A Novel Route Selection And Resource Allocation Approach To Improve The Efficiency Of Manual Material Handling System In 200 Mm Materia

User Manual:

Open the PDF directly: View PDF .
Page Count: 14

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 13, NO. 4, OCTOBER 2016 1567

A Novel Route Selection and Resource Allocation

Approach to Improve the Efficiency of Manual

Material Handling System in 200-mm

Wafer Fabs for Industry 3.5

Chen-Fu Chien, Member, IEEE, Che-Wei Chou, and Hui-Chun Yu

Abstract—Motivated by realistic needs to enhance the

productivity for 200-mm wafer fabs, this paper aims to propose a

novel approach for manual material handling system (MMHS) to

mimic functionalities of the automated material handling system

in the advanced fabs without intensive capital investment to

deliver the wafer lots manually and systematically. In particular,

a mathematical model is developed to optimize the routing plan

with two objectives that minimize the total traveling distance in

all routes or minimize the number of manpower needed in all

routes. Furthermore, a route planning approach is proposed to

utilize the routes that reduce the technician traveling distance

and transportation time for implementation. Also, a manpower

loading index was developed for evaluating the number of needed

technicians in the proposed MMHS. To estimate the validity of the

proposed MMHS, we developed a simulation environment based

on empirical data with different transportation requirement

scenarios for comparison. The results have shown practical

viability of the proposed approach.

Note to Practitioners—As advanced manufacturing strategies

such as Industry 4.0 are proposed for smart production, 200-mm

wafer fabs cannot be equipped with fully automation facilities

such as the automated material handling system to enhance

overall productivity. To address the needs in real settings,

a disruptive innovation manual material handling system was

developed, on the basis of existing 200-mm fab facility, to

organize the technicians to mimic the setting of a virtual material

handling system manually to enhance productivity. Indeed, the

developed solution has been implemented in this case company,

in which the results have validated the proposed approach that

can be a hybrid between the existing Industry 3.0 and to-be

Industry 4.0.

Index Terms—Fab economics, Industry 3.5, manpower allo-

cation, manual material handling system (MMHS), productivity,

route planning.

Manuscript received April 30, 2016; accepted June 11, 2016. Date of

publication July 19, 2016; date of current version October 4, 2016. This

paper was recommended for publication by Associate Editor M. Liu and

Editor H. Ding upon evaluation of the reviewers’ comments. This work

was supported in part by the Ministry of Science and Technology, R.O.C.

under Grant NSC 102-2221-E-007-057-MY3, Grant NSC 103-2622-E-007-

002-TM1, Grant NSC 102-2622-E-007-013, and Grant MOST 103-2218-E-

007-023, in part by the Toward World Class University Project within the

Ministry of Education under Grant 105N536CE1, and in part by Taiwan

Semiconductor Manufacturing Company under Grant 102A0287JC.

The authors are with the NTHU-TSMC Center for Manufacturing Excel-

lence, Department of Industrial Engineering and Engineering Manage-

ment, National Tsing Hua University, Hsinchu 30013, Taiwan (e-mail:

cfchien@mx.nthu.edu.tw).

Color versions of one or more of the figures in this paper are available

online at http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TASE.2016.2583659

I. INTRODUCTION

SEMICONDUCTOR fabrication facilities (fabs) are the

most capital-intensive and complex manufacturing plants

that consists of lengthy re-entrant processes including clean-

ing, oxidation, deposition, metallization, lithography, etching,

ion implantation, photoresist strip, inspection, and measure-

ment [1]. The wafers pass through approximately several

hundred processing steps for wafer fabrication, in which opera-

tional efficiency and productivity enhancement via maximizing

the throughput and yield, while minimizing cycle time, are

critical for maintaining competitive advantages [2], [3].

Automation in modern fabs enables efficient material han-

dling between resources to reduce cycle time and manufactur-

ing cost [4]. In particular, the advanced 300-mm fabs rely on

automated material handling system (AMHS) to manage the

wafer transportation in fabs [5], [6]. Furthermore, Germany

has proposed a manufacturing strategy, Industry 4.0 [7], for

smart factory via cyber-physical systems and decentralized

decisions within a smart and networked platform. However,

most existing 200-mm fabs that find it difficult or cost effective

to install AMHS employ technicians maneuvering the trolleys

for moving the wafer lots [8].

Motivated by realistic needs to empower 200-mm wafer

fabs, this paper aims to propose a disruptive innovation via

manual material handling system (MMHS) that mimics the

AMHS functionalities by technicians and reduces the trolley

accidents effectively. However, since the technicians may

decide by themselves the wafer lots and the corresponding

transportation route, some lots may be delayed causing cycle

time increase, while serious trolley accidents happen causing

injuries and yield loss. It is important to determine the

standard operating procedures and manage the transportation

routes for the technicians to avoid these issues. In particular,

a mathematical model is developed to optimize the routing

plan with two objectives that minimize the total traveling

distance in all routes or minimize the number of manpower

needed in all routes. Furthermore, a route planning approach

is also proposed to utilize the routes that reduce the technician

traveling distance and transportation time in a short time for

implementation. Also, a manpower loading evaluation model

is developed for determining the appropriate number of tech-

nicians for the MMHS. In order to estimate the validity of the

proposed MMHS approach, a simulation model was developed

1545-5955 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.

See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

1568 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 13, NO. 4, OCTOBER 2016

based on empirical data collected in the largest 200-mm fab

in a leading semiconductor manufacturing company in Taiwan

for comparison under different scenarios. The results have

shown practical viability of the proposed approach. Indeed,

the proposed MMHS has been employed in the case study.

Indeed, the proposed approach can be Industry 3.5, namely,

a hybrid approach between the existing Industry 3.0 and the

to-be Industry 4.0 platform.

The remainder of this paper is organized as follows.

Section II reviews related approaches for material handling for

semiconductor manufacturing. Section III describes the pro-

posed approaches for route planning and manpower evaluation.

Section IV estimates the validity of the proposed approach

with simulation and scenario analyses. Section V discusses

the implementation of this approach. Section VI concludes this

paper with a discussion of contributions and future research

directions.

II. LITERATURE REVIEW

The wafers pass through hundreds of production steps of re-

entrant processing flows in the fab, in which material handling

of the wafers is critical to enhance productivity. Thus, AMHS

is effectively employed in advanced fabs to enable efficient

delivery of the wafer lots [5], [6]. In particular, the facility

design and material handling system design are two major

designs that will influence the fab productivity [5]. Many

studies have addressed the design of the AMHS. For example,

Peters and Yang [9] proposed a network flow formulation

to integrate the layout and material handling system design

for both spine and perimeter layout configurations. More-

over, Ting and Tanchoco [10] proposed two rectilinear layout

configurations, the unidirectional loop and bidirectional loop

layout, which connected the tools to stockers to minimize the

total loaded travel distances in fabs.

Indeed, the spine unidirectional configuration has no inter-

section and the involved traffic management, vehicle routing,

and dispatching decisions are simple [11]. Therefore, the spine

configuration is a general layout of the AMHS with a loop in

one direction.

Similar to a general tandem configuration, segmented flow

configuration design consists of one or more mutual zones

that are nonoverlapping segments with a single vehicle serving

each segment [12]. Indeed, the segmented flow configuration

design has higher efficiency than conventional systems by

eliminating the congestion and blocking of vehicles [13].

However, the segmented flow configuration design requires

additional transfer stations [14]. Hsieh et al. [15] proposed

a segmented configuration design for AMHS which used a

dual-track bidirectional loop design to eliminate congestion

and blocking to reduce the cycle time and increase stocker

utilization.

Yu and Egbelu [16] developed a partitioning algorithm

that grouped the workstations into a number of single-vehicle

zones based on variable path routing for nonoverlapping

tandem configuration while minimizing the number of vehicles

and satisfying the total workload. In addition, meta-heuristic

approaches were employed to deal with the partitioning

problem of the tandem configuration design [17], [18].

Aarab et al. [19] developed a hierarchical classification

approach to group the workstations based on a similarity

coefficient considering the flows and the distances between

workstations.

Vehicle fleet sizing problem for determining the number

of vehicles is critical for achieving the performance and

minimizing the vehicle cost for AMHS. A number of studies

have developed mathematical programming models [20]–[22],

queuing models [23]–[25], or simulation models [26]–[28] to

solve the vehicle fleet sizing problem. Furthermore, multiload

vehicles were proposed to picks up additional wafer lots while

transferring a previously assigned load, indicating better per-

formance than single-load vehicles via simulation [29], [30].

Hung and Liu [31] modified previous single-load vehicle

studies and proposed a model to estimate the number of

multiload vehicles needed to improve the system performance.

However, a complicated material handling control system is

required to employ multiload vehicles [14].

The material handling control systems [12], [14] were

developed to fulfill transportation demands and avoid traffic

conflicts among the vehicles with the following functions:

1) vehicle dispatching rule to select a vehicle to execute a

transportation demand;

2) vehicle scheduling to dispatch the vehicles for trans-

portation demands under certain constraints such as

capacity and priority;

3) vehicle routing to find a suitable path for a pickup or

delivery of a dispatched wafer lot.

It is difficult to schedule the vehicles over a long horizon

in a fab with high arraival stochasticity. In order to quickly

respond to transporation requests, heuristic dispatching rules

are employed to reduce cycle time, wafer waiting time, and

delivery time, as well as to improve throughput, vechicle

utilization, and machine utilization [32], [33]. A number of

studies have been done to address route planning that decides

the route for a vehicle to picks up the working pieces and the

visiting sequence with the most efficient resource utilization

and the minimum traveling distance and cost [34]–[36].

In conventional unidirectional (CU) route planning, Sinriech

and Tanchoco [37] proposed the optimal single-loop route

planning based on integer programming in an automated

guided vehicle system. While the vehicles ran with the same

direction path and uniform speed, the material handling system

would not congest with the optimal single-loop route. How-

ever, the automated guided vehicle system may not be suitable

for a large-scale material handling system with great number

of vehicles and stations [34]. Asef-Vaziri et al. [38] developed

an optimal unidirectional loop route to serve manufacturing

cells in an automated guided vehicle system, in which the

vehicle transported working piece following the assigned loop

route including corresponding pickup and delivery stations.

In AMHS, the material flow is influenced by routing strate-

gies such as static routing [39] and dynamic routing [40], [41].

In particular, Kim et al. [39] proposed a simple blocking

prevention method for a path-based AMHS based on swap-

ping the load assignments between retrieval vehicles on the

same path. The simulation showed that their approach could

improve productivity under various vehicle dispatching rules.

CHIEN et al.: NOVEL ROUTE SELECTION AND RESOURCE ALLOCATION APPROACH TO IMPROVE THE EFFICIENCY OF MMHS 1569

Fig. 1. Illustration of MMHS configuration in a 200-mm fab.

Bartlett et al. [40] proposed a congestion-aware dynamic

routing approach to efficiently reroute vehicles as congestion

status changes such as vehicle breakdowns and thus reduced

the frequency of heavy congestion. Lau and Woo [41] devel-

oped an agent-based dynamic routing strategy for a loop-

based network including the generation of feasible routes,

reconfiguration of system parameters, and management of fault

situations that can generate good routes in terms of cycle time,

utilization, and ability to balance network loading.

However, the present problem addressed material han-

dling in 200-mm fabs that cannot install the AMHS and

in which the technicians maneuver the trolleys to carry

the wafer lots dynamically. There is a need to deter-

mine manpower allocation and route planning. For example,

Faaland and Schmitt [42] developed a cost-based heuristic to

determine the manpower and machines needed in the assembly

workshop. Süer et al. [43] focused on the cell loading and

product sequencing problems and proposed a three-phase

methodology to allocate manpower to the operation cells.

Guyon et al. [44] proposed an approach to simultaneously

assign a work pattern to each operator and form a feasible

schedule to solve the employee scheduling and production

scheduling problems. Chien et al. [45] employed genetic

algorithm and response surface methodology to determine the

operator–machine assignment under different product mix to

minimize machine interference time and labor cost in the IC

final testing facility.

Indeed, there is no study to address related issues for the

MMHS problem for 200-mm wafer fabs in real settings.

Therefore, a mathematical model and an analytical model

were developed in this paper to address the routing plan and

manpower evaluation.

III. METHODOLOGY

A. Problem Description

Focusing on real needs in the largest 200-mm fab in a

leading semiconductor company in Taiwan that consists of two

production lines equipped with tools in six production bays:

diffusion (DIF), chemical vapor deposition, physical vapor

deposition, etching (ETH), photo lithography (PHO), and ion

implant (IMP). The proposed MMHS has a loop layout that

connects twin fab tools with a length of more than 200 m as

shown in Fig. 1. Although the X-line and Y-line can be treated

as independent production lines, they can back up each other

to enhance overall productivity. Therefore, it is inevitable to

have cross-line transportation of the wafer lots.

TABLE I

FROM–TOFLOW TABLE

The wafer material handling system is a single-loop interbay

design to deliver the wafer lots manually by technicians.

A number of stockers are located among the production bays

to store temporary wafer lots before they are transported.

A stocker has 32 temporary racks to store the output wafers

from tools and the input wafers waiting for next processing

tool. The wafer lot was first placed onto a stocker by techni-

cian. Each stocker is regarded as an entry (or exit) point to

(from) an intrabay.

The trolleys used in the MMHS are multiload trolleys

that can store up to 12 wafer lots. According to the pri-

ority information provided by the manufacturing execution

system (MES), the technicians load the wafer lots from the

beginning stockers to the terminal stockers in the unidirec-

tional loop configuration following the interbay design of the

AMHS. Then, another technician in charge of the correspond-

ing production bay will move the wafer lots from stockers

to the tools within the production bay following the intrabay

design of the AMHS.

The from–to flow table was employed to denote the wafer

lot that moved among stockers. fij denotes the number of

wafer lots that are expected in the planning period to be moved

from stocker ito stocker j,wherei= j. An example of four

stockers in the material handling system is illustrated. Given

the number of technicians, Table I extracts the from–to flows

indicating the relationships among stockers, which illustrates

the demands via CU transportation.

The efficiency of the MMHS is difficult to evaluate through

the number of technicians and from–to flow table. Therefore,

when the technicians cannot fulfill the transport demand, the

conventional approach will add technicians to resolve the

shortage problem, yet will increase the manpower cost.

It is difficult to control and manage the MMHS system.

Since there are many stockers, numerous multiload trolleys,

and cross-line transportation in the 200-mm fab, the route

planning is complicated. The transportation control system of

MMHS is not similar to AMHS which could receive all the

transportation requests and assign vehicles immediately by the

computational control system. In addition, the technicians are

not automatic vehicles. They could not execute the compli-

cated routing plan accurately. The technician tends to have his

or her discretion for transporting the wafer lots and thus may

cause traffic accidents and wafer damages. Hence, it is very

important to develop a simple and highly efficient routing plan

to meet the transportation request in MMHS.

This paper focused on the interbay design of the MMHS

to assign the route and allocate the corresponding stockers

to the technicians to improve the efficiency and reduce the

1570 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 13, NO. 4, OCTOBER 2016

average transportation time of the MMHS. Based on domain

knowledge, three indices including efficiency (E), minimal

number of manpower needed (G), and average transportation

time (TT) were used to evaluate the MMHS performance and

thus derive the optimal solution. In particular, the efficiency

is calculated by dividing the total transportation demand (F)

by the total traveling distance (D) in the planning period as

follows:

E=F

D(1)

where Fdenotes the total transportation demand, which is the

sum of all wafer lots from stocker ito stocker j, i.e., F=

ijfij.Dindicates the total traveling distance, which is

the sum of all trolley’s traveling distance, i.e., D=gdg,

where dgrepresents the traveling distance by technician (g).

That is, the more wafer lots are transported per distance, the

more efficient the MMHS is.

The manpower needed is the minimal technicians required

to fulfill transportation demand while keeping the MMHS

uncongested. The average transportation time is the sum of

the average waiting time on the stocker (ws) and the average

transportation time on the trolley (wt). The waiting time begins

from the wafer lot put on stocker until the wafer lots are

loaded onto the transportation trolleys. The transportation time

on the trolley is from the time the wafer lots are loaded

onto the trolleys to the time they arrive at the terminal stocker.

The transportation time is regarded as the service level of

the MMHS.

B. Mathematical Model

Before further discussion, the notations used below are

summarized as follows.

Indices:

i,j,p,qIndex of stockers, i,j,p,q=1,2,...,N.

sIndex of beginning stockers, s=1,2,...,N.

rIndex of routes, r=1,2,...,R.

Parameters:

fij Number of wafer lots expected to be moved from

stocker ito stocker jin the planning period,

where i= j.

λrTrolley’s traveling frequency in route rin the

planning period.

vijr Wafer transported from stocker ito stocker jin

route r.

tsi Transportation time from beginning stocker sto

stocker i.

dsi Distance between stocker sand stocker i.

vTransportation velocity of a trolley.

Lir Number of wafer lots on a trolley while it stops

at an intermediate stocker iin route r.

ArTotal trip time of trolley in route r.

tl Loading time from beginning stocker to trolley.

tu Unloading time from trolley to destination stocker.

LDrTotal wafer loading time by technician in route r.

ULDrTotal wafer unloading time by technician in

route r.

OT Operation time for transportation technician.

drTotal trip distance of all trolleys based on route r.

cTrolley capacity.

MLarge positive number.

Decision Variables:

grNumber of trolleys needed in route r.

xir =1,if trolley picks up wafer from stocker iin route r

0,otherwise

xsr =1,if trolley stops at stocker sin route r

0,otherwise.

The mathematical programming formulation of the MMHS

routing problem is described as follows:

Objective function

Min

R



r=1

dr(2)

or

Min

R



r=1

gr(3)

subject to

dr≥2dsiλr−M(1−xir)∀r,s,i=s+1,...,N

(4)

Lir ≥

i−1



p=s

N



q=i+1

vpqr +

N



q=i+1

i−1



p=s

vqpr

−M(1−xir)∀r,s,i=s+1,...,N−1(5)

R



r=1

vijr =fij ∀i,j(6)

N



i=s+1

xir ≤(N−s)xsr ∀r,s(7)

λr≤Mxsr ∀r,s(8)

N



j=1

vijr +

N



j=1

vjir ≤Mxir ∀r,s(9)

j−1



p=s

N



q=j

vpqr ≤cλr∀r,s,j=s+1,...,N(10)

j−1



p=s

N



q=j

vqpr ≤cλr∀r,s,j=s+1,...,N(11)

Ar≥2λrmax

i{tsixir}−M(1−xir)

∀r,s,i=s+1,...,N(12)

LDr=tl

N



i=1

N



j=1

vijr ∀r(13)

ULDr=tu

N



i=1

N



j=1

vijr ∀r(14)

gr≥(Ar+LDr+ULDr)/OT ∀r(15)

xir ∈{0,1}∀i,r(16)

CHIEN et al.: NOVEL ROUTE SELECTION AND RESOURCE ALLOCATION APPROACH TO IMPROVE THE EFFICIENCY OF MMHS 1571

xsr ∈{0,1}∀s,r(17)

gr≥0,and integer ∀r(18)

λr≥0,and integer ∀r(19)

where

E=N

i=1N

j=1fij

R

r=1dr

(20)

G=

R



r=1

gr.(21)

Objective function (2) is used to minimize the total traveling

distance in all routes. Alternatively, objective function (3)

arms to minimize the number of manpower needed in all

routes. Constraint (4) specifies the total trip distance drbased

on route r.dris obtained by multiplying two (for round

trip), the distance between the beginning stocker sand the

end stocker iof the route r, and its service frequency λr

together. The large positive number Mand the 0–1 variable

xir in constraint (4) ensure that dris the maximum of the

distance dsi between stocker sand all stockers iin the

route r. If a route ris not formed, i.e., all xir =0,

the large positive number Mwill make dr=0. Constraint

(5) defines Lir that is determined by the wafer lots pickup

at a previous stocker pand alight at a subsequent stocker q,

i.e., Lir is the total wafer lots (vpqr and vqpr)servedbythe

round trip of the route rbetween stocker pand stocker q.

Constraint (6) denotes the transportation demand fij between

stockers iand jwhich must be fulfilled by all wafer lots

vijr served by all routes. Constraints (7) and (8) obtain route

r. All trolley trips based on route rstart from a beginning

stocker s,wherexsr =1, i.e., the trolley stops at stocker s,

and load and unload wafer lots. If xsr =0, route rcannot

be formed while trolley does not stop at stocker s,i.e.,all

xir in constraint (7) and the trip frequency λrin constraint

(8) are equal to zero. Constraint (9) ensures that no wafer

lot can load or unload at stocker i, if route rdoes not stop

at stocker i. Constraints (10) and (11) are capacity limitation

constraints that denote total wafer lots served by a one-way

trip of route rwhich must be accommodated by the trolley

capacity of the corresponding trips. Constraint (12) defines the

total trip time of trolley Arin route rwhich is restricted by

multiplying its traveling frequency λrand its running time of

round trip between the beginning stocker sand ending stocker

i(maxi{tsixir }), i.e., considering the running time by maximal

distance of a round trip. Constraints (13) and (14) denote

the total wafer loading and unloading time by technician

in route r. Constraint (15) determines the minimum number

of trolley needed, gr, in route r. For each route r,the

number of trolley required is the sum of total trolley traveling

time, loading time, and unloading time, divided by operation

time of transportation technician. Constraints (16) and (17)

define the binary variables of decision variables xir and xsr.

Constraint (18) obtains decision variable gras a nonnegative

real number. Constraint (19) denotes the trolley’s traveling

frequency λras a nonnegative real number. Constraint (20)

obtains the efficiency of a MMHS system which is cal-

culated by dividing the total transportation demand by the

TABLE II

FROM–TOFLOW TABLE OF THE SMALL-SCALE NUMERICAL EXAMPLE

TABLE III

DISTANCE BETWEEN EACH STOCKER OF THE

SMALL-SCALE NUMERICAL EXAMPLE

TABLE IV

RESULT OF THE SMALL-SCALE NUMERICAL EXAMPLE

total traveling distance. Constraint (21) indicates the minimal

manpower needed which is the sum of number of trolleys that

maneuvered by technicians in route r.

To validate the mathematical model, a small-scale numerical

example was demonstrated. In this case, there are six stock-

ers in the MMHS. The transportation demand of MMHS is

shown in Table II, and the distance between each stocker is

listed in Table III. The transportation velocity of a trolley (v)

was set at 0.56 m/s. The operational time for loading (tl)and

unloading (tu) of wafer lots from stockers was set as 10 s. The

capacity limitation of trolley is 12 wafer lots. In this paper, the

small-scale test problem was solved through the mathematical

model by using LINGO software.

First, we considered objective function (2) to minimize

the total traveling distance in all routes that were subject

to constraints (4)–(21). Then, LINGO was used to generate

the optimal solution to the mathematical model. Second, we

alternatively considered objective function (3) to minimize

the manpower needed in all routes that were also subject to

constraints (4)–(21). Table IV illustrates an optimal route plan

that includes the manpower needed, total traveling distance,

and the efficiency of the MMHS system.

1572 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 13, NO. 4, OCTOBER 2016

In Table IV, there are two routes allocated to two tech-

nicians for minimizing total traveling distance, 1306.8 m,

in objective function (2). One technician transported wafer

lots in round trip with first route (1-2-3-4-5-6) that stops at

each stocker. The other technician transported wafer lots only

between stocker #2 and stocker #3 with the second route (2-3).

Alternatively, one route was allocated to one technician

for minimizing total manpower in objective function (3).

The technician was assigned to the route (1-2-3-4-5-6) to

transport wafer lots with total traveling distance in MMHS.

In conclusion, the optimal route plan was determined by the

objective function. The proposed mathematical model can

address the present MMHS problem.

However, the proposed mathematical model may not be able

to generate the optimal route plan in real settings because

of computational limitations. Indeed, the MMHS problem in

200-mm fab is a complex and large-scale problem. It takes

much computational time to solve the mathematical model

in MMHS. It is not efficient in semiconductor manufacturing

environment. Furthermore, the mathematical model that is

formulated with different objectives could solve the MMHS

problem to minimize the total traveling distance in all routes

or minimize the manpower needed in all routes. However, it

is crucial to simultaneously consider multiple and conflicting

objectives to optimize the route plan and manpower allocation

in real MMHS settings. Therefore, this paper proposed an

analytical model to solve the MMHS problem in real setting.

C. Proposed Analytical Model

This paper proposes a route planning approach to deter-

mine the routes and to allocate the corresponding number of

technicians simultaneously. Two types of transportation flows

are considered: forward transportation (FWDT) and backward

transportation (BWDT). A technician starts the FWDT by

moving forward in a loop configuration to transport the wafer

lots from the left-side stocker to the right-side stocker. When

a technician reaches the end of the lane, he or she starts

the BWDT that rolls over to deliver the wafer lots from the

right-side stocker to the left-side stocker. The BWDT needs to

confirm all the transportation demands to be fulfilled, while the

demands arise from left-side stockers to right-side stockers.

The path flow (pij)is the total number of wafer lots

that pass along the corridor between stocker iand the next

stocker jin the planning period, that is

pij =

p≤i

q≥j

fpq ∀i,j,p,q=1,2,...,N;i= j(22)

where fpq is the wafer lots needed to be transported from

stocker pto stocker q,andNdenotes the total number of

stockers. While fpq is given in this paper, the path flow is not

only a metric to measure the congestion of the MMHS, but is

also regarded as a consequence of routing decision.

Fig. 2 shows an example of path flows. A technician

carries wafer lots from stocker 1 which consists of different

destination stockers. The path flow p12 is the accumulated

number of wafer lots that transport from stocker 1 to stocker 2,

which is equal to f12 +f13 +f14 based on (22). When the

Fig. 2. Illustration of path flow.

Fig. 3. Illustration of the manpower work loading of the MMHS.

technician passes through stocker 2, he or she unloads ( f12)

and loads wafer lots ( f23 +f24)to next stocker. Then, the

path flow p23 is f13 +f14 +f23 +f24. Finally, before arriving

at stocker 4, the path flow p34 is f14 +f24 +f34,wherethe

wafer lots are delivered to stocker 4.

This paper also evaluates the manpower workload for people

productivity by considering two indices: the traveling distance

and the number of wafer lots for transportation in the planning

period. This evaluation assumes that each technician follows a

repetitive working procedure of first loading the wafers from

the stockers, then transporting the wafer lots by trolleys, and

finally unloading the wafer lots to stockers. Fig. 3 illustrates

manpower workload, where the x-axis represents the total

traveling distance by all technicians and the y-axis stands for

the total number of wafer lots the technicians transported. Each

line represents the cumulative transportation capacity as each

additional technician is added. The dot shows the workload

of the technicians. When the dot is located on the diagonal

line, denoting a specific number of technicians, they are all

working at full capacity. The manpower workload is evaluated

by the transported wafer lots and the corresponding traveling

distance. The optimal number of needed technicians can thus

be derived. As shown in Fig. 3, the minimal manpower needed

to satisfy the transportation requirements is eight technicians

in the MMHS.

Based on the transportation requirements from a given

from–to flow table, the manpower can be determined for

the MMHS. In addition, this paper develops a novel route

planning (NRP) approach to design the routes to be assigned to

each of the corresponding technicians that satisfied technician

workload limitation with high transportation efficiency and

good service without congestion.

Following the route planning in Fig. 4, the ranges of routes

can be generated, in which the technicians have been assigned

a distinct route that defined the corresponding beginning and

ending stockers. While the route is assigned with a wide range,

CHIEN et al.: NOVEL ROUTE SELECTION AND RESOURCE ALLOCATION APPROACH TO IMPROVE THE EFFICIENCY OF MMHS 1573

Fig. 4. Proposed route planning procedures.

the technicians can load and unload wafer lots from more

stockers, which confirms that wafer lots could be delivered in

a reasonable waiting time. On the other hand, the technicians

that were assigned a route with a narrow range can move

more wafer lots contributing to the increase in the frequency

of round trip. That is, the technicians with a wide route could

cover all stockers fulfilling transportation demand. The tech-

nicians with a narrow route can transport more lots to increase

efficiency. Hence, the MMHS efficiency can be improved. The

proposed route planning procedures are as follows.

Step 1) Calculate the path flow pij for stocker iand the next

stocker jbased on (22). According to the from–to

table, the transportation flow can be specified for

each demand. The overall transportation flow can

be derived by computing the path flow of FWDT

and BWDT. Note that pij can be converted into

pi,i+1in FWDT and pi,i−1in BWDT.

Step 2) Identify the bottleneck path flow, max{pij}, that

contains the maximal accumulated wafer lots from

stocker ito next stocker j. Indeed, stocker iis the

earliest stocker to be blocked when using CU route

planning in which technicians traverse from the first

left-side stocker to the last right-side stocker.

Step 3) Determine the total number of technicians

needed (G)using the manpower loading evaluation

approach based on the from–to flow table without

considering route planning, i.e., all technicians

transport in the same unidirectional loop routes.

First, calculate the minimal traveling frequency

(λ) based on trolley capacity (c)considering the

bottleneck path as follows:

λ=max

i,j{pij}/c.(23)

Note that the technician is at least loaded λ−1 times with full

loads plus one more load with (0, full] load, so that Stocker i

will not be congested. Then, the total traveling distance (D)

is determined by multiplying two (for round trip), the distance

between the first stocker and the end stocker (d1,N), and its

minimal traveling frequency (λ) together as follows:

D=2d1,Nλ. (24)

Then, compute the total traveling distance and obtain all

transportation flows from the from–to flow table to derive the

total workload of the technicians in the MMHS. Finally, derive

the total manpower workload and thus determine the minimal

number of needed technicians based on their loading including

traveling distance and transportation flows. In addition, the

slope that formed by the dot and the origin in Fig. 3 is

derived by total taking wafer lots (ijfij) divided by total

traveling distance (D). The slope denotes the efficiency which

measures the transportation efficiency, i.e., an assigned flow

route, where the shorter the traveling distance is, the more the

efficiency is.

Furthermore, the following procedures are employed to

assign the route to the technician, respectively.

Step 4.1) Calculate the minimal responsible loading for each

technician, i.e., max{pij}/G. That is, each techni-

cian needs to transport at least max{pij}/Gwafer

lots from stocker i; otherwise, stocker iwill be

blocked and congested.

Step 4.2) Path flows of both FWDT and BWDT are parti-

tioned by dividing with the minimal responsible

loading. After splitting the path flows of the stock-

ers, the overall traffic flow in the MMHS can be

estimated, in which the minimal required wafer lots

for the stockers to be transported are managed to

avoid congestion.

Step 4.3) Determine the route for each technician based on

the corresponding transportation flow of techni-

cian g,(G−g/G)·max{pij}. In FWDT, the route

of technician gstarts at stocker i,wherethepath

flow is greater than (G−g/G)·max{pij}, and ends

at stocker Nof the loop configuration. Thus, the

initial stocker and the end stocker for FWDT and

BWDT are derived as follows:

FWDT:min i|pij ≥G−g

G·max{pij},N

(25)

BWDT:max i|pij ≥G−g

G·max{pij},1.

(26)

Note that the gth technician was assigned to be in

charge of the boundary stocker at the end of the

configuration. It is designed to let the technician

load as many wafer lots as possible, while the

trolley still has empty space in one trip. Hence,

the gth technician can improve utilization and

share the transportation loading of other (G−g)

technicians.

1574 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 13, NO. 4, OCTOBER 2016

Step 4.4) Ensure (G−g) technicians can handle the trans-

portation flows of inverse transportation, i.e., when

the first assigned is FWDT, the inverse trans-

portation is BWDT. While the gth technician has

been assigned a route, his traveling distance can

be reduced by reducing the transportation scope.

Thus, the other (G−g) technicians need to trans-

port the wafer lots in the inverse transportation that

the gth technician has not covered. Go to Step 4.8

and check the other technicians that have been

assigned routes; otherwise, continue to Step 4.5 to

assign the next route allocation for the (g+1)th

technician.

Step 4.5) Assign the gth and (g+1)th technicians to take

charge of the routes according to (25) and (26),

respectively, where g=g+1. If the gth technician

cannot fulfill the transportation demands with the

corresponding gth route in both FWDT and BWDT

flows, the inverse transportation will be congested

while using the gth route and the other unassigned

routes. Thus, we need to assign both the gth and

(g+1)th technicians to satisfy the transportation

requirements.

Step 4.6) Confirm that the assigned gth technician, the

(g+1)th technician, and the other unassigned tech-

nicians can satisfy the transportation requirements.

Ifnot,gotoStep4.7.Otherwise,gotoStep4.8.

As the gth and (g+1)th technicians are assigned

to the corresponding routes in Step 4.5, confirm

that the gth technician can back up transportation

requirements for the (g+1)th technician who is in

charge of the inverse transportation, respectively.

That is, the gth technician should be able to carry

out twice the required transportation in the range of

his corresponding route that substitutes the (g+1)th

transportation flow. Also, the inverse transportation

should be checked at the same time. The gth route

and (g+1)th route are determined to ensure that

the transportation system will not be congested,

since the gth and (g+1)th technicians may share

the loading of other technicians. Therefore, the

other technicians who were not given assigned

routes can still transport the wafer lots in the same

unidirectional route.

Step 4.7) If the assigned technicians, gth and (g+1)th,

cannot satisfy the transportation requirements, the

transportation system will be congested. Therefore,

the unassigned technicians are all allocated to

unidirectional routes to transport the wafer lots.

Step 4.8) Stop the route planning, if all routes have been

assigned. Otherwise, return to Step 4.3.

IV. SIMULATION AND RESULTS

Given the complex and lengthy processing steps in the fabs,

simulation has been widely used for analyzing the AMHS for

semiconductor manufacturing [27], [46], [47]. A simulation

model of the proposed MMHS was developed with eM-Plant

based on empirical data collected in a 200-mm twin fab in a

leading semiconductor manufacturing company in Taiwan to

test the feasibility and estimate the validity of the proposed

approach under different transportation requirement scenarios.

The tradeoff among the performance indices are also illustrated

in scenario analyses.

A. Simulation Assumptions

The proposed MMHS was employed for the largest

200-mm twin fab with two production lines in this empirical

study. The material handling system consisted of 42 stockers,

and the transportation requirements were obtained from real

settings. For confidentiality issues, the data and transportation

flow of the MMHS were transformed without loss of generality

for validation. The settings are designed as follows.

1) The transportation velocity of a trolley (v)was set

at 0.56 m/s.

2) The operational time for loading (tl) and unloading (tu)

of wafer lots from stockers was set at 10 s.

3) The wafer lot output time was exponentially distributed.

4) The capacity limitation of stocker was 32 wafer lots.

5) The capacity limitation of trolley c=12 wafer lots.

6) The trolley dispatch rule of the MMHS was set as the

first come first serve (FCFS).

7) The wafer lot dispatch rule of the MMHS was designed

as a weighted priority index (WP

m)considering lot

priority (LP

m), remaining queue time index (QIm),

waiting time on the stocker (WI

m), and transportation

distance index (DIm)of wafer lots, that is

WP

m=ym(α1LP

m+α2QIm+α3WI

m+α4DIm)

(27)

where there are mwafer lots on the stocker, and α1,α2,

α3,andα4denote the weight parameters, α1+α2+α3+

α4=1, that were determined by the preference of the

domain expert. LP

mrepresents the production priority

of wafer lot, scoring from 0 to 1, which is determined

by the production planner. QImis determined by the

remaining queue time (RQm)divided by its processing

limited queue time (LQm),i.e.,QIm=RQm/LQm.

WI

mis equal to the maximum waiting time of wafer lots

on the stocker (maxm{WT

m})subtracted from its waiting

time (WT

m)and divided by maxm{WT

m}, i.e., WI

m=

(maxm{WT

m}−WT

m/maxm{WT

m}).DImdefines the

transportation distance of the wafer lot (TD

m)divided

by the maximum transportation distance of wafer lots

on the stocker, i.e., DIm=TD

m/maxm{TD

m}.ym

is the Boolean binary that controls the priority index.

If the queue time (QTm)of the wafer lot longer

than its processing limited queue time, ym=0;

otherwise, ym=1.

While the trolley stopped at a stocker, the technician

checked the priority information of the wafer lot from the MES

and then transports the wafer lots according to their priorities

determined by the dispatch rule to satisfy multiple objectives.

Note that the wafer lot with smaller priority index indicates

higher priority. The technicians load and unload the wafer lots

CHIEN et al.: NOVEL ROUTE SELECTION AND RESOURCE ALLOCATION APPROACH TO IMPROVE THE EFFICIENCY OF MMHS 1575

TABLE V

SIMULATION SCENARIO

from the stockers and use trolleys to move them in a clockwise

direction.

The wafer lots whose processing from the tools in the

production bay is completed will be moved to the preassigned

stocker until they are ready to be transported by technicians

for the next process. If the storage racks of the stockers

are full, the stockers will not be available until technicians

transport previously loaded wafer lots to their destination

stockers. Meanwhile, the completed wafer lots will remain

inside the tools if they cannot be moved to next location. The

transportation will not be affected during the meal time in the

simulation since other technicians can back up.

B. Experimental Design

To validate the proposed route planning approach, three

scenarios of different utilization levels, i.e., high, medium,

and low, were examined with historical data in 200-mm

fab as shown in Table V. Six route planning approaches

were employed to compare the performances of the

proposed MMHS.

Hung and Liu [31] proposed an analytical model (AM) by

modifying the vehicle fleet size estimation model [20] to esti-

mate the multiload vehicle requirement in an automated guided

vehicle system for minimizing empty vehicle travel time

(see the Appendix). Thus, the AM approach can determine

the manpower needed and assign the technician to transport

wafer lots in the same direction in the 200-m fab.

The present approach (PS) assigns the technicians to trans-

port wafer lots according to their corresponding production

bay. For example, the technicians who are in charge of PHO

production bay carry wafer lots from the stockers located in

PHO production bay to its destination stockers (Fig. 1), i.e., the

technician’s route is assigned to only take the wafer lots from

his or her corresponding production bay and transport them to

the destination stocker in the loop configuration layout.

The CU approach computes the manpower needed by using

the manpower loading evaluation approach which is proposed

in this paper. However, the technician’s route planning follows

the CU path. i.e., CU approach assigns every technician to

transport wafer lots in one direction in the loop configuration

layout of the MMHS.

To comprehensively compare the route planning methods

in this paper, we also ran Dijkstra’s algorithm [48] which

is used to solve the shortest path (SP) problem. We applied

AM and CU approaches to determine the manpower needed

and assigned the technicians to follow Dijkstra’s algorithm for

generating route plan that is named as AM-SP and CU-SP,

respectively. When the technicians picks up wafer lots based

on weighted priority index (WPm), the system will calculate

the shortest path by Dijkstra’s algorithm. Then, the technician

Fig. 5. Partial transport output of the MMHS.

follows this request to move to the destination. In addition,

while wafer lots are loaded and unloaded, the route plan will

be updated depending on transported wafer lots that are loaded

on a trolley.

The proposed NRP approach determines the needed man-

power and allocates technicians who follow routes to transport

wafer lots. To reduce the randomness of the simulation model,

we ran ten replications for each scenario. The pilot run was

designed to reach a steady state. Therefore, only after a steady

state was reached, the warm-up period could be determined.

As shown in Fig. 5, a transient period was for 60 min and

then the data collection began and lasted for 10 days.

C. Route Planning

To demonstrate the proposed route planning approach, the

medium utilization scenario of the MMHS is used for the

following Illustration.

First, in Step 1, the path flows based on the given from–to

flow table were calculated for FWDT and BWDT, respectively.

In Step 2, max{pij}=4132 was specified, which was between

the 17th stocker and the 16th stocker in the BWDT mode.

In Step 3, the minimal traveling frequency,

λ=4132/12=345, was computed based on the

bottleneck path flow and the trolley capacity. Also, the total

traveling distance, D=2d1,Nλ=161 561, was determined.

Then, we evaluated manpower workload based on traveling

distance and transporting wafer lots. Fig. 3 shows the

transformed working hour into the number of loaded wafer

lots and the distance moved, respectively. Thus, the minimal

manpower needed for the MMHS was derived to be eight

technicians.

Second, the routes were designed as follows. In Step 4.1,

the calculated max{ pij}/Gwas 517 wafer lots, which was

equal to the minimal responsible loading. In Step 4.2, both

FWDT and BWDT were partitioned by dividing the path

flows from the minimal responsible loading of 517 wafer

lots. In Step 4.3, we assigned the first technician with the

first route {28,1} traveling in the BWDT mode using (26).

In Step 4.4, we checked whether seven technicians can

meet the transportation requirements at inverse transporta-

tion, FWDT. However, FWDT still had to be done by eight

technicians. Thus, we needed to simultaneously allocate the

first and second technicians to take charge of the respective

BWDT {28,1} and FWDT routes {14,42} in Step 4.5. As the

first technician had transported the wafer lots between the 28th

stocker and the first stocker, the second technician transported

1576 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 13, NO. 4, OCTOBER 2016

Fig. 6. Path flow chart of Step 4.6 for the medium utilization scenario.

Fig. 7. Route planning result of medium utilization in the MMHS.

wafer lots between the 14th stocker and the 42nd stocker.

To guarantee the assigned technicians fulfilling the minimal

required wafer lots, we need to check, in Step 4.6, whether

the assigned first and the second routes met the transportation

requirements in the BWDT and FWDT modes simultaneously.

For the BWDT mode, seven technicians were needed

to transport wafer lots in the range of {42,28}, yet eight

technicians were required to reach the bottleneck stocker of

the range {17,1}. However, the second technician had been

assigned to be responsible for the range of {14,42}in the

FWDT mode. Thus, it was necessary to confirm whether the

first technician could transport two-eighths of the original

transportation requirements in the bottleneck stocker of the

range {17,1}, and the other unassigned six technicians could

deliver the remaining six-eighths of the transportation require-

ments in the BWDT mode. The path flows were recalculated

based on the two-eighths transportation requirements of the

from–to flow table in the range of {17,1}. We found that

the first technician could meet the transportation flow under

his minimal responsible loading, i.e., 517 wafer lots. For

the FWDT mode, the checking procedure is the same with

different ranges, respectively. Fig. 6 shows the path flow chart

of the two-eighths requirements. That is, it was confirmed

that the assigned first and second routes could satisfy the

transportation requirement in Step 4.6. The route planning

procedures stopped after all of the technicians were assigned

in Step 4.8.

That is, as shown in Fig. 7, the route planning of the eight

routes in the MMHS is as follows: #1:{28,1}, #2:{14,42},

#3:{32,1}, #4:{10,42}, #5:{40,1}, #6:{2,42}, #7:{42,1},and

#8:{1,42}.

Thus, the range of the routes was narrowed down con-

tributing to the increase in the frequency of round trip.

TABLE VI

SIMULATION RESULTS OF THREE SCENARIOS

The technicians moved the wafer lots on the stockers with

higher transportation frequency in their corresponding routes.

Consequently, the efficiency of the MMHS and the trans-

portation time are improved with this proposed route planning

approach.

D. Simulation Results

Six approaches were compared in the simulation model

developed for the real setting MMHS problem under three

scenarios. Table VI lists the comparison results with three

measures, i.e., the efficiency (E), the manpower needed (G),

and the average transportation time (TT), in which each result

denotes the average of ten replications.

As shown in Table VI, the number of technicians determined

by the proposed manpower loading approach was fewer than

those derived from PS and AM approaches. Also, the results

are all significantly different from the two approaches, PS and

NRP, at the level α=0.05 in each scenario. Furthermore,

the efficiency comparisons showed 20.7%, 53.3%, and 67.7%

improvement rates for the high, medium, and low utilization

scenarios, respectively, showing that the proposed NRP was

more efficient than the PS approach.

In addition, the AM approach overestimates the manpower

needed because of assuming that each time an empty trolley

could load all wafer lots equal to its fully loading capacity

with the same destination stocker. Alternatively, the technician

CHIEN et al.: NOVEL ROUTE SELECTION AND RESOURCE ALLOCATION APPROACH TO IMPROVE THE EFFICIENCY OF MMHS 1577

may load wafer lots from different stockers and often

have different destination stockers. In practice, the trolley

may not be fully loaded during every trip because of the

wafer lot output rate. Therefore, the AM approach has a

long traveling distance that deteriorates the transportation

efficiency.

Compared with the average transportation time, AM-SP and

CU-SP have shorter transportation times on the trolley due

to application of the shortest path for route plan. However,

the traffic congestion happened in MMHS, which resulted

in a long transportation time on the stocker. The wafer lots

were queued on the stockers. The efficiency was worse than

that obtained using other approaches. Indeed, AM-SP and

CU-SP cannot fulfill the transportation demand of three sce-

narios even if the transport manpower is the same for AM and

CU approaches, respectively.

Furthermore, the NRP approach had a shorter transportation

time than the CU approach, implying that the waiting time

on the stocker was reduced. Note that the NRP and CU

approaches have used the same manpower loading evaluation

approach except routing planning. The NRP approach con-

siders the route planning that designed shorter empty trip for

transportation of trolley, and thus the technician could visit the

stockers and pick the corresponding wafer lots up as early as

possible to reduce the wafer waiting time on stockers. Note

that the transportation time of the PS was shorter than that

of others. However, its efficiency was significantly worse than

that of others at the level α=0.05.

Therefore, both the NRP and CU approaches were more

efficient than the PS, AM, AM-SP, and CU-SP approaches.

The NRP approach reduced the transportation time signifi-

cantly compared with the CU approach. Indeed, the proposed

NRP approach assigns the efficiency routes to the technicians

reducing the transportation time with the best performance

among other approaches.

V. DISCUSSION

The developed solution has been implemented in this case

company. The results have shown practical viability of the

proposed approach and performance indices to enhance the

efficiency of the proposed MMHS and improve the overall

fab productivity. In particular, each work shift can save two

transport technicians in MMHS, i.e., 16% people productivity

improvement, while the average transportation time is still

keep at the same service level. Meanwhile, the efficiency is

also improved by at least 20% subject to the transportation

demands.

A number of studies have investigated the control strategies

for AMHS. However, few studies have addressed similar

control strategies for MMHS in real settings. While the pro-

posed MMHS mimics AMHS, this paper developed effective

approach and performance indices to determine the routes and

allocate the corresponding manpower needed. The proposed

approach provides an adaptive and flexible control policy for

efficiently maintaining transportation performance of MMHS

in the real setting of 200-mm fab.

While AMHS was constrained in 200-mm fabs due to fab

configuration, the proposed approach provides an adaptive and

TABLE VII

COMPARISON OF AMHS AND MMHS

flexible manpower allocation and route planning to empower

the proposed MMHS. Table VII addresses the differences in

features between AMHS and MMHS.

An AMHS spends a significant share of a 300-mm fab

investment cost, U.S. $2–3 billion, as much as 3%–5% of the

total fab cost [23], [49]. Alternatively, the proposed MMHS

requires only labor cost for transportation and associated

variable cost which are much lower than the capital investment

of AMHS.

The AMHS should not be a bottleneck in wafer manufac-

turing fab, and thus AMHS design is critical for constraints

and assumptions. The minimum number of vehicles has to be

determined while AMHS is designed [12] considering different

scenarios. However, a fab configuration, capacity portfolio,

and product mix are often change as semiconductor manu-

facturing technologies migrate, causing difficult adjustments

of the constructed AMHS. Alternatively, the proposed MMHS

can adjust the needed manpower and the corresponding routing

flexibly in light of the transportation demands considering

production plan and product mix change.

The vehicle capacity is single load in AMHS [22]–[28].

However, the proposed MMHS employs multiload trolley

to picks up additional lots while transferring a previously

assigned wafer lot to improve system performance [29]–[31].

The AMHS in a 300-mm fab generally employs unidirectional

loop configuration [9], while some AMHS studies [10], [15]

1578 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 13, NO. 4, OCTOBER 2016

designed bidirectional loop configuration to eliminate conges-

tion and blocking.

Vehicle dispatch rules help AMHS and MMHS control

vehicles when a vehicle picks up wafer lots, a vehicle reaches

its destination stocker, and a vheicle unloads wafer lots.

Depending on the objective of dispatch rules, three types of

dispatch rules were addressed in AMHS [14].

1) Distance-based dispatch rules based on traveling dis-

tances or time, including shortest travel time first rule

and nearest vehicle first rule.

2) Workload-based dispatch rules considered queue sizes

of wafer lots in the stockers. There rules include maxi-

mum outgoing queue size rule and minimum remaining

outgoing queue space rule.

3) Time-based dispatch rules depended on waiting time of

wafer lots, such as the FCFS rule.

In MMHS, the time-based dispatch rule, FCFS, was applied.

While a trolley stops at a stocker, the technician will follow

a weighted priority index (WP

m)which determines the wafer

lot priority to load. The technician usually takes several wafer

lots onto the trolley due to multiload trolley design.

There are two types of routing strategies in AMHS such

as static routing [39] and dynamic routing [40], [41]. The

static route planning employs static strategies such as shortest

distance, least utilization, random assignment, round robin,

and CU algorithms for route planning. The dynamic route

planning needs to change periodically to fulfill the trans-

portation demands. The proposed MMHS has a central lane

designed for multiload trolley transferring wafer lots in a

200-mm fab. The designed routing can be adjusted dynami-

cally in light of new demands by using the proposed approach.

In general, the optimization methodologies of AMHS rout-

ing problem are classified into three catagories: mathemati-

cal optimization, heuristic, and meta-heuristic. Mathematical

optimization approaches search optimal routing plan based

on objective functions, such as transportation cost, vehicle

cost, balancing of routes, and wait time [36], [50]. Heuristics

are problem-specific approaches that take advantage of the

problem properties to derive the routing plan [39], [40].

Meta-heuristics are general heuristic schemes that can be

applied to many complicated and large-scale routing problems

in AMHS [51]. In this paper, a mathematical model that

is formulated with different objectives, minimizing the total

traveling distance in all routes or minimizing the manpower

needed in all routes, was addressed to solve the MMHS

routing problem. However, this model cannot generate the

optimal route plan because of computational limitations with

real settings. And it is also crucial to simultaneously consider

multiple and conflicting objectives to optimize the route plan

and manpower allocation. Therefore, this paper also devel-

oped a novel heuristic approach to provide an efficient route

planning.

AMHS traffic management is complicated due to large-

scale stockers and vehicles used in fab [32], [33], [39], [40].

It is a very important task to prevent the traffic congestion

and keep the wafer transportation efficient. In this paper, we

provided an NRP approach to manage all trolley’s routes with

high efficiency. Besides, this routing plan is also adaptive and

flexible, which was validated through a simulation model and

really implemented on MMHS system in a 200-mm fab based

on transportation request.

VI. CONCLUSION

To empower existing 200-mm fabs, this paper developed

a novel MMHS approach as an illustration of Industry 3.5

as a hybrid strategy between the existing manufacturing for

Industry 3.0 and to-be Industry 4.0. In particular, a mathe-

matical model was constructed to address the MMHS routing

problem that was formulated with two objectives to minimize

the total traveling distance in all routes or minimize the

manpower needed in all routes. Since it cannot generate the

optimal route plan because of computational limitations in

real settings, it is crucial to simultaneously consider multiple

conflicting objectives to optimize the route plan and manpower

allocation. This paper also developed a novel approach for

route planning that considers the path flows of segmented

configuration to obtain the route planning in light of trans-

portation requirements and a data-driven from–to flow table

and determines the manpower needed for the corresponding

routes. Empirical data collected in the largest 200-mm fab of

a leading semiconductor manufacturing company in Taiwan

were used for validation. The results have shown that the

proposed approach has fairly good performance than others

under different transportation scenarios, showing practical

viability of the developed solution. The proposed approach

can dynamically adjust both the manpower needed and the

routes in light of the change of production system to improve

the efficiency and reduce the transportation time to enhance

overall productivity of the fab. Indeed, the proposed approach

has been implemented in this company.

Further research can be done to integrate the interbay

and intrabay configurations of the MMHS while considering

extra factors such as fab capacity, product mix, and various

layout configurations. Furthermore, since most of the existing

manufacturing facilities cannot easily adopt advanced manu-

facturing strategies such as Industry 4.0 platform, the proposed

hybrid approach, i.e., Industry 3.5, should be extended to

other industries that do not have fully automated systems

for handling the materials to empower existing facilities to

address the needs for smart production. Future research can

be done to examine human–machine collaborations between

humans, trolleys, and robotics with the information collected

by multimode sensors and mobile devices in advanced man-

ufacturing systems. In addition, AMs and big data analytics

can be employed to structure the present problem to generate

the optimal solutions, provide alternative estimates, and derive

empirical rules to enhance the overall effectiveness of MMHS

in real settings.

APPENDIX

The integer linear programming model [20], [31] for esti-

mating the minimal number of multiload vehicles needed is

as follows:

Objective function

min Te=

n



i=1

n



j=1

tijxij (A1)

CHIEN et al.: NOVEL ROUTE SELECTION AND RESOURCE ALLOCATION APPROACH TO IMPROVE THE EFFICIENCY OF MMHS 1579

subject to

n



j=1

xij =

n



j=1

fij/c,∀i(A2)

n



j=1

xji =

n



j=1

fij/c,∀i(A3)

xii ≤⎡

⎢

⎢

⎢⎛

⎝

n



j=1

fji/c

⎞

⎠·⎛

⎝

n



j=1

fji/c/

n



k=1

n



l=1

fkl/c

⎞

⎠⎤

⎥

⎥

⎥

,∀i

(A4)

xij ≥0,and integer ∀i,j(A5)

where

Tt=

n



i=1

n



j=1fij/ctij (A6)

Tl=(tl +tu)

n



i=1

n



j=1

fij (A7)

G=Tt+Tl+Te

Ta.(A8)

Objective function (A1) aims to minimize the total empty

travel time (Te)that aggregates total empty trips from the

ith stocker to the jth stocker (xij)multiplied by its traveling

time (tij), where the vehicle can load with the maximal

capacity (c). Constraint (A2) guarantees that the total num-

ber of empty trips departing from the ith stocker is equal

to the number of multiload of wafer lots delivered there.

Constraint (A3) denotes that the total number of empty trips

terminating at the ith stocker is equal to the total number of

multiload of wafer lots loaded from there. Constraint (A4)

indicates that the number of empty trips depart from the ith

stocker and stop there itself (xii).xii should not be more than

the number of multiload of wafer lots delivered at the ith

stocker multiplied by the fraction of multiload of wafer lots

loaded from the same stocker. Therefore, at most xii-loaded

trips ending at the ith stocker will again start as loaded from

the same stocker. Constraint (A5) defines the positive integer

variable of decision variable xij. Constraint (A6) denotes that

the total vehicle traveling time (Tt)is equal to summation the

traveling time of all loaded trips from the ith stocker to the

jth stocker. Constraint (A7) represents the total loading and

unloading time (Tl), i.e., the summation of accessing wafer

lots loading time (tl) and unloading time (tu). Consequently,

constraint (A8) derives the total vehicles needed (G)during

the planning period, where Tais the total vehicle effective

time.

REFERENCES

[1] C.-F. Chien and C.-H. Chen, “Using genetic algorithms (GA) and a

coloured timed Petri net (CTPN) for modelling the optimization-based

schedule generator of a generic production scheduling system,” Int. J.

Prod. Res., vol. 45, no. 8, pp. 1763–1789, 2007.

[2] C.-F. Chien, Y.-J. Chen, and J.-T. Peng, “Manufacturing intelligence

for semiconductor demand forecast based on technology diffusion and

product life cycle,” Int. J. Prod. Econ., vol. 128, no. 2, pp. 496–509,

2010.

[3] C.-F. Chien, C.-Y. Hsu, and C.-W. Hsiao, “Manufacturing intelligence to

forecast and reduce semiconductor cycle time,” J. Intell. Manuf., vol. 23,

no. 6, pp. 2281–2294, 2012.

[4] R.L.Francis,L.F.McGinnis,Jr.,andJ.A.White,Facility Layout

and Location: An Analytical Approach. Englewood Cliffs, NJ, USA:

Prentice-Hall, 1992.

[5] G. K. Agrawal and S. S. Heragu, “A survey of automated material han-

dling systems in 300-mm SemiconductorFabs,” IEEE Trans. Semicond.

Manuf., vol. 19, no. 1, pp. 112–120, Feb. 2006.

[6] J. R. Montoya-Torres, “A literature survey on the design approaches and

operational issues of automated wafer-transport systems for wafer fabs,”

Prod. Planning Control, Manage. Oper., vol. 17, no. 7, pp. 648–663,

2006.

[7] H. Kagermann, W. Wahlster, and J. Helbig, Recommendations

for Implementing the Strategic Initiative Industrie 4.0: Final

Report of the Industrie 4.0 Working Group. Berlin, Germany:

Federal Ministry of Education and Research, 2013.

[8] W. C. Chen, N. Chiou, J. Y. Lu, J. Yeh, and J. Kuan, “Manual material

handling system,” in Proc. Int. Symp. Semicond. Manuf., 2008, pp. 152–

154.

[9] B. A. Peters and T. Yang, “Integrated facility layout and material

handling system design in semiconductor fabrication facilities,” IEEE

Trans. Semicond. Manuf., vol. 10, no. 3, pp. 360–369, Aug. 1997.

[10] J.-H. Ting and J. M. A. Tanchoco, “Optimal bidirectional spine layout

for overhead material handling systems,” IEEE Trans. Semicond. Manuf.,

vol. 14, no. 1, pp. 57–64, Feb. 2001.

[11] A. Asef-Vaziri and G. Laporte, “Loop based facility planning and

material handling,” Eur. J. Oper. Res., vol. 164, no. 1, pp. 1–11, 2005.

[12] I. F. A. Vis, “Survey of research in the design and control of automated

guided vehicle systems,” Eur. J. Oper. Res., vol. 170, no. 3, pp. 677–709,

2006.

[13] D. Sinriech and J. M. A. Tanchoco, “An introduction to the segmented

flow approach for discrete material flow systems,” Int. J. Prod. Res.,

vol. 33, no. 12, pp. 3381–3410, 1995.

[14] T. Le-Anh and M. B. M. De Koster, “A review of design and control of

automated guided vehicle systems,” Eur. J. Oper. Res., vol. 171, no. 1,

pp. 1–23, 2006.

[15] C.-H. Hsieh, C. Cho, T. Yang, and T.-J. Chang, “Simulation study for

a proposed segmented automated material handling system design for

300-mm semiconductor fabs,” Simul. Model. Pract. Theory, vol. 29,

pp. 18–31, Dec. 2012.

[16] W. Yu and P. J. Egbelu, “Design of a variable path tandem layout

for automated guided vehicle systems,” J. Manuf. Syst., vol. 20, no. 5,

pp. 305–319, 2001.

[17] G. Laporte, R. Z. Farahani, and E. Miandoabchi, “Designing an efficient

method for tandem AGV network design problem using tabu search,”

Appl. Math. Comput., vol. 183, no. 2, pp. 1410–1421, 2006.

[18] T. Y. ElMekkawy and S. Liu, “A new memetic algorithm for optimizing

the partitioning problem of tandem AGV systems,” Int. J. Prod. Econ.,

vol. 118, no. 2, pp. 508–520, 2009.

[19] A. Aarab, H. Chetto, and L. Radouane, “Flow path design for AGV

systems,” Stud. Inform. Control, vol. 8, no. 2, pp. 97–106, 1999.

[20] S. Rajotia, K. Shanker, and J. L. Batra, “Determination of optimal AGV

fleet size for an FMS,” Int. J. Prod. Res., vol. 36, no. 5, pp. 1177–1198,

1998.

[21] P. J. Egbelu, “Concurrent specification of unit load sizes and automated

guided vehicle fleet size in manufacturing system,” Int. J. Prod. Econ.,

vol. 29, no. 1, pp. 49–64, 1993.

[22] Y.-M. Tu, C.-W. Lu, and A. H. I. Lee, “AMHS capacity determination

model for wafer fabrication based on production performance optimiza-

tion,” Int. J. Prod. Res., vol. 51, no. 18, pp. 5520–5535, 2013.

[23] D. Nazzal and L. F. McGinnis, “Analytical approach to estimating

AMHS performance in 300 mm fabs,” Int. J. Prod. Res., vol. 45, no. 3,

pp. 571–590, 2007.

[24] D. Nazzal and L. F. McGinnis, “Expected response times for closed-

loop multivehicle AMHS,” IEEE Trans. Autom. Sci. Eng., vol. 4, no. 4,

pp. 533–542, Oct. 2007.

[25] D. Nazzal, “A closed queueing network approach to analyzing multi-

vehicle material handling systems,” IIE Trans., vol. 43, no. 10,

pp. 721–738, 2011.

[26] J. T. Lin, F. K. Wang, and C. J. Yang, “The performance of the number

of vehicles in a dynamic connecting transport AMHS,” Int. J. Prod. Res.,

vol. 43, no. 11, pp. 2263–2276, 2005.

[27] C.-J. Huang, K.-H. Chang, and J. T. Lin, “Optimal vehicle allocation for

an automated materials handling system using simulation optimisation,”

Int. J. Prod. Res., vol. 50, no. 20, pp. 5734–5746, 2012.

1580 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 13, NO. 4, OCTOBER 2016

[28] K.-H. Chang, Y.-H. Huang, and S.-P. Yang, “Vehicle fleet sizing for

automated material handling systems to minimize cost subject to time

constraints,” IIE Trans., vol. 46, no. 3, pp. 301–312, 2014.

[29] Ü. Bilge and J. M. A. Tanchoco, “AGV systems with multi-load carriers:

Basic issues and potential benefits,” J. Manuf. Syst., vol. 16, no. 3,

pp. 159–174, 1997.

[30] J. R. Van der Meer and R. de Koster, “Using multiple load vehicles

for internal transport with batch arrivals of loads,” in New Trends

in Distribution Logistics. Berlin, Germany: Springer-Verlag, 1999,

pp. 197–214.

[31] P.-C. Hung and F.-H. Liu, “Estimation of the fleet size for a multi-load

automated guided vehicle system,” Proc. Nat. Sci. Council, Phys. Sci.

Eng., vol. 25, no. 4, pp. 244–253, 2001.

[32] J. T. Lin, F.-K. Wang, and P.-Y. Yen, “Simulation analysis of dispatching

rules for an automated interbay material handling system in wafer fab,”

Int. J. Prod. Res., vol. 39, no. 6, pp. 1221–1238, 2001.

[33] S. C. Sarin, A. Varadarajan, and L. Wang, “A survey of dispatching rules

for operational control in wafer fabrication,” Prod. Planning Control,

Manage. Oper., vol. 22, no. 1, pp. 4–24, 2011.

[34] L. Qiu, W.-J. Hsu, S.-Y. Huang, and H. Wang, “Scheduling and routing

algorithms for AGVs: A survey,” Int. J. Prod. Res., vol. 40, no. 3,

pp. 745–760, 2002.

[35] M. Saravanan and G. S. Kumar, “Different approaches for the loop

layout problems: A review,” Int. J. Adv. Manuf. Technol., vol. 69,

nos. 9–12, pp. 2513–2529, 2013.

[36] H. Fazlollahtabar and M. Saidi-Mehrabad, “Methodologies to optimize

automated guided vehicle scheduling and routing problems: A review

study,” J. Intell. Robot. Syst., vol. 77, no. 3, pp. 525–545, 2015.

[37] D. Sinriech and J. M. A. Tanchoco, “Impact of empty vehicle flow on

performance of single-loop AGV systems,” Int. J. Prod. Res., vol. 30,

no. 10, pp. 2237–2252, 1992.

[38] A. Asef-Vaziri, G. Laporte, and R. Ortiz, “Exact and heuristic procedures

for the material handling circular flow path design problem,” Eur. J.

Oper. Res., vol. 176, no. 2, pp. 707–726, 2007.

[39] B.-I. Kim, J. Shin, and J. Chae, “Simple blocking prevention for bay type

path-based automated material handling systems,” Int. J. Adv. Manuf.

Technol., vol. 44, nos. 7–8, pp. 809–816, 2009.

[40] K. Bartlett, J. Lee, S. Ahmed, G. Nemhauser, J. Sokol, and B. Na,

“Congestion-aware dynamic routing in automated material handling

systems,” Comput. Ind. Eng., vol. 70, pp. 176–182, Apr. 2014.

[41] H. Y. K. Lau and S. O. Woo, “An agent-based dynamic routing strategy

for automated material handling systems,” Int. J. Comput. Integr. Manuf.,

vol. 21, no. 3, pp. 269–288, 2008.

[42] B. Faaland and T. Schmitt, “Cost-based scheduling of workers and

equipment in a fabrication and assembly shop,” Oper. Res., vol. 41,

no. 2, pp. 253–268, 1993.

[43] G. A. Süer, J. Cosner, and A. Patten, “Models for cell loading and

product sequencing in labor-intensive cells,” Comput. Ind. Eng., vol. 56,

no. 1, pp. 97–105, 2009.

[44] O. Guyon, P. Lemaire, É. Pinson, and D. Rivreau, “Cut generation for

an integrated employee timetabling and production scheduling problem,”

Eur. J. Oper. Res., vol. 201, no. 2, pp. 557–567, 2010.

[45] C.-F. Chien, J.-N. Zheng, and Y.-J. Lin, “Determining the operator-

machine assignment for machine interference problem and an empirical

study in semiconductor test facility,” J. Intell. Manuf., vol. 25, no. 5,

pp. 899–911, 2014.

[46] G. T. Mackulak and P. Savory, “A simulation-based experiment for

comparing AMHS performance in a semiconductor fabrication facility,”

IEEE Trans. Semicond. Manuf., vol. 14, no. 3, pp. 273–280, Aug. 2001.

[47] N. Govind, T. M. Roeder, and L. W. Schruben, “A simulation-based

closed queueing network approximation of semiconductor automated

material handling systems,” IEEE Trans. Semicond. Manuf., vol. 24,

no. 1, pp. 5–13, Feb. 2011.

[48] E. W. Dijkstra, “A note on two problems in connexion with graphs,”

Numer. Math., vol. 1, no. 1, pp. 269–271, Dec. 1959.

[49] J. T. Lin, F. K. Wang, and Y. M. Chang, “A hybrid push/pull-dispatching

rule for a photobay in a 300 mm wafer fab,” Robot. Comput.-Integr.

Manuf., vol. 22, no. 1, pp. 47–55, 2006.

[50] A. Asef-Vaziri, N. G. Hall, and R. George, “The significance of

deterministic empty vehicle trips in the design of a unidirectional loop

flow path,” Comput. Oper. Res., vol. 35, no. 5, pp. 1546–1561, 2008.

[51] W. Qin, J. Zhang, and Y. Sun, “Multiple-objective scheduling for inter-

bay AMHS by using genetic-programming-based composite dispatching

rules generator,” Comput. Ind., vol. 64, no. 6, pp. 694–707, 2013.

[52] M.-J. J. Wang, H.-C. Chung, and H.-C. Wu, “The evaluation of manual

FOUP handling in 300-mm wafer fab,” IEEE Trans. Semicond. Manuf.,

vol. 16, no. 3, pp. 551–554, Aug. 2003.

Chen-Fu Chien (M’03) received the B.S. degree

(Phi Tao Phi Hons.) in industrial engineering and

electrical engineering from National Tsing Hua Uni-

versity (NTHU), Hsinchu, Taiwan in 1990, and

the M.S. degree in industrial engineering and the

Ph.D. degree in operations research and decision

sciences from the University of Wisconsin–Madison,

Madison, WI, USA, in 1994 and 1996, respectively.

He was a Fulbright Scholar with the University

of California at Berkeley, Berkeley, CA, USA, from

2002 to 2003. He received the PCMPCL Training at

the Harvard Business School, Boston, MA, USA, in 2007. From 2005 to 2008,

he was on-leave as the Deputy Director of the Industrial Engineering Division

at Taiwan Semiconductor Manufacturing Company (TSMC), Hsinchu. He is

currently a Tsing Hua Chair Professor with NTHU. He is also the Director

of the NTHU–TSMC Center for Manufacturing Excellence and the Princi-

pal Investigator for the Semiconductor Technologies Empowerment Partners

(STEP) Consortium, Ministry of Science and Technology, Taiwan. He has

received 20 invention patents on semiconductor manufacturing and published

4 books, more than 140 journal papers, and 12 case studies with the Harvard

Business School. His current research interests include decision analysis,

big data analytics, modeling and analysis for semiconductor manufacturing,

manufacturing strategy, and manufacturing intelligence.

Dr. Chien received the National Quality Award, the Distinguished Research

Awards and Tier 1 Principal Investigator (Top 3%) from NSC, the Distin-

guished University-Industry Collaborative Research Award from the Ministry

of Education, the University Industrial Contribution Awards from the Ministry

of Economic Affairs, the Distinguished University-Industry Collaborative

Research Award, the Distinguished Young Faculty Research Award from

NTHU, the Distinguished Young Industrial Engineer Award, the Best IE

Paper Award, the IE Award from the Chinese Institute of Industrial Engi-

neering, the Best Engineering Paper Award and Distinguished Engineering

Professor by the Chinese Institute of Engineers in Taiwan, the 2011 Best

Paper Award of the IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND

ENGINEERING, and the 2015 Best Paper Award of the IEEE TRANSACTIONS

ON SEMICONDUCTOR MANUFACTURING. He is an Area Editor of the

Flexible Services and Manufacturing Journal, an Editorial Board Member

of Computers and Industrial Engineering, an Associate Editor of the Jour-

nal of Intelligent Manufacturing, and an Advisory Board Member of OR

Spectrum. He has been invited to give keynote speech in various confer-

ences, including APIEMS, C&IE, IEEM, and IML, and leading universities

worldwide.

Che-Wei Chou is currently pursuing the

Ph.D. degree with the Department of Industrial

Engineering and Engineering Management, National

Tsing Hua University, Hsinchu, Taiwan.

He has published papers in the IEEE

TRANSACTIONS ON AUTOMATION SCIENCE AND

ENGINEERING and the Journal of Management

and System. His current research interests include

production planning and scheduling, modeling

and analysis for manufacturing, intelligent

manufacturing, and decision analysis.

Hui-Chun Yu received the Ph.D. degree in statis-

tics from National Cheng Kung University, Tainan,

Taiwan, in 2011.

He was a Research Fellow with the NTHU TSMC

Center for Manufacturing Excellence, NTHU,

Hsinchu, Taiwan. He is currently a Senior Data Sci-

entist with TSMC, Hsinchu. He has published papers

in the International Journal of Computational Intel-

ligence Systems and the Journal of Intelligent Man-

ufacturing. His current research interests include

data mining, tool health, modeling and analysis for

manufacturing, and intelligent manufacturing.