Proceedings of the 1999 Winter Simulation Conference

P. A. Farrington, H. B. Nembhard, D. T. Sturrock, and G. W. Evans, eds.

PROCESS AND MATERIAL HANDLING MODELS INTEGRATION

Michael Norman Jerry Barksdale Philip Campbell

Deron Tinsley Otto Wiersholm Edward MacNair

AutoSimulations, Inc. Dominion Semiconductor IBM Microelectronics Div.

655 Medical Drive 9600 Godwin Drive Mail Stop – 56A
Bountiful, UT 84010, U.S.A. Manassas, VA 20110, U.S.A. 1580 Hopewell Jct., NY 12533,
U.S.A.

ABSTRACT handling system, developed in the AutoMod software,

with a manufacturing model developed in the AutoSched
AutoSimulations has developed and applied a new method AP software, using a new product call the Model
of integrating separate models of manufacturing process Communications Module (MCM). Material handling
and material handling systems that exploits the strengths of moves requested via messages by the AP model are
two different products to provide users with maximum executed in the AutoMod model. Other inter-model
productivity and flexibility. As applied to semiconductor messaging coordinates processes in each model dependent
wafer fabrication facilities in the examples presented here, on equipment availability or other material handling
this approach replaces the traditional method of building a system conditions. Each model may also be run
single large, complex model of the entire scope of independently to facilitate specific uses by both groups.
operations. Two case studies illustrate the technique and
demonstrate the benefits of this new modeling architecture. 1 INTRODUCTION
IBM recently contracted with AutoSimulations to
upgrade an existing model of their prototype 300mm wafer The current “standard practice” for modeling wafer
fabrication facility. The existing model was developed in fabrication operations as a whole is to develop a single
1997 using the AutoSched application, a long-time model that encompasses the entire scope of fab operations.
manufacturing system modeling product from This effort and scope is necessary however to capture the
AutoSimulations. It incorporated various material constraining interactions of material handling and storage
handling systems (different configurations of overhead systems, process equipment, human operators and product
monorail layouts and use of operators) and manufacturing flow throughout wafer fabrication. Due to the size these
processes within one model – the standard practice for facilities and the enormous complexity of the wafer
modeling wafer fabrication facilities with the intent of fabrication process, these models are typically very large
understanding material handling requirements and process and sophisticated. Months are often spent developing and
requirements for varying order demand. validating these models. The models are applied to a
In 1998, Dominion Semiconductor, a subsidiary of variety of purposes, including facility layout planning, new
IBM and Toshiba, contracted with AutoSimulations to equipment justification, staffing analysis, order dispatch
develop a coordinated tool set to meet a diverse set of and control management, and many other planning support
requirements to support their fabrication of 200mm wafers. functions. Their use has become integral to many business
The manufacturing support group needed assistance in planning processes and in the daily operation of many
planning process capacity, in planning future staffing wafer fabs, to the point of daily scheduling using “live”
requirements, and in analysis of cost reduction initiatives input data in some applications.
and production scheduling, including lot start and dispatch Various simulation software vendors have developed
policies. The automated material handling system “templates” or entire products addressing the modeling
(AMHS) support group wanted a dynamic and graphical challenges associated with semiconductor wafer fabs.
means of justifying new equipment and analyzing potential AutoSimulations’ flagship product for many years in this
system modifications, as well as a tool to assist with other market was AutoSched. AutoSched provided a custom
planning and daily system support activities. user interface, an existing customizable template model,
AutoSimulations’ solution for both projects was to and a dispatch rule editor. The product was built “on top
integrate a detailed model of the automated material of” AutoMod, a general-purpose simulation product

1262
 Norman, Tinsley, Barksdale, Wiersholm, Campbell, and MacNair

providing built-in material handling constructs, a full- • Material handling systems modeled with a full-
featured programming language and 3-D graphics. This featured simulation language containing built-in
architecture provided for data-driven models using material handling modules and 3-D animation
AutoSched’s input file structures and 3-D animation. One • Independent model development, validation,
drawback to this broad functionality was execution speed. experimentation and support
Addressing this issue, AutoSimulations’ next-generation • Distributed processing, taking advantage of
product, AutoSched AP, was optimized for runtime multiple networked computers or processors
performance. The new object-oriented, C++ environment for improved runtime performance
of AutoSched AP sacrificed the computationally intensive
graphics for exceptionally fast execution speed. Many of 2 PROCESS OVERVIEW
AutoSimulations’ clients were willing to live without
graphics for the drastically improved model run times and Wafer fabrication is an extremely expensive business.
other new functionality of AutoSched AP. Typical fabs today cost several billion dollars to build and
Recently (with the release of AutoMod version 8.7 in equip, due primarily to the extreme “cleanroom”
1998) AutoSimulations has developed a product that environment needed to prevent contamination of the wafers
provides for communication between different simulation during fabrication, the tight physical tolerances of the
models or between models and other applications. This imbedded circuitry of the wafers, and the exotic production
technology is based on “sockets”. Sockets communication materials used. Many elaborate procedures and measures
is provided with most computer operating systems, and the are taken to keep the wafers from being exposed to any dust
TCP/IP standard allows for sockets on different hardware or other foreign matter and all of the process and handling
platforms to communicate seamlessly. Most computer equipment is designed specially for cleanroom use. The
systems have support for communications through TCP/IP equipment and fabs themselves also have a limited useful
Sockets. In Windows NT sockets are called Windows life due to rapidly changing technology and market forces.
Sockets, or WinSock. To be competitive in this environment, semiconductor
Socket communications allow multiple models to be companies need to closely monitor and control production
attached to each other via messages. This attachment can processes and costs. Relatively small improvements, or
be between models on the same computer or with models problems, usually result in big cost impacts.
on multiple networked computers. The messages may The processes involved in wafer fabrication provide
contain text information or data; the message formats are many challenges. For example:
user-defined. The use of multiple models communicating
saves time in model building, because different models can • Individual product routings are diverse and
be developed concurrently by different simulation analysts, highly variable with frequent re-entrant flows
and instead of merging models or code, the models can (meaning the same sets of equipment are
work independently and communicate with each other revisited many times)
where needed. This feature enables users to take • Changing priorities (the “what to work on
advantage of using multi-processor hardware and networks next” decision) constantly require
of machines. adjustments to schedules
Considering the many opportunities now available • Time-critical interdependencies may interrupt
given the ability for models to communicate with one planned sequences or lead to rework if time
another, the obvious new architecture for models of wafer constraints are violated
fabs is to exploit the strengths of different products and • Different batching conditions at various
link them via messaging. For the case studies described in process steps require complex dispatching
this paper, the manufacturing systems (consisting of part decisions which may impact later steps (i.e.
routings, equipment definitions, dispatch algorithms, create starvation or bottlenecks)
operations schedules, etc.) are modeled with the
• High quality standards necessitate frequent
AutoSched AP software. The material handling systems inspections and require periodic tool and
(automated overhead electrified monorails, operators, operator requalifications
storage units, operators, etc.) are modeled with the
• Engineering tests and stringent handling
AutoMod software. The two models communicate via
requirements interrupt or delay equipment use
sockets with the Model Communications Module
• Varying customer demand and new product
application. This design provides many benefits:
introductions result in constantly changing
product mixes.
• Wafer fab process modeling in an
environment specifically designed for that
All of these factors, along with normal process
application, data-driven and extremely fast
variations, create a very dynamic scheduling environment

1263
 Process and Material Handling Models Integration

and a daunting environment to analyze or attempt to ongoing process design issue at Dominion and is an
control. The need for automation (not to mention important area of investigation.
simulation!), both in the physical processing of product and The PRI system is controlled by Dominion’s Material
in the control and management of the system is obvious. Control System (MCS), which communicates directly with
The high level of automation and the many systems the Manufacturing Execution System (MES). Together
supporting production lead to the production of vast these systems track the location and status of each lot and
amounts of data. The need exists to better organize, make OEM vehicle in the fab. The AMHS is supported around-
sense of, and present important information from this data the-clock (24x7) by on-site PRI engineers and a small group
in a systematic and timely manner for process monitoring, of Dominion engineers. The Dominion staff is responsible
control and improvements. for recommending efficiency improvements, for new
equipment justification and for ensuring the system is
3 MATERIAL HANDLING functioning properly to support production. A staff member
is always on-call to address any problem that arises.
Sets of wafers (referred to as “lots”) are transported in sealed The IBM facility is the same basic “spine-type” layout
containers called pods. These pods, each containing up to as the Dominion fab. The exact type of material handling
25 individual wafers, are stored within “stockers” inside system was yet to be determined at the time of modeling,
each bay of the clean room. Stockers are automated storage so rather than a specific vendor solution, various
units with an internal robot moving lots between I/O ports alternative material handling systems were modeled for
and shelves; the typical capacity of a stocker is about 200 comparative analysis. One system resembled the
pods. The pods are transported between the bays (isolated Dominion AMHS, another consisted of isolated interbay
rooms connected to a central aisle) via an automated and intrabay monorail systems, another serviced the entire
overhead electrified monorail system. The Dominion fab fab with an interconnected AMHS, and yet another
uses an AMHS provided by Precision Robotics Incorporated consisted of a manual system. Each of the material
(PRI), which uses cars on the rail system to transport the handling schemes could be selected by flag settings in an
pods. The movement of pods between bays is referred to as input data file.
“interbay” transport. Figure 1 below shows the track layout
of the PRI system. Each small loop extending from the 4 MODEL ARCHITECTURE
central aisle (horizontal tracks in the diagram) services one
or more stockers inside a bay. The system architecture provides the flexibility of separate
The PRI system at Dominion currently services 11 use of the models and for tightly integrated operation.
bays on about 3,000 ft. of track, servicing 20 stockers (the Running both models together (referred to as “integrated”
AMHS simulation model included a planned expansion, mode) is necessary any time product routings, demand or
which resulted in 39 stockers in 22 bays). Routing of cars mixes change. Both models’ simulation clocks are
takes place at redirection devices called turntables, which synchronized, with the models communicating via
as the name implies, may rotate and proceed cars to messaging technology provided by the MCM. This mode
different sections of track; there are 23 turntables in the is activated from the AutoSched AP environment, with that
system. Complex routing and parking migration model acting as a “server” to the AutoMod “client”.
algorithms determine the actual paths taken by the cars. Upon model start in integrated mode, a single data file
is read by both models. This file contains definition
information regarding the stockers and local buffer storages.
The use of one input file for both models simplifies the
change management of the storage devices, the only
equipment common to both models (an operator input file
may be added to the IBM model to handle dynamic sharing
of operators for material handling and process tasks).
While running in integrated mode, the AutoSched AP
Figure 1: Dominion AMHS Plan View model messages to the AMHS model when material
handling moves are needed, then waits for the AMHS
For material handling moves inside each bay (so- model to execute the move and return a message that the
called “intrabay” moves), human operators stock in and out move was completed. A unique message type for each of
pods from I/O ports at the stockers. At Dominion, the pods these messages (signified by a particular type of message
may be temporarily stored on racks inside the bay or used defined for the MCM) contains:
immediately on one of the machines in the bay. The racks
in each bay act as local storage buffers to individual or sets • the string name of the lot
of machines. The use and sizing of these local buffers is an • a lot pointer value (an integer)

1264
 Norman, Tinsley, Barksdale, Wiersholm, Campbell, and MacNair

• the string name of the local buffer the lot is

moving from (if it applies)
• a pointer for that buffer
• the string name of the stocker the lot is
moving from (if it applies)
• a pointer value, the string name of the stocker
the lot is moving to (if it applies)
• a pointer value for that stocker
• the string name of the local buffer the lot is
being sent to (if it applies)
• a pointer value for that buffer
• the priority of the lot
• a unique message number (for validation)
• an integer move type (1 – 4):
Type 1 - a stocker to stocker move Figure 2: Integrated Models Schematic
Type 2 - a stocker to local buffer move
Type 3 - a local buffer to stocker move The from/to move rate and duration tables generated
Type 4 - between two local tool buffers during integrated use of the models provides the capability
to later drive each model separately (see Figure 3). If
When a move message is received by the AMHS operated independently, the AMHS model may use the
model, a lot load is created, attributes are assigned as read from/to rate data to generate moves independently. These
from the message, and a material handling move is rates may also be factored (using some rate multiplier for
initiated. At times, the destination directed by the instance) to induce more or less overall load on the AMHS.
manufacturing model may not be available (i.e. the device Other, separate experiments may be performed with the
may be down or in a maintenance mode), in which case AMHS without the need to run both models together.
algorithms in the AMHS model choose an alternate
delivery destination. Once the move is completed, the
AMHS model sends a new message to the manufacturing
model. This message returns most of the data originally
sent (for validation by the AutoSched AP model and
retrieval of the proper delayed lot) along with a flag
indicating whether or not the lot arrived in the destination
directed by the manufacturing model. If the lot did not
make it to the “primary” destination, a list is updated for
later “migration” moves (explained below).
A third type of message is defined for the AMHS
model to signal the AutoSched AP model any time a
storage device returns to service after being down or now
has available capacity after having none. This message Figure 3: Independent Models Schematic
simply passes the string name of the device back. The
manufacturing model recognizes this type of message as a Independent running of the manufacturing model
signal that it may be able to request “migration moves” provides the best runtime performance and is used for long
from alternate storage devices to their “primary” time-horizon experiments (cycle time studies, for
destinations. A list the model has kept of lots not reaching example). Average delay times for all from/to
their primary destination is searched for any that may need combinations generated in integrated mode are used to
to move to the newly available storage device and new model typical delays encountered in material handling.
move messages to that effect are issued to the AMHS The use of the delay times can also be defeated entirely to
model. compare simulated overall cycle times to ideal total
Figure 2 shows the relationship of the models and their process times. Updates to the rate and delay files are made
data files. Although the messages are depicted as any time a routing, product mix, or equipment change is
travelling via a network connection, the models may in fact made to either model.
reside on the same computer. Two output files are In addition to the flexibility of use, this architecture
generated by the AMHS model during “integrated mode” provides the capability of supporting each model
operation. They are both flat ASCII tables of “from/to” separately. The AMHS support engineers at Dominion use
average move rates and durations. and maintain the AutoMod model independently of the

1265
 Process and Material Handling Models Integration

Process Support Group; the process equipment and 6 POTENTIAL INTEGRATION

automation support engineers at IBM are at different sites.
Independent use also provides better skill mapping to each Upon completion and validation of the manufacturing and
application. The modular design is highly extensible, an AMHS models, Dominion management decided to move
important factor in the planned on-line integration of the forward with further integration and development of the
models (see Figure 4). models with other AutoSched Productivity Family (APF)
products and their business systems. Temporal (time-
7HPSRUDO 0(6
based) archival and reporting capability will be added with
5HSRVLWRU\ the ISS Reporter application. This technology provides a
powerful means of capturing MES transactions in a
temporal database and creating ad-hoc custom reports and
7HPSRUDO 0DQXIDFWXULQJ 5HDO0WLPH business graphs of fab activity. Because the repository is
5HSRUWHU 0RGHO 'LVSDWFKHU temporal, reports can be developed that cannot be created
using traditional databases, which simply store “snapshots”
of shop floor status, not a complete history. The tool may
$0+6 be used to “drill-down” to the specific cause of problems to
#0RGHO determine the root cause of reported anomalies.
AutoSimulations’ Real-Time Dispatcher (RTD) 
Figure 4: Potential Systems Integration product is in the process of being integrated with
Dominion’s MES (refer to Figure 4). Once complete, fab
5 CURRENT STATUS dispatching rules may be developed and tested using the
manufacturing model, then safely implement into the MES
The AMHS engineers at Dominion are currently using the scheduling system for real-time dispatching based on the
system to justify new stocker requirements and determine current status of the fab. The RTD application will be
the optimum number of cars under various operating seamlessly integrated within the MES so that operators will
scenarios. They are also analyzing the cause of vehicle continue to see the same dispatch screens as today.
gridlocks and other control system issues. In the future, Dominion may chose to integrate the
Process support IEs at Dominion are currently using AMHS control system (MCS) into the architecture. This
the models to identify and analyze capacity constraints in addition could further refine the dispatching and control of
the fab. The effectiveness of adding local buffering (racks vehicles in real-time based on current conditions. Plans for
for staging pods inside the bays) is also being evaluated. this enhancement have been postponed until the APF
New or modified dispatch policies are being tested in components are functional.
attempt to reduce WIP and reduce cycle times. At the time of publication of this paper, IBM
Dominion cost accountants are anxious to apply management was considering a proposal by
costs to modeled activities to better evaluate proposed AutoSimulations to further integrate the manufacturing and
process changes and cost reduction initiatives. They see AMHS models with the APF tools.
this tool as a new and better way to determine whether
or not local improvements have a significant effect on 7 SUMMARY
“the bottom line” – in other words, whether spending
money on improvements in particular areas or processes The integrated model solution provided by
returns the investment from an entire process-level AutoSimulations is an effective way to model separate, but
perspective. inter-dependent business functions. Applying the “best-fit”
IBM engineers are using their AMHS model to tool to its particular strengths, and linking the tools with
evaluate the efficiency and tradeoffs associated with messaging technology, increases flexibility and
different material handling system designs. Many new productivity throughout the project life cycle. Use of a
handling issues are becoming significant concerns with the common input data set for both models facilitates
design of 300mm wafer facilities, such as the higher costs consistent application of input information and provides for
of wafer damage during handling and ergonomic future live integration of MES data.
constraints imposed by the larger and heavier pods. Independent use of each model provides the capability
Process planning and facility engineers at IBM are for detailed analysis and experimentation with the AMHS
currently using the models to determine tool set model, and extremely fast execution of the manufacturing
requirements for various production rates and product model for scheduling and capacity analysis. The individual
mixes. Different layout patterns are being compared for models may reside in different physical locations,
space requirements, and overall effect on throughputs and connected over the company’s network, an important
product cycle times.

1266
 Norman, Tinsley, Barksdale, Wiersholm, Campbell, and MacNair

aspect since personnel involved with these projects come engineering positions in process equipment engineering,
from different organizations in different physical locations. automation systems development, and mechanical systems
The model architecture has proven effective and design. Philip holds a M.S. degree in Mechanical
versatile; it could be implemented in a wide range of Engineering from Rensselaer Polytechnic Institute, a B.S.
similar applications. in Mechanical Engineering from Polytechnic University,
and a BS in physics from St. John’s University.
REFERENCES
EDWARD MACNAIR, specializes in modeling
Barksdale, J., Norman, M., Tinsley, D., Wiersholm, O., contention systems in IBM Research’s Mathematical
1999. Integration of Manufacturing & Material Sciences Department. He has more than 30 years of
handling Simulation Models. IIE Simulation Solutions experience at IBM and has worked in the areas of
’99 Proceedings. modeling systems and model tool development. Mr.
Rohrer, M., 1998. Vision of Emulation, AutoSimulations MacNair holds a M.S. in Operations Research from New
Symposium ’98 Proceedings. York University and a B.A. in Mathematics from Hofstra
Stafford, R., 1998. AutoMod/ASAP Integration, Auto- University.
Simulations Symposium ’98 Proceedings.

AUTHOR BIOGRAPHIES

MICHAEL NORMAN is a Senior Simulation Analyst for

AutoSimulations and has worked in simulation for the past
10 years. Michael holds a B.S. in Industrial Engineering
from the University of Washington, is a Professional
Engineer and a Senior Member of IIE.

DERON TINSLEY is a Senior Simulation Analyst for

AutoSimulations. He provides AutoSched AP and
AutoMod consulting services to AutoSimulations’ clients
in many areas of manufacturing and material handling.
Deron holds a M.S. degree in Mechanical Engineering
from the University of Utah.

JERRY BARKSDALE, Mechanical Engineer for the

Automated Material Handling System at Dominion
Semiconductor, joined the company in 1996. Prior to
working at Dominion, Jerry was a consultant at Booz,
Allen, and Hamilton for 7 years. He holds a B.S. in
Mechanical Engineering and Applied Mathematics from
The George Washington University in Washington, D.C.

OTTO WIERSHOLM, Sr. Industrial Engineer in the

Industrial Engineering and Planning Department at
Dominion Semiconductor, joined the company in 1997.
Otto worked for Zilog Semiconductor in 1996 and 1997,
and at the Idaho National Engineering Laboratory
from1989 to 1996 prior to coming to Dominion. He holds
a M.S. degree in Industrial Engineering from Montana
State University.

PHILIP CAMPBELL is currently an Advisory Engineer

with International Business Machines Corp.,
Microelectronics Division in East Fishkill, NY. His
present assignment is with the Factory Technology
Integration group analyzing new 300mm fab capacity.
Over the past sixteen years, he has held various

1267