Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Journal of Systems Integration, 10, 189±206, 2001

# 2001 Kluwer Academic Publishers. Manufactured in The Netherlands.

Supply Chain Model for the Semiconductor


Industry of Global Market
YOUNG HOON LEE youngh@yonsei.ac.kr
Yonsei University, Department of Information and Industrial Systems Engineering, 134 Shinchon-Dong,
Seodaemun-Ku, Seoul 120±749 Korea

Received July 15, 2000; Accepted February 1, 2001

Abstract. Supply chain management (SCM) is a new approach to satisfy customers of products and services via
an integrated management for the whole business processes of the manufacturing from the raw material
procurement to the product or service delivery to customers. With the rapid development of the information
technology and global market environment the company is required to manage whole supply chain network to
gain the competitiveness in business. Typically semiconductor industry is the one whose supply chain network is
distributed all over the world, and its manufacturing process has the particular characteristics which has to be
considered in the modeling of supply chain. In this paper we suggest the push and pull type supply chain models
and their mathematical formulation for the semiconductor industry of the global market. Die bank between the
front-end and back-end in the semiconductor manufacturing process is assumed to be the decoupling point by
which two models are divided. Push supply chain model is based on the high throughput and the balance of the
WIP ¯ow, which takes the characteristics of re-entrant mechanism into account. Pull supply chain model is
constructed considering order-based manufacturing, distribution and transportation process in order to meet
customer's request appropriately. Development issues of the system and its implementation are also discussed.

Keywords: supply chain management, semiconductor industry, push and pull model

1. Introduction

Supply chain management (SCM) is a new approach to satisfy customers of products and
services via an integrated management for the whole business processes of the
manufacturing from the raw material procurement to the product or service delivery to
customers. The upstream and downstream of manufacturing, e.g., the procurement of raw
material, distribution and transportation process of ®nal goods, has to be managed
effectively. In most industries, market environments are getting more globalized, and
whole business processes are being implemented on the complex network of supply
chain. Therefore consistent and integrated management is needed for the whole supply
chain. Supply chain network can be grouped into three stages: supplier stage, plant stage,
and distribution stage [7]. The supplier stage consists of all upstream suppliers who
provide inputs to the manufacturing or service transformation. Shipments from suppliers
can be made directly to the plants or to intermediate stocking points, that is, the supply
distribution center. The plant stage may consist of several steps: feeder plant supplying
assembly plants, intermediate inventory bank, and ®nal assembly, test and packaging.
The distribution stage consists of warehouse, distribution center, retailer and ®nal
customers. Many kinds of decisions have to be made on the supply chain networks:
190 LEE

con®guration of the stage-to-stage network, amount of inventory at each network node,


volume of transportation, number of network nodes such as suppliers, supply distribution
centers, plants, warehouse and distribution center, etc. Each network node or stage can be
operated by local decisions only considering the node itself or within the stage.
Globalization and severe competition on industries ask the competitiveness over the
whole supply chain network, but it is not easy to achieve the global optimization. Total
quality management (TQM) made it possible to deliver the ¯awless products to the
customers, and ¯exible manufacturing systems (FMS) achieved the quick response and
agile manufacturing at low cost. Effective supply chain management mechanisms are
expected to deliver products quickly with low inventories through the ef®cient routes,
which lead to customer satisfaction and hence new competitiveness in the global market.
Semiconductor manufacturing is one of the industries that require most advanced
technology. Figure 1 shows the overview of the semiconductor manufacturing process.
Wafer fabrication and probe are generally referred to as front-end operations, and the
following stages, assembly and the ®nal test, are referred to as the back-end. Wafer
fabrication process consists of 15±30 masking layers, each of which has 20±40 process
steps, which are called as recipes. Speci®c operations may vary depending on the product
and the technology in use. In the wafer probe the individual circuits, of which there may
be hundreds on each wafer, are tested electrically by means of thin probes [27], and
circuits that fail to meet speci®cations are marked with an ink dot. The wafers are then
cut into individual circuits, which are also called as die, and the defective circuits are
discarded. In the back-end operations, lots may vary in size from several individual dies
to several thousands depending on the product and on customer speci®cation. In assembly
the dies are placed with lead-frame in plastic or ceramic packages that protect them from
the environment. Final testing process ensures that customers receive a defect free
product by using automated testing equipment to interrogate each integrated circuit and
determine whether it is operating at the required speci®cations.

Figure 1. Semiconductor manufacturing process.


SUPPLY CHAIN MODEL 191

Final goods are distributed to another manufacturing site to be made into modules or to
be a part of the set such as PC or telecommunication media. They may be stored in
warehouses temporarily or transported to the distribution centers for delivery
convenience or cost saving. Supply chain network of the semiconductor industry is
typically spread out all over the world. For example, fabrication and probe manufacturing
may be processed in Eechon, Korea, and the assembly and ®nal test may be processed on
Simyang in China. Malaysian PC maker receives the packaged devices from China to
assemble desktop PC that are exported to the United States. The primary management
goal has been set to optimize certain performance measurement within the site or the
stage. Maximizing the throughput and the utilization of the bottleneck machines is
important in the wafer fabrication. The back-end line has been measured on due date
performance and the customer satisfaction. Since throughput and due date performance
are con¯icting each other, the result was to hold large inventory buffers between the
front-end and the back-end. The die bank is the necessary inventory point to absorb the
¯uctuations of the production and the demand. The problem is that it was not managed by
the global point of view in the supply chain network. The market of the semiconductor
industry becomes more buyer-oriented rather than seller-oriented, and gets more pressure
to cut down on cycle times and achieve high levels of customer satisfaction.
The objective of this paper is to suggest the model and the framework of the supply
chain network for the semiconductor industry, in which the characteristics of its industry
are considered. It is organized as follows: Literature survey and related works are in
Section 2, supply chain modeling and formulations are in Section 3, considerations for
the development of the system and its implementations in Section 4. In Section 5 are
conclusions and discussion.

2. Literature Survey and Related Works

Most researches for the semiconductor manufacturing are emphasized on the production
planning and scheduling, especially on the fabrication process that is the most
complicated one. Flow control, which is the combination of the lot release, dispatching
and batching strategies to control the ¯ow of lots, is one of areas on which lots of studies
are done and their results are implemented. The objectives of their study are achieving
high throughput, high utilization of the bottleneck machines, and keeping the work-in-
process (WIP) as balanced as possible to avoid the starving of the machines. Traditional
scheduling and sequencing approach has the limit in the study of the fabrication line since
it is too complicated to analyze them in detail. Therefore dynamic ¯ow control methods
such as dispatching rules and lot release rules have been received more attention, and
regarded as more practical by researchers and practitioners, while traditional scheduling
and sequencing methods are too static for daily use, and are not ¯exible to meet the
dynamic change. Most dispatching rules used in the research are not new, e.g., ®rst in ®rst
out (FIFO), earliest due date (EDD), shortest remaining processing time (SRPT), least
slack (LS), and so on, while lot release control rules suggested in 1988 are considered to
be seminal in terms of rousing the interest of researchers. One of these is the starvation
avoidance rule suggested by Glassey and Resende [9, 10] in which a virtual inventory
192 LEE

measured over a lead time is used to regulate order release on a single bottleneck work
center. More speci®cally a lot is released when total expected time to consume WIP at the
bottleneck workstation is smaller than the lead time taken to arrive for a lot released, in
other words, only when starvation of the bottleneck is expected. Wein [29] introduced the
workload regulating input method, in which a lot is released so that the expected
workload for the bottleneck maintain the target load. These two rules were tested with
several dispatching rules on the bottlenecks and they concluded that order release is more
important than dispatching. Lou and Kager [20] suggested the ¯ow rate control rule for
the order release strategy by considering the re-entrant ¯ow process into a strictly serial
production line using a virtual workstation concept. Although these papers were the
landmarks of the systematic approaches for the fabrication control, and gave the good
conceptual points, they have lots of ¯aws to be implemented in the real fabrication lines.
The model on which their suggestions are tested is not suf®cient to represent the
characteristics of the real fabrication line. The complicated combination of the order
release rules and the dispatching rules made it dif®cult to evaluate the performance of
speci®c rules. Furthermore, as due date performance and cycle time reduction are more
important under the global supply chain environment, keeping the low WIP and the
balance of wafer ¯ow are their primary concerns.
CONWIP methods by Spearman et al. [25] appears to be a simple but robust
control philosophy which can manage the WIP and the throughput simultaneously.
Cycle time reduction was studied by Ehteshami et al. [6], Wein [30], and Demeester
and Tang [5]. They suggested how to de®ne and compute the proper WIP for the layer
of the semiconductor processes that are used in computation of the balance of the
layer, and how to schedule the bottleneck machines using dispatching rules with the
information of the balance. Case studies using the system development for the
semiconductor industry were reported by Bitran and Tirupati [3], Glassey [11], and
Kraft [15]. Intensive survey for the study in semiconductor can be found in Uzsoy et
al. [27, 28], and Leachman [16]. Although much advance in the research for the
semiconductor manufacturing is achieved as listed as above, their aspects are
restricted on the speci®c area in the semiconductor supply chain rather than aggregate
or globally viewed over the whole network. Ovacik and Weng [23] points out the
con¯icting objectives in the semiconductor supply chain management, and designed
the framework of supply chain for the semiconductor industry, especially three
processes: distribution planning, manufacturing planning and the order promising. Lee
and Billington [17] and Lee et al. [18] are good references, although they are not for
the semiconductor industry, which describe a supply chain model consisting of several
manufacturing facilities and distribution centers developed for the Hewlett and
Packard to minimize the inventory cost in the supply chain and to achieve the high
customer satisfaction. Mathematical programming is the fundamental and the ®rst
approach to formulate the supply chain model to minimize the production, inventory,
transportation and more as shown in Shapiro et al. [24] and Arntzen et al. [2] for the
Digital Equipment Corporation.
Case study for the semiconductor industry was reported by Murty and Bienvenu [22]
for Harris Semiconductor, in which academic pursue was implemented with information
technology to have excellent performance in supply chain management. Harris
SUPPLY CHAIN MODEL 193

Semiconductor, in 1990, after the acquisition of RCA and GE facilities to expand, found
that on-time-delivery performance was bad with only 70%. Harris needed an integrated
system to manage its entire supply chain, and adopted the BPS (Berkeley Planning
System), the academic version of which was renamed IMPReSS, to develop a company-
wide information network and a semiconductor-speci®c planning engine. They reformed
the engine so that its planning engine, IPE, had to orchestrate the factories for optimum
throughput, manufacturing supply for order quotations, procurement to work out
economic purchasing deals. IPE was designed to maximize the revenue, utilization of the
bottlenecks, and customer satisfaction by allocating the order-based and forecasted
products properly on the manufacturing processes through MRP and linear programming
(LP). After several upgrades of the system and endeavor to improve the data integrity,
they achieved 94% on-time-delivery performance in 1994. Harris Semiconductor case
has to be noti®ed for the integration of the front-end and back-end of the semiconductor
manufacturing in planning and order promising. Bouff [4] de®ned three basic
performance goals to resolve in the semiconductor industry as the performance to
customer request, responsiveness, and performance to promise. In his paper, the die bank
inventory is assumed to be a decoupling or ``push-pull'' point at which execution is
effectively decoupled from actual demand and is driven instead by forecast. He explained
in detail the basic processes occurring to implement the ef®cient supply chain, but did not
suggest the exact mathematical model with which the proper solution or guideline for the
decision making in supply chain can be obtained.
Frederix [8] and Makatsoris et al. [21] reported on the project, ESPRIT 20544,
dealing with the design of a planning, scheduling and control system which aims at
improving semiconductor manufacturing performances by integrating the hetero-
geneous manufacturing systems. They suggested the concept of virtual enterprise,
according to their de®nition, which is a subset of the units and processes within the
supply chain network behaving like a single company through strong co-ordination and
co-operation towards mutual goals. The objective of the project is to provide
manufacturing solutions and services for the semiconductor virtual enterprise with high
performances as follows: shorter customer order lead times, improved delivery
precision, reduced inventory and stocks, improved resource utilization, quick response
to customer enquiries, and cost reduction in the logistic processes. The supply chain
model and its solution system, named X-CITTIC, have been developed for the
operation of virtual enterprise to the needs of the semiconductor industry. They include
the following components: order promise handling fast responsive multiple request,
rough planner for the fast reactive order planning, ®ne planner scheduler for the
optimized detail plan, reactive controller, and information manager. The project
involves three industrial companies on the semiconductor sector, one software
company with three academic institutions, all of which are located in European
Union. This was known to be the ®rst multi-organizational approach for the
semiconductor industry including both within the enterprise and across the whole
supply network. Their research emphasizes the design and modeling for the multiple
network units such as plants, logistic centers and storage facilities, using the
information technology. To my knowledge, only a few have been published on the
supply chain modeling focusing on the semiconductor industry.
194 LEE

3. Supply Chain Modeling for Semiconductor Industry

Semiconductor companies are running global businesses covering multiple manufac-


turing sites, warehouses and distribution centers, subcontractors and suppliers.
Manufacturing sites may consist of multiple fabrication sites, probe sites, assembly
sites, ®nal test and packaging sites over the world. Warehouses are more closely related
with manufacturing sites while distribution centers are with the customers and marketing.
Semiconductor design teams, raw wafer manufacturers and photolithography mask shops
are typical examples of suppliers in the supply chain. It is necessary for the supply chain
model for semiconductor industry to include the entire network, starting with the
suppliers of raw materials to the customers of the ®nal products. They may consist of
several companies, some of which are in competition, and virtual or real, sometimes in
vertical or horizontal relationships [23]. There are also issues regarding security, data
integrity and authority. In our discussion, the scope of the supply chain is limited to the
network consisting of semiconductor manufacturing companies, suppliers of raw
materials, distributing facilities, and their immediate customers such as retailers, the
PC manufacturers, and telecommunication media set assemblers.
Semiconductor manufacturing has a pattern of cyclic or re-entrant process, in other
words, similar sequence of steps are repeated 20±30 times, and each repetition is called as
a layer, which consists of photolithography, etching, thin ®lming, diffusion, etc. Some of
the equipment is shared in competition with processes of different layers, and different
devices. One of the characteristics different from the normal assembly manufacturing
process is binning, whereby multiple quality graded products may emerge from a single
lot of source product. Each bin de®nes a speci®c range of performance for electrical
attributes, and percentage of the output of the source product falling into each bin are
probabilistic [19]. Binning may occur at either the wafer probe or device test process.
Devices can be classi®ed into several grades depending on the binning, to be sold with
different prices. Output products are produced probabilistically, not by order, hence some
products with high grade have to be sold at the price of lower graded products to optimize
the total revenue and to meet the order, which is called substitution. These characteristics,
cycling, binning and substitution, of the manufacturing have to be included in the supply
chain model since they work critically in the wafer ¯ow and contribution of the revenue.
The primary management goal is to maximize pro®ts through maximization of the
production throughput and the utilization of bottlenecks. In order to achieve the maximal
throughput, the traditional push type management is known to be the best, in which all
machines are made to work as much time as possible with maximal supply of WIP and raw
material. Material requirement planning (MRP) system are adopted in most factories to
support and guarantee the supply of the required materials in time. In mass production
lines with relatively small number of production items, the push type management is
preferred to maintain high utilization of equipment and high production rate. Customer
satisfaction pressure that is getting tighter in the new market, leads to management of on-
time-delivery and tardiness related performances. Pull type management is known to be
proper for the manufacturing lines producing items based on orders. Production lines are
managed by order information from the end of the manufacturing process, in which items
are prioritized and allocated to process on machines by the order and due date information.
SUPPLY CHAIN MODEL 195

Figure 2. Semiconductor supply chain framework.

Semiconductor manufacturing process has both attributes, push and pull type. Common
sequence of steps is processed in the fabrication as long as they have same
photolithography masks, which is divided into many items at the assembly and
packaging steps. The front-end needs a typical push type management while back-end
process does a pull type. There exists then a point at which push type and pull type
management encounter each other, called a decoupling point or push-pull point [4]. In
semiconductor manufacturing the die bank at wafer probe is considered to be a
decoupling point at which dies are accumulated as a WIP after wafers are cut into small
chip. The cycle time to ®nal products after the die bank is 4±6 days, and the
manufacturing process is continuous without cycling, which makes it easy to expect the
output time. Supply chain from the raw material input to the die bank has to be modeled
on the different concept from the one of die bank to ®nal products, ultimately to customer
delivery. The former is said to be the push supply chain, and the latter the pull supply
chain. Figure 2 shows the framework of the supply chain model for the semiconductor
industry.

3.1. Push Supply Chain Model

In the push supply chain model, throughput and utilization are the primary management
goals while on-time-delivery and customer satisfaction in the pull model. These two
areas, front-end and back-end, divided by the borderline of the die bank, are actually
located in the different plants in most major semiconductor companies, and are managed
by different organizations, even with different culture and manufacturing environment.
Fabrication line in the front-end can be characterized by technology-intensiveness, heavy
capital investment, long cycle time due to cycling, setup time loss of the products on
196 LEE

equipment. For the process of back-end, more emphases are given on maintaining the
¯exibility for the order requirement, hence on short cycle time, fast delivery to customers,
and tardiness minimization. Shorter cycle time under the low WIP is necessary even at
the front-end to achieve customer satisfaction on the back-end, but this is con¯icting to
the throughput maximization that can be achieved by holding high WIP to minimize the
setup time loss of equipment. In general, the production target at the front-end is
predetermined by the plan or forecasting, and has not been changed for a certain period of
time, while at the back-end it may be changed dynamically depending on the customer's
order status. Dealing two parts of a single supply chain model in an integrated form may
give a global solution to the whole network, but there is a possibility that the push model
can lose the ef®ciency and the pull model can lose the ¯exibility.
The push supply chain takes in general the typical make-to-stock policy. Devices are
itemized based on the output type of binning process at the wafer probe process. The
proportion of the bin split can be found from the history data analysis. Some devices can
be grouped together if they follow the same manufacturing process. The most important
objective function is the throughput under the lowest WIP as possible. The lower the WIP,
the better performance can be delivered to the following supply chain. Since the
fabrication and probe manufacturing consists of 600±800 process steps, all these steps
cannot be controlled always in our model. We de®ne the managing point as the process
steps in which lot ¯ow on them give critical effects on subsequent steps, and hence needs
to be managed with high priority. The aligning processes on the stepper machines, which
are known to be the most expensive equipment, can be typical managing points. We have
then as many managing points as the number of layers, and more if we add some steps in
wafer probe and other steps not in photolithography. The objective of the push supply
chain model is to maximize the weighted sum of wafer ¯ow on all managing points
within the capacity of corresponding equipment, where the weight is the degree of
importance of the production items. Wafers are moving to the next layer if they are
scheduled to process on equipment of the managing points, and join in WIP of the next
layer after processing. Wafers on WIP in the same layer are processed based on FIFO, but
wafers on different layers are competing to be processed, which are determined by the
solution of the model. In our model, volumes of wafers to be processed on equipment are
determined for each device and for each time period. The objective is to maximize total
volumes allocated to be processed on equipment at managing point subject to equipment
capacity allowed and working time for the time period given. Total volumes scheduled on
equipment correspond to the throughput of the manufacturing line, and high utilization of
equipment can be achieved under maximal throughput. Notations and formulations are as
follows:

Notations

i: index for devices.


j: index for managing points on the process.
k: index for equipment on the managing points.
t: index for time periods, e.g., shift or day.
SUPPLY CHAIN MODEL 197

X i; j; k; t†: volume of device j on the managing point j to be processed on the equipment


at time t (to be determined).
W i; j; t†: weight or degree of importance of device i on the managing point j at
time t.
WIP i; j; t†: WIP amount of device i on the managing point j at time t.
STD i; j; k†: time required to process devices i on the managing point j on machine
k.
WT: total working time during one unit of period.

Formulation A:

X
Max W i; j; t†6X i; j; k; t†
i; j; k; t
X X
s.t. WIP i; j; t† ˆ WIP i; j; t 1† X i; j; k; t† ‡ X i; j 1; k; t† Vi; j; t
k k
X
X i; j; k; t†  WIP i; j; t† Vi; j; t
k
X
X i; j; k; t†6STD i; j; k†  WT Vk; t
i; j

WIP i; j; t†  0; X i; j; k; t†  0 Vi; j; k; t

The ®rst constraint shows the change of WIP after moving or processing volume at the
end of the time period. Figure 3 shows the mechanism of the WIP change as time elapses
and the scheduling decisions are made. The second is the WIP constraint that moving
amount does not exceed the WIP amount, and the last constraint is for the working
time.
Maximization of total throughput in the semiconductor has one ¯aw that it has a
tendency of allocating wafers to fast machines as possible, and the output of devices at
the last layer varies widely from period to period. The cycling characteristics in
fabrication line requires the balanced production on all layers, otherwise it may make a
heap of wafers on certain layer, and result in ultimate decrease in the production and
utilization in the future. The balanced line implies that each process has the WIP
necessary to meet the required throughput within a desirable cycle time. In an assembly
line without the re-entrant process, perfect balance means combination of the elements of
work to be done in such a manner that at each station the sum of the required time for
each step just equals the desired cycle time. The same amount of WIP then can move to
the next station continuously with keeping the utilization as high as possible. In a re-
entrant process such as the fabrication process, it can be said that the perfect balance has
been achieved when each process has the WIP proportional to its cycle time, which can
be obtained using history data they have. The total WIP amount has to be determined by
management based on the target cycle time they want to reach. The target cycle time
cannot be smaller than the sum of the minimal required processing time of each process.
198 LEE

Figure 3. Framework of WIP change.

Keeping the balance of wafer ¯ow makes it possible to provide wafers to layers properly,
and hence it rarely happen to be piling up of wafers on certain layers. Since steps of the
layer can be processed on the given set of equipment, unbalance may incur the starvation
of WIP at certain layer. If equipment workload is appropriately distributed over all layers
with balanced WIP, high level of equipment utilization can be achieved without
starvation. The target WIP, which is the proper WIP level in terms of the production
target and cycle time, can be determined in many ways, one of which is WIP movement
variance method [14]. The objective in the second formulation considering the WIP
balance, is to maintain WIP as desired as possible assuming that wafer movement of all
layers is at least the production target. The objective function is to minimize the sum of
the differences between the target WIP level and the resultant WIP level after the
scheduling. Constraints are equal to those of the formulation A except the last one, which
makes wafers of each layer processed at least the amount of the production target. This
model can be infeasible in case that the current WIP level is less than the production
target, but slight modi®cation, such as taking the minimal value of production target and
the current WIP level, can give a feasible solution.

Notations

TW i; j; t†: target WIP amount of device i on the managing point j at time t.


Q i; t†: production target of device i at time t.
SUPPLY CHAIN MODEL 199

Formulation B

X
Min A i; j; t† ‡ B i; j; t†
i; j; t

s.t. A i; j; t† B i; j; t† ˆ WIP i; j; t† TW i; j; t† Vi; j; t


X X
WIP i; j; t† ˆ WIP i; j; t 1† X i; j; k; t† ‡ X i; j 1; k; j† Vi; j; t
k k
X
X i; j; k; t†  WIP i; j; t† Vi; j; t
k
X
X i; j; k; t†6STD i; j; k†  WT Vk; t
i; j
X
X i; j; k; t†  Q i; t† Vi; j; t
k
A i; j; t†  0; B i; j; t†  0; WIP i; j; t†  0; X i; j; k; t†  0 Vi; j; k; t

3.2. Pull Supply Chain Model

Primary goal for the pull type supply chain from the die bank to customers is to meet the
orders with minimal cost incurred. The cycle time from the die bank to the ®nal products
is in general 4±6 days and another similar time period is required for the delivery to
customers. Although customers who need the ®nal products of semiconductor are from
all over the world, the transportation cost are not different signi®cantly from each other
since they are transported usually by aircraft. In our model we assume that the cost on
the delivery and transportation does not give critical effects on decision making,
although not negligible. If all orders are promised within the production capacity, they
then can be met within the promise of delivery. Decisions are made on the production
volume at the ®nal manufacturing point for each production item, customers and time
period. Manufacturing capacity for the managing points on the back-end may restrict the
production volume on decision points. WIP volume on the back-end has its own
destination of customers and delivery time. Final goods can be stored at warehouses, and
transported to the distribution centers to wait for ultimate customer delivery. At
distribution centers devices can be classi®ed and packaged based on orders. Size of the
storage and inventory policy may affect the production volume at the ®nal manufacturing
point to be determined. For simplicity we assume that time taken to transport devices
from the manufacturing plant to warehouse or distribution center depends on only their
facility, not by devices and time, which is true for semiconductor. The warehouse and
distribution center where devices are stored, are usually determined according to their
customers, since logistics center for semiconductor industry are operated not for the
storage of long time, but for the purpose of marketing. The objective function for the pull
200 LEE

supply chain is to maximize the revenue achieved through the production and delivery,
subject to equipment capacity of manufacturing, storage capacity of stocking point, and
delivery time taken between the logistic centers. Notations and formulation are as
follows:

Notations

i: index for devices of ®nal products.


j: index for managing points such as bottleneck process, warehouse, distribution centers.
b: index for bin splits.
c: index for customers.
t: index for time periods, e.g., shift or day.
r c†: expected delivery time after the production to customer c.
r j†: cycle time from the managing point j to the ®nal production.
r w†: expected time to be taken from the ®nal production to the warehouse or distribution
center w.
X i; c; t†: production volume at the end of the manufacturing position of device i for
customer c at time period t (to be determined).
Y i; b; c; t†: volume of device i to be sold at the price of bin grade b for customer c at time
t (to be determined).
C i; b; c; t†: selling price of device i of bin grade b for customer c at time period t.
d i; b; c; t†: demand of device i of bin grade b from customer c at time period t.
g i; b†: proportion of bin grade b from device i.
STD i; j†: time required to process device i on the managing point j.
CAP t; w†: capacity of warehouse or distribution center w at time period t.
WT: total working time during one unit of period.

Formulation C

X
Max C i; b; c; t†6Y i; b; c; t†
i; b; c; t
X X
s.t. X i; c; t†6 g i; b0 †  Y i; b0 ; c; t† Vi; c; t
b0  b b0  b

Y i; b; c; t r c††  d i; b; c; t† Vi; b; c; t
X
X i; c; t r j††6STD i; j†  WT Vt; j
i;c
X
X i; c; t ‡ r w††  CAP t; w† Vt; w
i;c

X i; c; t†  0; Y i; b; c; t†  0 Vi; b; c; t
SUPPLY CHAIN MODEL 201

Y i; b; c; t† denotes the volume of devices sold at bin grade b including one with higher
grade than b, which represents the substitution. The ®rst constraint is on the binning and
substitution. The value of g i; b† can be found or predetermined from the history data.
The second constraint describes production volumes to be determined at least customer's
order quantity. The working time constraint has to be applied on the managing points on
manufacturing process as shown in the third constraint. The fourth speci®es the
warehouse and distribution center capacity constraint on the logistics to the customers.
Our model assumes the ®nal assembly and test factory is located at one place and
managed by one organization. For the multiple ®nal factories this model can be modi®ed
easily by adding an index for each.

4. System Modeling and Implementation

Individual functions discussed in supply chain management are in fact being


implemented currently in most companies. They are partially connected each other via
the database from which information and knowledge are extracted. Problems are
formulated to ®nd solutions within a local network depending on organizations and
locations they are concerning. Ef®ciency and effectiveness in a local network must be
restrictive. Global solutions on an integrated network are necessary as the market is
distributed all over the world, especially for the semiconductor industry. One of the
factors expediting the global approach is the use of information technology on all
business processes. Large scale of data transactions and repetitive data computation is
possible on the system that makes the business decisions more accurate and fast. MRP
and enterprise resource planning (ERP) are typical examples for the transaction and
decision making support systems using information technology. Information technology
makes the market more competitive, and enables the company to meet the dynamic
change of market. Without help of information technology it is not possible to make
decisions on integrated supply chain model. Models suggested above can be implemented
after corresponding systems are developed and interfaced with the database needed to
obtain solutions. Even though only two supply chain models we have for full business
process, several systems have to be developed for implementation, and several layers of
system hierarchy are required. The push and pull supply chain model described as above
need more considerations.
New modeling for the supply chain of the semiconductor industry may lead to easier
development of the information system. Many functions and the modules the most supply
chain commercial package has are not necessary for the semiconductor industry.
Inef®ciency that comes from the heavy system for the supply chain can be eliminated
through the new modeling. Umeda and Jones [26] and Jain et al. [12, 13] have also
suggested the simpler model than the generic supply chain system. The system can be
developed based on the solution modeling as above. Information on the supply chain has
been delivered backward from the customer to the raw material suppliers via the serial
information block (Figure 4). The number of steps of the information transition in the
model can be reduced.
202 LEE

For the push supply chain of the front-end, the so-called push chain manager has to be
developed in pursuit of the high production and balanced line. Production planning and
scheduling functions are most important to support the push chain manager. Solutions for
push supply chain can be obtained using formulations A or B as suggested in a previous
section, and they have to be transferred to the lower level of systems, e.g., production
planning and scheduling system to generate manufacturing data to be implemented in
detail. Manufacturing orders obtained are sent to the shop ¯oor control level to operate
the equipment loading and unloading, automated guided vehicles, stacks, etc.
In the pull supply chain of the back-end, continuous and real-time type information
transfer is required to achieve the high performance of customer satisfaction. The system
controlling the pull chain in the model is, say, the pull chain manager. Its role is to control
the distribution of the product according to the customer order and inventory policies of
the intermediate distribution points. Solutions for pull supply chain can be obtained using
formulation C, and they can be used to make decisions on orders for the distribution and
allocation for each manufacturing bank, warehouse, distribution centers, and customers.
Manufacturing process from the decoupling point to the ®nal products is performed on
the concept of pulling system by order information from the customers. Their behavior
must be different from those of push manufacturing portion. Equipment is allocated to
devices by their priority and urgency that are the objective performance measure in the
pull supply chain model. More emphases are given on the revenue rather than utilization
of equipment. Final products are stored and transported to warehouses and distribution
centers by their destination. As the shop ¯oor level system is mandatory for the push

Figure 4. The suggested supply chain model.


SUPPLY CHAIN MODEL 203

supply chain model, the logistics control system is required for implementation of the
pull supply chain model. Solutions on the distribution, transportation and delivery to their
destination are transferred to the logistics control systems that manages the resources of
corresponding facilities. They include the inventory control system for the logistic
centers, transportation and delivery control systems of which module generates orders for
the vehicle operation and routing decisions. Figure 5 shows the general mechanism that
the solution of the supply chain model can be implemented.
The main difference from the general supply chain model is that the part of the
production process, the back-end, is regarded as the logistics process in this model. The
supply chain manager integrating the push and pull chain manager works as the
coordinator over the whole processes. The ¯exibility in controlling the market can be
achieved by the proper management of two managers. For instance, on the buyer's market
the supply chain manager applies the pull supply chain tightly, and makes sure that input
data for the push chain system is collected and adjusted from the pull supply chain. The
front-end production system then behaves to satisfy the downstream process within the
capacity allowed. On the other hand, when the product demand is more than the capacity,
the pull chain manager receives the input data from the front-end system to implement
the logistics within the maximum volume the fabrication line can produce.
Order promising is an activity whose purpose is to produce a quotation in response to a
customer request, and quotation is a statement of price, description of products, due dates,

Figure 5. System models of supply chain functions.


204 LEE

and other terms of sale offered by a supplier to a prospective customer [1]. They need to
answer an incoming inquiry within a very short term, to con®rm another feasible due date
if the customer request cannot be ful®lled, and to establish accurate and highly reliable
commitments with the customers. Companies have their own policies for the order
promising and quotation. One of them is a hybrid policy consisting of soft pegging and
hard pegging, where hard pegging allocates the speci®c orders to the real WIP in
manufacturing or ®nal products on storage and transportation while soft pegging does on
volume of the production target, not by the speci®c items. For example, devices allocated
to certain orders according to hard pegging policy have to be delivered to designated
customers whatever the situation is, while devices allocated on soft pegging policy can be
delivered to others by the sequence of output. Hard pegging policy is appropriate for the
production items on pull supply chain model, and soft pegging for the push supply chain
model. Quotation process can be implemented with help of ERP system, but results of
them are to be used as input data for the push supply chain model, hence they should be
correct within the production capacity. Functional description and system models for the
order promising and quotation can be found in Azevedo et al. [1]. We did not include
procedures on the raw material replenishment into the supply chain model since it is
relatively simple in the semiconductor industry, and not critical in the decision making
process of supply chain. Traditional MRP can be applied for the procedure of raw
material procurement after the production targets for each managing points are
determined.

5. Conclusions and Discussion

In this paper, we have discussed the supply chain model for the semiconductor industry of
global market. Semiconductor manufacturing process has the special characteristics such
as re-entrant cycling, probabilistic binning split, and resultant substitution for the graded
productions. For the manufacturing itself, many researchers have studied for the effective
planning and scheduling policies. We have suggested the integrated supply chain model
from the start of the manufacturing to the delivery to customers, which is distributed all
over the world. In our model, die bank between the front-end and back-end in the
manufacturing process is assumed to be the decoupling point by which two models are
divided. Push supply chain model is based on the high throughput and the balance of the
line, which takes the characteristics of re-entrant mechanism into account. Pull supply
chain model is constructed considering order-based manufacturing, distribution and
transportation to satisfy customers. Commercial software packages are in the market and
implemented in many companies, but most of them are modeled based on the general
assembly manufacturing industry and converted for application in the semiconductor
industry. Competition in the semiconductor industry is getting tighter these days, and
prices of the devices are dropping rapidly. Ef®ciency that can be obtained by applying the
appropriate supply chain model will be important. Our model needs to be tested on the
real ®eld with real data, which is our next research endeavor.
SUPPLY CHAIN MODEL 205

References

1. A. Azevedo and J. P. Sousa, ``On the design of an order promise system for virtual enterprises,'' http://
www.nimblesite.com/xcittic, 1996.
2. B. C. Arntzen, G. G. Brown, T. P. Harrison, and L. L. Trafton, ``Global supply chain management at Digital
Equipment Corporation,'' Interfaces 25(1), pp. 69±93, 1995.
3. G. R. Bitran and D. Tirupati, ``Development and implementation of a scheduling system for a wafer
fabrication facility,'' Operations Research 36(3), pp. 377±395, 1988.
4. J. L. Bouff, ``Semiconductor manufacturing: A supply chain management perspective,'' in Conference on
Global Supply Chain Management, Singapore, 1997.
5. L. Demeester and C. Tang, ``Reducing cycle time at an IBM wafer fabrication facility,'' Technical Paper,
University of California at Los Angeles, 1994.
6. B. Ehteshami, R. G. Petrakian, and P. M. Shabe, ``Trade-offs in cycle time management: Hot lots,'' IEEE
Transactions on Semiconductor Manufacturing 5(2), pp. 101±106, 1992.
7. S. S. Erenguc, N. C. Simpson, and A. J. Vakharia, ``Integrated production/distribution planning in supply
chain : an invited review,'' European Journal of Operational Research 115, pp. 219±236, 1999.
8. F. Frederix, ``Planning and scheduling multi-site semiconductor production chains,'' Esprit Integration in
Manufacturing Conference, Galway, Ireland, 2±4 October, 1996.
9. C. R. Glassey and M. G. C. Resende, ``Closed-loop job release control for VSLI circuit manufacturing,''
IEEE Transactions on Semiconductor Manufacturing 1(1), pp 36±46, 1988.
10. C. R. Glassey and M. G. C. Resende, ``A scheduling rule for job release in semiconductor fabrication,''
Operations Research Letters 7, pp. 213±217, 1988.
11. P. G. Glassey, ``A comparison of release rules using BLOCS/M simulations,'' ESRC Report 90-15, 1990.
12. S. Jain, B. Gan, C. Lim, and Y. Low, ``Criticality of detailed modeling in semiconductor supply chain
simulation,'' Proceedings of the 1999 Winter Simulation Conference, pp. 888±896.
13. S. Jain, B. Gan, C. Lim, and Y. Low, ``Bottleneck based modeling of semiconductor supply chains,''
Proceedings of the MASM 2000 Conference, pp. 340±345.
14. S. Kim, S. Yea, and B. Kim, ``Stepper scheduling in semiconductor wafer fabrication process,''
Proceedings of the International Conference on Modeling and Analysis of Semiconductor Manufacturing
2000, pp. 157±162.
15. C. Kraft, ``Dynamic kanban semiconductor inventory management system,'' Technical Paper, Texas
Instruments Inc., 1993.
16. R. C. Leachman, ``The competitive semiconductor manufacturing survey: Second report on results of the
main phase,'' CSM-08 Report, 1994.
17. H. L. Lee and C. Billington, ``Material management in decentralized supply chains,'' Operations Research
41, pp. 835±847, 1993.
18. H. L. Lee, C. Billington, and B. Carter, ``Hewlett Packard gains control of inventory and service through
design for localization,'' Interfaces 23(4), pp. 1±11, 1993.
19. C. Liu, ``A modular production planning system for semiconductor manufacturing,'' Ph.D. Dissertation,
University of California, Berkeley, 1992.
20. S. X. C. Lou and P.W. Kager, ``A robust production control policy for VLSI wafer fabrication,'' IEEE
Transactions on Semiconductor Manufacturing 2(4), pp. 159±164, 1989.
21. C. Makatsoris, N. P. Leach, H. D. Richards, C. Besant, and M. Ristic, ``Addressing the planning control
gaps in a semiconductor virtual enterprises,'' Esprit Integration in Manufacturing Conference, Galway,
Ireland, 2±4 October, 1996.
22. S. V. Murty and J. W. Bienvenu, ``Global planning at Harris Semiconductor,'' IEEE International
Symposium on Semiconductor Manufacturing, 1995.
23. I. M. Ovacik and W. Weng, ``A framework for supply chain management in semiconductor manufacturing
industry,'' IEEE/CPMT International Electronics Manufacturing Technology Symposium, 1995.
24. J. F. Shapiro, V. M. Singhal, and N. W. Wagner, ``Optimizing the value chain,'' Interfaces 23(2), pp. 102±
117, 1993.
25. M. L. Spearman, D. L. Woodruff, and W. J. Hopp, ``CONWIP: A pull alternative to kanban,'' International
Journal of Production Research 28(5), pp. 879±894, 1990.
206 LEE

26. S. Umeda and A. Jones, ``An integrated test-bed system for supply chain management,'' Proceedings of the
1998 winter simulation conference, pp. 1377±1385.
27. R. Uzsoy, C. Y. Lee and L. A. Martin-Vega, ``A review of production planning and scheduling models in the
semiconductor industry, Part I: System characteristics, performance evaluation, and production planning,''
IIE Transactions 24(4), pp. 47±60, 1992.
28. R. Uzsoy, C. Y. Lee, and L. A. Martin-Vega, ``A review of production planning and scheduling models in
the semiconductor industry, Part II: Shop-¯oor control,'' IIE Transactions 26(5), pp. 44±55, 1994.
29. L. M. Wein, ``Scheduling semiconductor wafer fabrication,'' IEEE Transactions on Semiconductor
Manufacturing 1(3), pp. 115±129, 1988.
30. L. M. Wein, ``On the relationship between yield and cycle time in semiconductor wafer fabrication,'' IEEE
Transactions on Semiconductor Manufacturing 5, pp. 156±158, 1992.

You might also like