predict how the tool–AMHS interactions
the details of the case study and compares in detail the perfor- affect the overall fab performance.
mance of integrated capacity/AMHS models and coupled ca- The need for predicting system performance at a higher level
pacity/AMHS models. Section V proposes the levels of detail is the reason why the detailed levels of capacity and AMHS
for specific simulation objectives. Finally, Section VI states the modeling have been combined. The models in Quadrant 4, re-
conclusions of this paper. ferred herein as the capacity/AMHS model, is capable of repre-
senting and predicting the performance of the actual wafer fab
and the material handling system very accurately. As one would
II. GENERATING AND CLASSIFYING LEVELS OF
expect, it also takes significantly longer time to simulate these
WAFER FAB DETAIL
models.
The key to managing the size of capacity/AMHS models,
A. Description of Levels of Wafer Fab Detail
and thus the execution time, is to interface the capacity model
If the fab model (i.e., a model representing the wafer fab) is and the AMHS model efficiently. There are currently two
separated into the level of detail of the capacity component and model interfaces. In coupled capacity/AMHS models, a ca-
the level of detail of the AMHS component, four basic fab mod- pacity model and an AMHS model are simulated separately
eling levels then result, as shown in Fig. 1. This figure describes and sequentially for the entire simulation and are then united
the goals, problem types, planning horizons, solution quality, by passing performance measures between them. As shown
etc. for each of the levels of detail. in the diagram in Fig. 2(c), the capacity model first generates
602 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 21, NO. 4, NOVEMBER 2008

Fig. 1. Fab modeling levels based on four different combinations of capacity and AMHS detail.

Fig. 2. Fab models at different levels of detail. (a) AMHS Model—Quadrant 2. (b) Capacity Model—Quadrant 3. (c) Coupled Capacity/AMHS Model—Quadrant
4. (d) Integrated Capacity/AMHS Model—Quadrant 4.

“From–To” material flow requirements and passes this data Fig. 2(d), the capacity model triggers the AMHS model to
into the AMHS model. The AMHS model then generates and execute the succeeding material handling activities that follow
passes “From–To” delivery times into the capacity model. On after a wafer completes a given production process. The AMHS
the other hand, in integrated capacity/AMHS models, a capacity model simulates the wafer delivery to the next process, and it
model and an AMHS model are combined into a single and then triggers the necessary production events in the capacity
unified simulation framework. As shown in the diagram in model once the lot is delivered.
JIMENEZ et al.: LEVELS OF CAPACITY AND MATERIAL HANDLING SYSTEM MODELING 603

TABLE I
FACTORS THAT AFFECT LEVEL OF CAPACITY/AMHS DETAIL

B. Development of Classification Method for Levels of development of a preliminary classification method and 2) eval-
Fab Modeling uation and modification of the preliminary method until valida-
tion issues are resolved.
There have been attempts to classify Fab models based As a part of the development of this classification scheme, we
on the increasing levels of capacity detail. For instance, the identify the factors affecting the levels of capacity/AMHS de-
classification scheme presented in [9] differentiates between tail, as shown in Table I. These are: modeling tool , model
spreadsheets, queuing networks (QN), and simulation models. type , factory level , AMHS model level , capacity
However, as was stated earlier, schemes like this do not elements , AMHS elements , study goals , and per-
offer sufficient information to be able to differentiate specific formance metrics . We define additional subfactors within
AMHS modeling elements. For example, they are incapable and to further specify capacity and AMHS modeling
of differentiating models with simple transportation delay components.
times and detailed representations of the interactions between Note that in studies with strategic goals, decision-making has
wafer lots, reticles, and AMHS movements. We have ex- a long-term projected impact on the system (usually reflected
tended the original concept of classification to the level of in more than six months). Examples of studies with strategic
detail of the capacity component and the level of detail of the goals include: finding number of tools, analyzing product mix,
AMHS component. conducting ramp-up analysis, identifying bottleneck worksta-
The classification method by the level of capacity/AMHS de- tions, designing the AMHS, among others. On the other hand,
tail was defined by using a procedure similar to that used in [10]. in studies with operational goals, decision-making has a short-
In general terms, this procedure consists of two major steps: 1) term projected impact on the system (usually reflected in less
604 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 21, NO. 4, NOVEMBER 2008

TABLE II
PROPOSED DIFFERENTIATING FACTORS FOR CAPACITY MODEL CLASSIFICATION

than six months). Examples of operational goals include: de- cies, downtimes, secondary resources, setups, tool dedication
termining lot sequencing rules, evaluating scheduling rules, an- policies, and setups. The following eight numbers indicate none
alyzing production operations concerning batching, setups, re- of the AMHS elements are modeled. Finally, the last two num-
work, lot sizes, reducing cycle times, and others. bers suggest that the study goals are strategic and the perfor-
To develop the preliminary classification method, we mance measure under study is cycle time.
analyzed approximately 30 published studies from the liter- The value of the proposed coding system is that cluster
ature which report primarily the applications of simulation analysis methodologies can be applied to discover groups of
to capacity planning and AMHS design. We coded each of studies whose levels of detail are similar. The idea is to define
these studies by using a multidimensional coding scheme the levels of capacity/AMHS detail within the preliminary
describing what factors and subfactors are considered in classification method according to the resulting groupings. The
the study. See the column labeled “Codes” in Table I for a development of the preliminary classification method is ac-
reference to the factor values. In our coding scheme, a digit complished through the following Agglomerative Hierarchical
and a digit represent the values taken by factor Clustering Algorithm [11].
and subfactor , respectively. The collection of all digits 1) Initialization: Start with groups, each containing one
constitutes the code. observation. An observation contains initially an individual
The following represents an example of how our reviewed simulation study. Therefore, if there are 30 studies, there
studies are coded: Shikalgar et al. (2003).- (3, 1, 3, 1, 1, 1, 0, are also groups. The factors affecting the level
1, 0, 0, 1, 1, 1, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2). of detail are the input variables to the analysis, thus each
In the example shown above, the first four numbers describe observation has dimensions, where is the total
a simulation study of the fab capacity at the factory level, but number of factors and subfactors. Groupings are formed on
without representing the AMHS. The following 13 numbers in the basis of similarities between input variables. As new
this code summarize the capacity components simulated, which groups are formed, observations may contain more than
include batching, dispatching rules, hot lots, order release poli- one study.
JIMENEZ et al.: LEVELS OF CAPACITY AND MATERIAL HANDLING SYSTEM MODELING 605

2) Computation of similarity matrix: An symmetric TABLE III


matrix is set up to indicate the similarity between pairs of PROPOSED DIFFERENTIATING FACTORS FOR AMHS MODEL CLASSIFICATION
groups. The similarity between two groups, say and
, is given by the Euclidean distance between their
-dimensional observations.
3) Combination of most similar groups in the similarity ma-
trix (i.e., Single Linkage Method): The pair of groups with
the minimum Euclidean distance is selected from the simi-
larity matrix. These groups are combined into a new group.
The similarity matrix is updated by deleting the records for
the selected groups and by adding a row and column indi-
cating the distances between the newly formed group and
the remaining groups.
4) Stopping rule: Steps 2 and 3 are repeated until an adequate
grouping of studies is formed. User judgment is required
to interpret the levels of capacity/AMHS detail from the
resulting groupings. The maximum number of repetitions
is times, which indicates that all studies will be in a
single group after the algorithm stops.
A clustering analysis is done by using the set of factors
, and (i.e., factors strongly related to the
levels of capacity detail). A second clustering analysis is done
by using the set of factors , and (i.e., factors
related to the level of AMHS detail). Conducting two separate
clustering analyses enables the development of a preliminary
classification scheme with an emphasis on a combined level of
capacity/AMHS detail, as proposed in Fig. 1. The results from
these clustering analyses suggest that the preliminary classifi-
cation method should consist of six levels of capacity detail,
six levels of AMHS detail, and two levels of capacity/AMHS
detail.
To assure the accuracy of the final method of classification by
the level of detail, further validation and modification of the pre-
liminary method is required. The modification process is done Winter Simulation Conference, among others. This literature
by an analysis of an additional set of 20 studies as follows. review has been restricted to publications dating from 1990 to
1) Individual studies are coded using our system for capacity/ 2005. The reason for this is because the applications of AMHS
AMHS model characteristics. to wafer fabs proliferated during this time period.
2) An individual study is assigned into one of the levels of
detail. If the study does not fit in an existing category, a
new level of detail is created to represent this model. III. RELATED LITERATURE REVIEW
Based on this validation analysis, we determined that the final
classification scheme consists of six levels of capacity detail, six A. Current State of Literature
levels of AMHS detail, and four levels of capacity/AMHS de-
A comprehensive list of our reviewed studies as classified
tail. The levels of capacity detail and the levels of AMHS de-
with our classification method by level of capacity/AMHS detail
tail are specified in Tables II and III, respectively. We used nu-
can be found in [13]. A summary of this information is presented
merical identifiers (i.e., 1-2-3-4-5-6) to represent the increasing
in Table IV. The numbers within each cell are the numbers of
levels of capacity detail, and alphanumerical identifiers (i.e.,
A-B-C-D-E-F) to represent the increasing levels of AMHS de- published articles fitting our classification scheme. This table is
tail. The levels of capacity/AMHS detail can be divided pri- not intended to represent every semiconductor manufacturing
marily into two groups, one representing integrated models and study published but instead to be a representative sample. In
the other representing coupled models. The factors shown in particular, many studies are available on pure capacity models
these tables represent the proposed framework for classifying (column in Quadrant 3) that we elected not to include in our
fab models by level of detail. classification since they provided little additional information.
The major publications that we referenced for this study A more comprehensive list of articles related to pure capacity
include the IEEE TRANSACTIONS ON SEMICONDUCTOR models can be found in [14] and [15].
MANUFACTURING, IIE Transactions, International Journal Some cells in Quadrant 4 contain two numbers to differentiate
of Production Research, Simulation, Robotics & Computer-In- between the model interfaces. The first number corresponds to
tegrated Manufacturing, and Semiconductor International. We the number of publications using the integrated interface, and
also referenced conference proceedings from the IEEE/SEMI the second number corresponds to those publications using the
Advanced Semiconductor Manufacturing Conference and the coupled interface.
606 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 21, NO. 4, NOVEMBER 2008

TABLE IV
CURRENT STATE OF LITERATURE ACCORDING TO CLASSIFICATION METHOD BY LEVEL OF CAPACITY AND AMHS DETAIL (I.E., 1-2-3-4-5-6 STANDS FOR
LEVELS OF CAPACITY DETAIL SHOWN IN TABLE II, AND A-B-C-D-E-F STANDS FOR LEVEL OF AMHS DETAIL SHOWN IN TABLE III)

What Table IV demonstrates is a lack for tools that allow tenance [3]. This tool is often the starting point for a more com-
for the combination of detailed capacity models with detailed plex fab modeling effort. In application, spreadsheets are often
AMHS models as illustrated by the small number of studies in used in strategic studies, especially if the time window allocated
Quadrant 4. The major deterrents for such models are the large for the project is small. In particular, these models help deter-
amount of CPU time needed to get statistically usable results mine the adequate number of workstations in the fab (for ex-
and the difficulty in creating these detailed models. ample, see [17]–[19]).
The remaining part of this section presents a literature review
associated with and organized according to increasing levels of C. Queuing Network Models (QN)
capacity and AMHS detail. Our review seeks to explain, in gen- Queuing network models (classified as ) estimate the fab
eral terms, the relevant model characteristics that can potentially capacity by using time-dependent performance measures such
affect the selection of the level of detail. as cycle time and work-in-progress (WIP). Unlike spreadsheets,
In Table IV, we show the current state of the literature ac- QN models are capable of representing the variability in the
cording to the classification method by level of capacity and system. QN models are also computational efficient. However,
AMHS detail (i.e., 1-2-3-4-5-6 stands for the levels of capacity the capacity predictions generated with QN models can be only
detail shown in Table II, and A-B-C-D-E-F stands for the levels used as approximations because queuing models cannot capture
of AMHS detail shown in Table III). the fab complexities accurately. QN models have been used in
strategic studies, especially for rough-cut capacity analysis (for
B. Spreadsheet Capacity Models example, see [20]) and for optimizing the equipment portfolio
Spreadsheet models (classified as ) provide a static esti- (for example, see [21] and [22]).
mation of the capacity of the fab by using performance metrics It is very unlikely that either spreadsheets or QN models can
such as throughput and equipment efficiency. The mathematical account for the effects of the material handling system on ca-
models used within spreadsheets can be very accurate for some pacity, as the AMHS is very dynamic and complex. This can
capacity loss agents, such as downtimes, yields, and setups [5]. be noted by the lack of studies in this area, e.g., levels 2B-2E,
Yet, capacity predictions tend to be inflated because other more 3B-3E.
complex capacity loss agents are modeled very crudely (as was
shown in [9] and [16]). D. Reduced Simulation-Based Capacity (RSC) Models
Despite their inaccuracy, spreadsheets are popular because RSC models (classified as ) study high-utilized worksta-
they provide fast computing times, ease of use, and ease of main- tions without having to model the entire fab in detail. The work-
JIMENEZ et al.: LEVELS OF CAPACITY AND MATERIAL HANDLING SYSTEM MODELING 607

stations that add little value to total cycle time are reduced to the the supply of AMHS vehicles is assumed infinite (i.e., level of
lowest possible detail without losing too much model accuracy. detail ), the lot waiting time for vehicles can be ig-
Thus, RSC models generate very accurate predictions of the CT nored. Thus, this approach helps reduce the simulation effort
within the workstation that is modeled in detail. In general, RSC because “From–To” delivery times can be calculated easily, i.e.,
simulations execute relatively fast, as was shown in [23]. They by computing delivery times as a function of distances between
are also easier to update and provide a starting point for building locations and the speed of the vehicles. Using such simple esti-
more sophisticated simulation models [24]. In application, RSC mation of delivery time will result in very optimistic predictions
models are used in operational studies, especially for the anal- of cycle time, which can then be used as lower bounds on the
ysis of system bottlenecks. Most of the related literature deals actual fab cycle times.
with how to abstract unimportant workstations (for example, see The use of RSC models with AMHS time delays (classified
[23] and [25]–[27]). as ) has also been reported. In particular, they are used for
analyzing the operations of bottleneck workstations and for an-
E. Basic Simulation-Based Capacity (BSC) Models alyzing the fab and AMHS very crudely [42].
BSC models (classified as ) simulate the basic and the
most representative capacity components, e.g., batching, setups, H. AMHS Models
downtimes, dispatching rules, and rework. Modeling the system AMHS models measure key performance measures such as
core with such constructs enables relatively accurate predictions transport times, lot-waiting times for free vehicles, delivery
of the entire fab capacity, provided that the AMHS does not times, stocker robot utilization, vehicle utilization, vehicle lost
cause any significant loss in capacity. However, the cycle time capacity, etc. AMHS models execute very slowly. However,
will be generally underestimated, especially if these estimations these models produce quality predictions provided that the
are compared to those produced by the capacity/AMHS model. “From–To” material flow requirements are represented accu-
BSC models have traditionally supported the configuration and rately [39]. Historical data or capacity models can be used to
scheduling of production processes (for example, see [28]–[30], estimate material flow requirements, yet model accuracy suffers
and [31]) and also the strategic analysis of wafer fabs (for ex- as these tools can only partially explain the AMHS dynamics
ample, see [32] and [33]). [43]. Previous experimentation has also demonstrated that
quality of prediction depends strongly on modeling the vari-
F. Full Simulation-Based Capacity (FSC) Models ability of the material flow requirements accurately by using
FSC models (classified as ) have the necessary details to probability distributions explaining large variance components,
virtually emulate the actual fab’s production operations, thus e.g., hyper-exponential distribution [44].
they can predict capacity accurately. The use is suggested for AMHS models (i.e., models) are used extensively
configuring complex production processes, such as those pro- for AMHS design. More specifically, these models are used in
cesses involving complicated batching policies, lot sequencing the design of AMHS layout (for example, see [8], [45], [46],
rules, time-bound process sequences, sequence-dependent and [47]), the analysis of the hallway transportation system (for
setups, reticles, etc. (for example, see [34]–[37], and [38]). example, see [7], [44], [48], and [49]), determination of vehicle
The downside is the significant amount of time required to and storage capacities (for example, see [50] and [51]), anal-
develop, validate, execute, analyze, and maintain these models. ysis of transportation technology (for example, see [52]), evalu-
However, the time to do this is significantly less than the times ation of vehicle control algorithms (for example, see [53]–[55]),
required for the capacity/AMHS model. and the analysis of AMHS delivery paradigms (for example, see
[56]–[58]), among others.
G. Capacity Models With AMHS Time Delays
At this level of detail, the capacity model includes a delay I. RSC/Intra-Bay AMHS Models
block to account for the transfer time of wafers between pro- RSC models with intra-bay AMHS models (classified as )
cesses. These models help address the same goals as those ad- help analyze an entire processing area within the fab and its
dressed by capacity models without the AMHS construct but intra-bay AMHS. These models provide a detailed analysis of
with higher solution quality and about the same computational the factors affecting the tools–AMHS interactions (e.g., vehicle
effort. Experiments have shown that the CT predictions can be capacity, tool buffer capacity, process area layout, etc.), so they
near to the predictions produced by the capacity/AMHS model are therefore used widely in intra-bay AMHS design (for ex-
provided that the “From–To” delivery times are represented ac- ample, see [59] and [60]). Unfortunately, up to this point, it is
curately [39]. unclear whether the time gains by using RSC/Intra-bay AMHS
The supply of vehicles within a capacity model with AMHS models can offset the loss in predictive quality caused by the in-
time delays can be assumed to be limited (i.e., level of detail accurate representations of the rest of the factory.
). An assumption like this implies that data such as
the time that lots spent waiting to be picked up by free vehicles J. Capacity/AMHS Models
has to be specified in the model. The problem with assuming Capacity/AMHS models (level of detail ) contain
a limited supply of vehicles is that lot-waiting time for vehi- detailed modeling constructs for the production operations, the
cles can only be computed crudely through a static model, so production equipment, and the AMHS. The performance pre-
model accuracy will suffer. Refer to [40] and [41] for exam- dictions produced by these models can be assumed to be repre-
ples of models within this level of detail. On the other hand, if sentative of the actual system performance, thus managers can
608 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 21, NO. 4, NOVEMBER 2008

potentially make more effective decisions. These models enable (TBF) and time to repair (TTR) are exponentially distributed.
higher level and integrated analysis of the equipment-AMHS Batching, setups, and reticles are also defined for the required
interactions [61]. For example, the contributions of inter-bay workstations. For planning purposes, the fab operates 24 hours
AMHS technology on decreasing cycle time, delivery time, and per day, 365 days per year. Lots are released at a constant rate.
costs were analyzed in [62] and [63]. In addition, the effect of The time between lot releases is deterministic.
lot sizes and priority lots on decreasing AMHS delivery times The AMHS model represents the machines in a 24-bay
was analyzed in [64]. layout. There are two stockers with a capacity of 800 lots for
The problem is that it takes too long to simulate and an- each bay, each consisting of two input–output ports. There is a
alyze all of the desired system configurations within an inte- stocker robot for each stocker. There is one main transportation
grated model. The total computational time of an integrated ca- loop responsible for the inter-bay AMHS and 24 bay loops
pacity/AMHS model is proportional to the execution time of the responsible for the intra-bay AMHS. In addition, there is one
capacity model, the execution time of the AMHS model, and a transportation loop dedicated to deliver reticles. Each bay has
large overhead (i.e., time to manage longer simulation lists, time 24 tool positions, each consisting of two I/O ports. Number,
to synchronize capacity model and AMHS model, etc.). To re- speed, loading times, unloading times, TTR, and TBF are
duce the overhead times practitioners have proposed coupling a defined for the vehicles in each transportation loop.
capacity model with AMHS time delays and an AMHS model. Users have the choice to run the SEMATECH Fab Model in
Experiments have shown that more computational efficient pre- one of the following modes.
dictions are generated by coupling these two abstract models 1) As a separate AutoSched AP (ASAP) Capacity Model.
(for example, see [1] and [65]). However, their validity depends 2) As a separate AutoMod AMHS Model.
on using system assumptions that are consistent in both models, 3) As a single Integrated Model consisting of the ASAP
thus each simulation module will probably need recalibration if Capacity Model and the AutoMod AMHS Model. These
assumptions change. Simulation studies using coupled models models are combined by using Brook’s Model Commu-
include [62], [65], and [66]. nication Module; see [70] for the details on the Model
Communication Module (MCM).
IV. ANALYSIS OF CAPACITY/AMHS MODELS 4) As two Coupled Models consisting of the ASAP Model
and the AutoMod Model. These models are united by using
A. Description of Simulation Models “From–To” Tables.
In this section, we will attempt to demonstrate via a sim- The coupling of ASAP/AutoMod models is achieved through
ulation-based abstracting methodology that coupled capacity/ the following process.
AMHS models can produce system predictions that are not sta- 1) pilot runs are done with a pure capacity model (i.e.,
tistically differentiable from those produced by the integrated a model without any AMHS simulation components) to
capacity/AMHS models. Thus, coupled models work more ef- approximate the “From–To” number of wafer moves.
fectively in simulation studies with fixed budget and short anal- 2) The resulting “From–To” Table with the number of moves
ysis times, as their computational times are significantly lower is fed into the AMHS Model.
than those observed by executing the detailed integrated models. 3) pilot runs are done with the AMHS Model to produce
Our objective in creating this methodology is to develop a fast “From–To” delivery times.
and accurate surrogate for a fully integrated capacity/AMHS fab 4) The resulting “From–To” Delivery Time Table is input into
simulation model. It has been proposed that in many cases a a delay simulation module within ASAP to simulate the
queuing approximation might be a likely alternative. While we AMHS activities as a black box. This forms the abstract
believe a queuing model might be capable of accurately pre- coupled model.
dicting the cycle time of a capacity simulation, and that a similar In this paper, pilot runs. Each “From–To” Table
queuing approximation might also be developed for the AMHS is thus based on the average of the five replications. It is as-
system, we do not see at this time how these two approximations sumed that both the number of moves and the delivery times
could be combined into an accurate predictor of a fully detailed are not stochastic, so the use of probability distributions is un-
fab simulation model. This paper therefore restricts the method- necessary. This assumption works effectively for the parameter
ological development to the improvement of simulation-based settings used in this case study, but it could change if the coupled
analysis methods. model is run with parameters that produce greater variability in
To carry out this experimentation, we used an existing simu- the From–To data.
lation model of a 300-mm wafer fab [67]; this model is currently
available at the SEMATECH website (www.sematech.org). The B. Experimentation and Analysis
SEMATECH 300-mm Fab Model consists of a capacity model To illustrate the accuracy and efficiency of abstract cou-
built in AutoSched AP [68] and an AMHS model built in Au- pled models in relation to the detailed integrated models, we
toMod [69]. produced Cycle Time-Throughput (CT-TH) Curves for the
The capacity model includes ten lot types with each lot con- system described above. A CT-TH curve is a tool often used by
taining 25 wafers. The processing route consists of approxi- managers to determine the projected cycle time for a desired
mately 316 operations. There are 60 workstations, each work- throughput level. In these curves, throughput (TH) is often
station consisting of identical machines operating in parallel. represented in terms of bottleneck capacity. The relationship
Interruptions due to breakdowns and preventive maintenance between bottleneck capacity and TH is empirically determined
are defined for these workstations. The time between failures by running simulations at different points along the curve and
JIMENEZ et al.: LEVELS OF CAPACITY AND MATERIAL HANDLING SYSTEM MODELING 609

Fig. 3. Projected fab cycle times estimated by two different modeling methods.

noting utilizations of all equipment. Refer to [71] for more Table V also shows the computational time in seconds that
details concerning CT-TH curves. coupled models and integrated models take to run one simula-
We produced cycle time (CT) estimates at the design points tion replication of 100 simulation days. These results suggest
% % % and % (i.e., percent capacity of the that the coupled capacity/AMHS models executes approxi-
system bottleneck that will generate a desired throughput mately 98% faster in relation to the integrated capacity/AMHS
level). Each CT point is based on ten independent replications model (i.e., column f in Table V). The computational time
of 100 simulation days; for each run, we cleared statistics after gained by using this simulation method can significantly ex-
the first 50 days to eliminate initialization bias. The results of tend the capabilities of the experimentation. For instance, if
these simulations are depicted in Fig. 3. The points with the a simulation project is allocated with the time and budget to
symbol were generated from simulation experiments using the estimate one point of the CT-TH curve with the integrated
integrated model, whereas the points with the symbol were capacity/AMHS model, then we could estimate between 65
generated from simulation experiments using a coupled model. to 87 points (i.e., column g in Table V) using the coupled
Table V shows the relative difference and the 95% confidence capacity/AMHS model within the same budget and time re-
intervals of the mean difference between the CT estimates quirements. With this larger experimental capacity, we could
produced by the coupled and integrated models (see [72] for analyze more system configurations by building more CT-TH
the technical details concerning confidence intervals). curves, or we could increase the accuracy of the estimates by
Our results suggest that the coupled model and the integrated increasing the number of replications for each data point in the
model can produce cycle time estimations that are almost iden- curve.
tical. Because the abstraction components used within coupled The accuracy and computational efficiency observed in this
capacity/AMHS models (i.e., “From–To” delivery times) will case study are specific to the SEMATECH Fab Model described
typically underestimate system variability, the loss in predictive above. However, the SEMATECH Fab Model is a reasonably
accuracy observed by using this more abstract modeling tech- accurate model of a somewhat simplified, generic fab and even
nique increases slightly as the bottleneck utilization approaches though the results do not exactly represent other fab models,
full capacity (i.e., from 0.09% at the 70% point to 0.59% at the similar levels of accuracy and efficiency are expected if the pro-
97% point). However, despite such small loss of accuracy, the posed procedures for coupling capacity/AMHS models are ap-
95% confidence intervals indicate that we cannot statistically plied. In fact, more realistic fab models would contain greater
differentiate between coupled models and integrated models at complication and applying our approach should therefore pro-
any of the four design points. vide even greater time reductions.
610 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 21, NO. 4, NOVEMBER 2008

TABLE V
EXPERIMENTAL RESULTS COMPARING ACCURACY AND COMPUTATIONAL EFFICIENCY BETWEEN COUPLED MODELS AND INTEGRATED MODELS

Fig. 4. Recommended levels of modeling for integrated capacity analysis and AMHS design.

In general, the coupling modeling procedure can be applied abstract CPU intensive modeling components into efficient and
effectively to any high-volume manufacturing system with an accurate simulation modules that can be easily incorporated
AMHS. It would be impractical to model the capacity system into existing fab models.
separately from the AMHS if the number of AMHS requests
is nominal, as in low-volume manufacturing systems. However, V. RECOMMENDED LEVELS OF WAFER FAB MODELING
the application of coupled models becomes significant as the It can be assumed that refereed articles published in the liter-
increase of AMHS transactions affects model complexity and ature have been carefully reviewed in terms of the capability
thus computational time. of the simulation model to accomplish the underlying study
The efficiency improvements obtained with the coupled goals efficiently and accurately. If this is the case, the pub-
model are not exclusive to Brook’s ASAP and AutoMod. The lished literature becomes a reliable source for determining the
coupled modeling procedure could be easily implemented in best matching level of capacity and AMHS modeling for typical
other simulation languages since the principle behind it is to semiconductor wafer fab problems.
JIMENEZ et al.: LEVELS OF CAPACITY AND MATERIAL HANDLING SYSTEM MODELING 611

Fig. 5. Recommended levels of modeling for pure capacity analysis.

We summarized the information provided by our literature not modeled). For studies seeking rough approximations (e.g.,
review into the diagrams shown in Figs. 4 and 5. These flow di- within 25% of the actual value) and little computational time
agrams present a series of questions that can be used to identify (e.g., less than an hour), the literature recommends using spread-
the requirements of the simulation study (i.e., questions are in- sheet models if throughput, equipment utilization, availability,
side boxes with solid borders). These questions are followed by etc., will be measured, or QN models if cycle time and WIP
a recommended level of capacity and AMHS detail fitting the need to be estimated. On the other hand, BSC models should
identified study (i.e., recommended levels of detail are inside be used instead if higher quality predictions are sought. Exper-
boxes with dotted borders). imentation has shown that in some problem instances pure ca-
The diagram in Fig. 4 attempts to recommend the levels of pacity models produce results that are approximately 15% less
detail within the context of capacity analysis and AMHS de- accurate than the integrated capacity/AMHS model (see [13] for
sign. As was demonstrated earlier in this paper, coupled ca- more details). In some cases, the accuracy of BSC models is
pacity/AMHS models will be preferred in studies with fixed not always sufficient to estimate production performance, espe-
budget and time constraints because they offer the same ac- cially when the fab capacity is constrained by complex manufac-
curacy and precision than detailed integrated models, but with turing processes (i.e., processes involving reticles, complicated
greater computational efficiency. This means that using cou- batching policies, time-bound production sequences, etc.). For
pled capacity/AMHS models allow for more experimentation such complex system conditions, the use of FSC models is rec-
and analysis within the time allocated for the study. However, ommended, as this level of detail represents the system more
coupled capacity/AMHS models are not capable of estimating accurately.
any measures describing the performance of AMHS (i.e., de-
livery times, AMHS utilizations, traffic congestion, etc.). For VI. CONCLUSION
such cases, the use of detailed AMHS models is recommended. In this paper, we described a method for classifying wafer
Experimentation has demonstrated that AMHS models provide fab models by their level of model detail. The six levels of
results that are near to the results produced by the integrated ca- detail of the capacity construct (i.e., material flow requirements,
pacity/AMHS model and the gain in computational time is sig- spreadsheet models, queuing network models, RSC models,
nificant (see [1] for more details). On the other hand, the inte- BSC models, FSC models) and the six levels of detail of the
grated capacity/AMHS models should be used over abstractions AMHS construct (i.e., no AMHS elements, AMHS delay times
whenever the study demands greater precision and accuracy, and with limited supply of AMHS vehicles, AMHS times with
at the same time, the budget and time allocated to carry out the unlimited supply of AMHS vehicles, AMHS models without
analysis are nearly unlimited. vehicle interference, AMHS models with vehicle interference,
The diagram in Fig. 5 depicts the recommended levels of de- full AMHS model) that we proposed were developed from past
tail in the context of a pure capacity analysis (i.e., the AMHS is published simulation studies. While it was designed for wafer
612 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 21, NO. 4, NOVEMBER 2008

fab systems, our classification framework by the level of detail [13] J. Jimenez, Simulation Modeling Levels to Support Integrated Capacity
can be applied in a more general context to other manufacturing and AMHS Decision Making in Semiconductor Wafer Fabs.. Tempe:
Arizona State Univ., to be published.
settings, especially those that use any sort of AMHS. [14] F. Chance and J. Robinson, FabTime’s Bibliography. FabTime. , CA,
An extensive literature review effort indicates that sufficiently 2005 [Online]. Available: http://www.fabtime.com/bibliogr.shtml,
[Online]. Available:
detailed fab models have not been frequently used due to the [15] J. W. Fowler, J. K. Cochran, and S. M. Horng, in Abstract MASMLAB
significant amount of time that it takes to run these models. Bibliography Website. Arizona State Univ., Tempe, AZ, May 2003 [On-
line]. Available: http://www.eas.asu.edu/~masmlab/Biblio.html, [On-
This review also indicates that the use of fab model abstrac- line]. Available:
tions has been also very limited. The case study presented in [16] S. S. Johal, “Non-linearity and randomness in a semiconductor wafer
fab,” in Proc. 1996 IEEE/SEMI Adv. Semiconductor Manufacturing
this paper demonstrated that working with a coupled capacity/ Conf., pp. 17–22.
AMHS model generated predictions that are not statistically dif- [17] J. D. Witte, “Using capacity modeling techniques in semiconductor
manufacturing,” in Proc. 1996 IEEE/SEMI Adv. Semiconductor Man-
ferent than those generated by a detailed integrated capacity/ ufacturing Conf., pp. 31–35.
AMHS model, and the savings in computational time are very [18] J. Neudorff, “Static capacity analysis using microsoft visual basic,” in
Proc. Int. Conf. Semiconductor Manufacturing Oper. Modeling and
significant. Thus, coupled models have been shown to be poten- Simulation, 1999, pp. 207–212.
tially efficient and accurate. It is our hope that the results ob- [19] Y.-C. Chou and R.-C. You, “A resource portfolio planning method-
tained in this paper motivate further the development and use of ology for semiconductor manufacturing,” Int. J. Adv. Manuf. Technol.,
vol. 18, pp. 12–19, 2001.
abstraction methodologies. [20] H. Chen, J. M. Harrison, A. Mandelbaum, A. V. Ackere, and L. M.
For future work, we plan to use the classification method by Wein, “Empirical evaluation for a semiconductor manufacturing wafer
fabrication,” Oper. Res., vol. 36, pp. 202–215, Mar.-Apr. 1988.
the level of capacity/AMHS detail in a methodology that will [21] D. P. Connors, G. E. Feigin, and D. D. Yao, “A queuing network
help model builders find an optimal abstraction level. As was model for semiconductor manufacturing,” IEEE Trans. Semicond.
Manuf., vol. 9, no. 3, pp. 412–427, Aug. 1996.
discussed in [3], this work requires the development of an IP [22] W. J. Hopp, M. L. Spearman, S. Chayet, K. L. Donohue, and E. S.
model using binary decision variables to represent whether a Gel, “Using an optimized queuing network model to support wafer fab
design,” IIE Trans., vol. 34, pp. 119–130, 2002.
level of capacity/AMHS detail is selected. The objective func- [23] Y. F. Hung and R. C. Leachman, “Reduced simulation models of wafer
tion will minimize the total loss in result accuracy subject to fabrication facilities,” Int. J. Prod. Res., vol. 37, pp. 2685–2701, 1999.
[24] Y. Shen, “Robust capacity modeling in semiconductor manu-
constraints specifying expected project deadlines. The key for facturing,” Semiconduct. Intl. Jul. 2002 [Online]. Available:
this research is to create the algorithms that will predict sim- http://www.reed-electronics.com/semiconductor/index.asp?layout=ar-
ticle&articleid=CA231596&pubdate=7/1/2002&spacedesc=webex,
ulation CPU times and solution accuracy for each level of fab [Online]. Available:
detail under consideration. With an optimal level of detail, it will [25] O. Rose, “WIP evolution of a semiconductor factory after a bottleneck
workcenter breakdown,” in Proc. 1998 Winter Simulation Conf., pp.
hopefully take less time to build and execute fab models. 997–1003.
[26] O. Rose, “Estimation of the cycle time distribution of a wafer fab by a
REFERENCES simple model,” in Proc. 1999 SMOMS, pp. 133–138.
[27] O. Rose, “Why do simple wafer fab models fail in certain scenarios?,”
[1] J. Shelton, “Fabless vision,” Future Fab Int. vol. 14, Nov. in Proc. 2000 Winter Simulation Conf., pp. 1481–1490.
2003 [Online]. Available: http://www.future-fab.com/docu- [28] L. M. Wein, “Scheduling semiconductor wafer fabrication,” IEEE
ments.asp?grID=208&d_ID=1635#, [Online]. Available: Trans. Semicond. Manuf. , vol. 1, no. 3, pp. 115–130, Aug. 1988.
[2] J. S. Pettinato, “An ITRS view on future factory integration, de- [29] S. Brown, F. Chance, J. W. Fowler, and J. Robinson, “A centralized
sign, and capabilities,” Future Fab Int. vol. 16, Feb. 2004 [Online]. approach to factory simulation,” Fut. Fab Int., vol. 1, pp. 83–86, 1997.
Available: http://www.future-fab.com/documents.asp?d_ID=3016, [30] J. W. Fowler, S. Brown, H. Gold, and A. Shoeming, “Measurable im-
[Online]. Available: provements in cycle-time-constrained capacity,” in Proc. 1997 IEEE/
[3] K. Lee and P. A. Fishwick, “OOPM/RT: A multimodeling method- UCS/SEMI Int. Symp. Semiconductor Manufacturing, pp. 21–24.
ology for real-time simulation,” ACM Trans. Modeling and Computer [31] M. Janakiram and J. R. Morrison, “Capacity planning and study of
Simulation, vol. 9, pp. 141–170, Apr. 1999. scheduling policies using simulation at Motorola’s ACT fab,” in Proc.
[4] F. Chance, J. K. Robinson, and J. W. Fowler, “Supporting manufac- Int. Conf. Semiconductor Manufacturing Oper. Modeling Simulation,
turing with simulation: Model design, development, and deployment,” 1999, pp. 201–206.
in Proc. 1996 Winter Simulation Conf., pp. 114–121. [32] N. S. Grewal, T. M. Wulf, A. C. Bruska, and J. K. Robinson, “Inte-
[5] J. Robinson, J. Fowler, and E. Neacy, Capacity loss factors in semicon- grating targeted cycle time reduction into the capital planning process,”
ductor manufacturing. FabTime CA, Mar. 2003 [Online]. Available: in Proc. 1998 Winter Simulation Conf., pp. 1005–1010.
http://www.fabtime.com/files/CapPlan.pdf, [Online]. Available: [33] R. Stafford and J. Rimington, “Use of goal seeking for capacity analysis
[6] G. Nadoli and D. Pillai, “Simulation in automated material handling with semiconductor simulation models,” in Proc. Int. Conf. Semicon-
systems design for semiconductor manufacturing,” in Proc. 1994 ductor Manufacturing Oper. Modeling Simulation, 1999, pp. 213–218.
Winter Simulation Conf., pp. 892–899. [34] N. S. Grewal, T. M. Wulf, A. C. Bruska, and J. K. Robinson, “Val-
[7] G. T. Mackulak, F. P. Lawrence, and T. Colvin, “Effective simulation idating simulation model cycle time at Seagate technology,” in Proc.
model reuse: A case study for AMHS modeling,” in Proc. 1998 Winter 1999 Winter Simulation Conf., pp. 843–849.
Simulation Conf., pp. 979–984. [35] L. Mönch, M. Prause, and V. Schmalfuss, “Simulation-based solution
[8] G. T. Mackulak and P. Savory, “A simulation-based experiment for of load-balancing problems in the photolithography area of a semi-
comparing AMHS performance in a semiconductor fabrication fa- conductor wafer fabrication facility,” in Proc. 2001 Winter Simulation
cility,” IEEE Trans. Semiconduct. Manuf., vol. 14, no. 3, pp. 273–280, Conf., pp. 1170–1177.
Aug. 2001. [36] L. Mönch, O. Rose, and R. Sturm, “A simulation framework for the per-
[9] M. Baudin, V. Mehrotra, B. Tullis, D. Yeaman, and R. A. Hughes, formance assessment of shop-floor control systems,” Simulation, vol.
“From spreadsheets to simulations: A comparison of analysis methods 79, pp. 163–170, Mar. 2003.
for IC manufacturing performance,” in Proc. 1992 IEEE/SEMI Int. [37] L. W. Schruben and T. M. Roeder, “Fast simulations of large-scale
Semiconduct. Manuf. Science Symp., pp. 94–99. highly congested systems,” Simulation, vol. 79, pp. 115–125, Mar.
[10] N. E. Ozdemirel, G. T. Mackulak, and J. K. Cochran, “A group tech- 2003.
nology classification and coding scheme for discrete manufacturing [38] S. Shikalgar, D. Fronckowiak, and E. MacNair, “Application of cluster
simulation models,” Int. J. Prod. Res., vol. 31, pp. 579–601, 1993. tool modeling to a 300 mm fab simulation,” in Proc. 2003 Winter Sim-
[11] R. A. Johnson and D. W. Wichern, Applied Multivariate Statistical ulation Conf., pp. 1394–1397.
Analysis, 5th ed. Englewood Cliffs, NJ: Prentice–Hall, 2002, pp. [39] N. G. Pierce and R. Stafford, “Simulating AMHS performance for
679–694. semiconductor wafer fabrication,” in Proc. 1995 Int. Symp. Semicon-
[12] R. Uzsoy, C. Y. Lee, and L. A. Martín-Vega, “A review of production ductor Manufacturing, pp. 165–170.
planning and scheduling models in the semiconductor industry part I: [40] R. Wright, C. Cunningham, K. Benhayoune, E. Campbell, V. Swami-
System characteristics, performance evaluation and production plan- nathan, and R. White, “300 mm factory layout and automated materials
ning,” IEE Trans., vol. 24, pp. 47–60, Sep. 1992. handling,” Solid State Technol., vol. 42, pp. 35–42, Dec. 1999.
JIMENEZ et al.: LEVELS OF CAPACITY AND MATERIAL HANDLING SYSTEM MODELING 613

[41] J. Hunter, D. Delp, D. Collins, and J. Si, “Understanding a semicon- [65] C. D. DeJong and S. A. Fischbein, “Integrating dynamic fab capacity
ductor process using a full-scale model,” IIE Trans. Semiconduct. and automation models for 300 mm semiconductor manufacturing,” in
Manuf., vol. 15, pp. 285–289, May 2002. Proc. 2000 Winter Simulation Conf., pp. 1505–1509.
[42] A. Peikert, S. Brown, and J. Thoma, “A rapid modeling technique [66] C. D. DeJong and S. P. Wu, “Simulating the transport and scheduling
for measurable improvements in factory performance,” in Proc. 1998 of priority lots in semiconductor factories,” in Proc. 2002 Winter Sim-
Winter Simulation Conf., pp. 1011–1015. ulation Conf., pp. 1387–1391.
[43] S. Wu, J. Rayter, I. Paprotny, G. T. Mackulak, and J. Yngve, “Increasing [67] SEMATECH (2002). Fab Simulation Modeling Software [On-
first pass accuracy of AMHS simulation output using legacy data,” in line]. Available: http://ismi.sematech.org/modeling/simulation/index.
Proc. 1999 Winter Simulation Conf., pp. 784–789. htm#models [Online]. Available:
[44] S. Sokhan-Sanj, G. Gaxiola, G. T. Mackulak, and F. B. Malmgren, “A [68] “Brooks-pri automation,” in AutoSched AP 7.2 User’s Guide, Salt Lake
comparison of the exponential and the hyperexponential distributions City, UT, Oct. 2002.
for modeling move requests in a semiconductor fab,” in Proc. 1999 [69] “Brooks automation,” in AutoMod User’s Manual, Salt Lake City, UT,
Winter Simulation Conf., pp. 774–778. Mar. 2001, vol. 1–2.
[45] T. D. Colvin and G. T. Mackulak, “A concurrent engineering approach [70] “Brooks-pri automation,” in AutoSched AP 7.2 Customization Guide,
for semiconductor handling system design: A case study,” in Proc. Int. Salt Lake City, UT, Oct. 2002.
Conf. Ind. Eng. Applications, 1997, pp. 123–128. [71] J. W. Fowler, S. Park, G. T. Mackulak, and D. L. Shunk, “Efficient
[46] T. D. Colvin, F. P. Lawrence, and G. T. Mackulak, “Soft simulation cycle time-throughput curve generation using a fixed sample size pro-
crucial for new automated fab decisions,” Solid State Technol., vol. 41, cedure,” Int. J. Prod. Res., vol. 39, pp. 2595–2613, 2001.
pp. 161–168, Jun. 1998. [72] D. C. Montgomery, G. C. Runger, and N. F. Hubele, Eng. Stat., 3rd
[47] T. D. Colvin, A. R. Jones, L. S. Hennessy, and G. T. Mackulak, “Fab ed. New York: Wiley, 2004, pp. 198–260.
design for 300 mm wafer handling,” Eur. Semiconduct., pp. 25–27,
May 1998.
[48] G. Nadoli and M. Rangaswami, “An integrated modeling methodology
for material handling systems design,” in Proc. 1993 Winter Simulation
Conf., pp. 785–789.
[49] I. Paprotny, W. Zhao, and G. T. Mackulak, “Reducing model creation
cycle time by automated conversion of a CAD AMHS layout,” in Proc.
1999 Winter Simulation Conf., pp. 779–783. Jesus A. Jimenez received the B.S. and M.S. de-
[50] G. Cardarelli and P. M. Pelagagge, “Simulation tool for design and grees in industrial engineering from The University
management optimization of automated interbay material handling and of Texas at El Paso, El Paso, TX, in 1997 and 1999,
storage systems for large wafer fabs,” IEEE Trans. Semicond. Manuf., respectively, and the Ph.D. degree in industrial
vol. 8, no. 1, pp. 44–49, Feb. 1995. engineering from Arizona State University, Tempe,
[51] G. Cardarelli, P. M. Pelagagge, and A. Granito, “Performance analysis AZ, in 2006.
of automated interbay material handling and storage systems for large He is an Assistant Professor in the Industrial Engi-
wafer fabs,” Robot. Comp. Integr. Manuf., vol. 12, pp. 227–234, 1996. neering program of the Ingram School of Engineering
[52] I. Paprotny, J.-Y. Shiau, Y. Huh, and G. T. Mackulak, “Simulation at Texas State University-San Marcos, San Marcos,
based comparison of semiconductor AMHS alternatives: Continuous
flow vs. overhead monorail,” in Proc. 2000 Winter Simulation Conf., TX. His research interests are in the modeling and
pp. 1333–1338. analysis of manufacturing systems, especially semi-
[53] N. Bahri and R. J. Gaskins, “Automated material handling system conductor manufacturing; automated material handling systems; the develop-
traffic control by means of node balancing,” in Proc. 2000 Winter ment of manufacturing applications of operations research, computer simula-
Simulation Conf., pp. 1344–1346. tion, and statistics; and the reduction of time required to build, experiment, and
[54] J. T. Lin, F.-K. Wang, and P.-Y. Yen, “Simulation analysis of dis- analyze complex computer simulation models.
patching rules for an automated interbay material handling system in Dr. Jimenez is member of the Institute for Operations Research and the Man-
wafer fab,” Int. J. Prod. Res., vol. 39, pp. 1221–1238, 2001. agement Sciences (INFORMS) and the Institute of Industrial Engineers (IIE).
[55] M. Fukunari, S. Rajanna, R. J. Gaskins, and M. E. Sparrow, “Data-
based node penalties in a path-finding algorithm in an automated ma-
terial handling system,” in Proc. 2002 Winter Simulation Conf., pp.
1383–1386.
[56] N. Bahri, J. Reiss, and B. Doherty, “A comparison of unified vs. segre- Gerald T. Mackulak received the B.Sc., M.Sc., and
gated automated material handling systems for 300 mm fabs,” in Proc. Ph.D. degrees industrial engineering, from Purdue
2001 IEEE Int. Semiconductor Manufacturing Symp., pp. 3–6. University, West Lafayette, IN.
[57] T. Inoue, “A comparison of AMHS delivery models in the fab of He is an Associate Professor of Engineering
the future,” Fut. Fab Int. vol. 13, pp. 185–187, Aug. 2002 [Online]. in the Department of Industrial Engineering at
Available: http://www.future-fab.com/documents.asp?d_ID=1319#,
Arizona State University. His primary area of
[Online]. Available:
[58] J. T. Lin, F.-K. Wang, and C.-K. Wu, “Simulation analysis of the research is simulation methodology with a focus
connecting transport AMHS in a wafer fab,” IEEE Trans. Semicond. on model abstraction, execution speed and output
Manuf., vol. 16, no. 3, pp. 555–564, Aug. 2003. analysis. He has just completed work on a grant
[59] T. Jefferson, M. Rangaswami, and G. Stoner, “Simulation in the design from SRC/International SEMATECH as part of the
of ground-based intrabay automation systems,” in Proc. 1996 Winter FORCE II program related to creating a simulation
Simulation Conf., pp. 1008–1013. on demand environment.
[60] G. T. Mackulak, F. P. Lawrence, and J. Rayter, “Simulation analysis Dr. Mackulak is has received several Engineering Teaching Excellence Award
of 300 mm intrabay automation vehicle capacity alternatives,” in Proc. nominations and is a Senior Member of the Institute of Industrial Engineering.
1998 IEEE/SEMI Adv. Semiconduct. Manuf. Conf., pp. 445–450.
[61] R. Sturm, F. Frauenhoffer, J. Dorner, and T. Kaufmann, “Implementing
manufacturing strategies for advanced capacity analysis and material
handling design in the planning process of a wafer fab,” in Proc. Int.
Conf. Semiconductor Manufacturing Oper. Modeling and Simulation, John W. Fowler is a Professor of Industrial Engineering at Arizona State
1999, pp. 92–97. University (ASU) and was previously the Center Director for the Factory Op-
[62] S. Murray, G. T. Mackulak, J. W. Fowler, and T. Colvin, “A simula- erations Research Center that was jointly funded by International SEMATECH
tion-based cost modeling methodology for evaluation of interbay ma- and the Semiconductor Research Corporation. His research interests include
terial handling in a semiconductor wafer fab,” in Proc. 2000 Winter modeling, analysis, and control of semiconductor manufacturing systems.
Simulation Conf., pp. 1510–1517.
Dr. Fowler is a member of IIE, INFORMS, and SCS. He is an Area Editor for
[63] M. Schulz, T. Stanley, B. Renelt, R. Sturm, and O. Schwertschlager,
“Simulation based decision support for future 300 mm automated mate- SIMULATION: Transactions of the Society for Modeling and Simulation Inter-
rial handling,” in Proc. 2000 Winter Simulation Conf., pp. 1518–1522. national and an Associate Editor of IEEE TRANSACTIONS ON SEMICONDUCTOR
[64] S. Shikalgar, D. Fronckowiak, and E. MacNair, “300 mm wafer fabri- MANUFACTURING.
cation line simulation model,” in Proc. 2002 Winter Simulation Conf., Dr. Fowler is an IIE Fellow and is on the Winter Simulation Conference Board
pp. 1365–1368. of Directors.

View publication stats

You might also like