On: 25 April 2014, At: 06:24 Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
International Journal of Computer Integrated
Manufacturing
Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/tcim20
The transport strategies for fully automated
manufacturing in 300 mm wafer fab
D. Y. Sha ab & C. J. Yang b a
Department of Industrial Engineering and System Management , Chung Hua University , 707, Section 2 WuFu Road, Hsinchu 300, Taiwan, R.O.C.
b
Department of Industrial Engineering and Management , National Chiao Tung University , 1001, Ta Hsueh Road, Hsinchu 300, Taiwan, R.O.C.
Published online: 24 Sep 2009.
To cite this article: D. Y. Sha & C. J. Yang (2009) The transport strategies for fully automated manufacturing in 300 mm
wafer fab, International Journal of Computer Integrated Manufacturing, 22:10, 962-975, DOI: 10.1080/09511920902942863
To link to this article: http://dx.doi.org/10.1080/09511920902942863
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
The transport strategies for fully automated manufacturing in 300 mm wafer fab
D.Y. Shaa,b* and C.J. Yangba
Department of Industrial Engineering and System Management, Chung Hua University, 707, Section 2 WuFu Road, Hsinchu 300, Taiwan, R.O.C.;bDepartment of Industrial Engineering and Management, National Chiao Tung University, 1001, Ta Hsueh Road,
Hsinchu 300, Taiwan, R.O.C.
(Received 19 March 2008; final version received 7 March 2009)
This paper addresses the operational issues of transport strategies in tool and vehicle dispatching integration (TVDI) architecture in a fully automated manufacturing wafer fab. At present, there are three transport strategies involved in vehicle dispatching, namely, avoid blocking, avoid starvation, and accelerate batch preparation. These strategies were developed to obviate production obstacles and to avoid capacity loss. Consequently, there are five levels in the decision-making process of TVDI, namely, dispatching request, conditions checking, candidate selection, dispatching rules, and result execution. Specifically, candidate selection was classified into five categories: FOUP-selects-tool (FST), FOUP-selects-stocker (FSS), tool-selects-FOUP (TSF), FOUP-selects-vehicle (FSV), and vehicle-selects-FOUP (VSF). The proposed transport strategies were implemented in VSF, and a simulation model abstracted from a wafer fab in Taiwan was used to evaluate the performance. The results show that the differences in the proposed strategies compared with ignoring the issues are statistically significant, and the performances of the wafer output, cycle time and waiting time can be improved.
Keywords: dispatching; integration; transport; AMHS; wafer fab; simulation
1. Introduction
Since increasing the technology of integrated circuit (IC) manufacturing from 0.18 mm to 0.13 mm, or even smaller than 0.1 mm, airborne molecular contaminants (AMC) in the clean room had negative effects during the process, which eventually affected the yield. For this reason, mini-environment (Brain and Abuzeid 1994) manufacturing, which emphasises maintenance of the cleanliness classification, is implemented. Aside from its economic benefits, the size of the wafer also increases from 200 mm to 300 mm. Also, the fully automated transport is implanted because of its ergonomic requirement. Fully automated manufactur-ing is consequently introduced in the 300 mm wafer fab. It brings the greatest challenge of integrating the transport and the production elements, particularly at the operational level, tool dispatching (TD) and vehicle dispatching (VD). Moreover, an automated material handling system (AMHS) for delivering the heavy 300 mm wafers without human involvement is the result of a breakthrough in the transport system (Kaempf 1997, Kurosaki et al. 1997, Bahri et al. 2001). Generally, the 300 mm AMHS is implemented as many separate loops, spreading out from a central loop, and is connected in front of each functional process bay. Both loops are located overhead to attain
zero footprints in transport and to minimise the fab footprint (see Figure 1). The wafer carrier, or the front-opening unified pod (FOUP), is a kind of closed carrier with an automated door at the front side. On the other hand, the vehicle, an overhead hoist transporter (OHT), is capable of carrying one FOUP at a time and has a hoisting mechanism that automatically loads and unloads one FOUP. Under this transport config-uration, vehicles are not restricted to one designated loop, but are allowed to travel all around the wide fab. At the same time, the FOUP can be delivered directly through stockers or tool-to-tool to their destination. Hence, a matrix of transport capabilities can also be provided (Plata 1997).
The dispatching is triggered by the production or transport request. The production request is that FOUP has to be pushed to the downstream tool when it has already been accomplished as a process step, or the tool pulls the FOUP for the next task when its capacity is released. Furthermore, the transport request is initialled by the production request for transporting the FOUP (either pushed or pulled) to the designated location. However, these requests might not be executed immediately owing to the limited resources. Thus, the dispatching is raised to check the available capability and to determine the resource allocation based on the designated rule.
*Corresponding author. Email: yjsha@chu.edu.tw Vol. 22, No. 10, October 2009, 962–975
ISSN 0951-192X print/ISSN 1362-3052 online Ó 2009 Taylor & Francis
DOI: 10.1080/09511920902942863 http://www.informaworld.com
Moreover, TD involves the determination of which FOUP should be process first, given that many FOUPs are waiting to be processed. Tyan et al. (2002) used theory of constraint (TOC) principles to propose the state-dependent dispatching rule that is especially designed for the bottleneck station. Meanwhile, Dabbas and Fowler (2003) classified the rules as local or global policies. The global policies include the look-behind and look-ahead strategies, which make a timely decision within and outside its immediate neighbour-hood because of their re-entrance characteristics. In addition, some issues are focused on particular restrictions, such as mask scheduling, tool dedication, and full-batch. The mask scheduling attempts to minimise the mask change times to reduce set-up time. For example, family-based rules, which group the same photo mask as a family, were proposed by Chern and Liu (2003) to balance the workload between two consecutive exposure operations. These rules were said to be similar to the workload-levelling algorithm proposed by Kim et al. (1998). Tool dedication, which aims to balance the workload of the selected tool, is like the evaluation of the flexible assignment policy and dedicated assignment policy by Akcalt et al. (2001), and the line balance algorithm by Wu et al. (2006), which also aims to smooth the flow rate between multiple photo layers. Furthermore, the full-batch process combines multiple lots with the same recipe for cleaning or oxidation deposition, and the related issue is intended to reduce the production variance owing to the batch collection. Weng and Leachman (1993) used the information about future arrival to develop the minimum cost rate heuristic for reducing the variation in lead times. Also, Kim et al. (1998) proposed the back and front queues levelling rule to avoid starvation of the workstation. In fact, the idea of a kanban card or the pull approach used in the time constraint between wet etch and furnace was introduced by Scholl and Domaschke (2000).
VD involves the determination of which FOUP to transport first, given that many FOUPs are waiting to be moved. The related topic has appeared earlier in an automatic guide vehicle (AGV) system, and similar issues are extended to the AMHS in fab. Egbelu and Tanchoco (1984) classified the AGV dispatching into two categories specifically the work centre-initiated and the vehicle-initiated. The work centre-initiated approach involves the work centre requesting a vehicle, from a set of available vehicles, to move out a load in its output queue. On the other hand, the vehicle-initiated dispatching approach involves the vehicle asking for the next load, from a set of loads which are waiting to be moved. Egbelu (1987) further classified the vehicle-initiated dispatching approach as source-driven and demand-source-driven. The decision criterion of
source-driven dispatching approach is focused on the conditions of the work centre where the loads originate, while the work centre conditions of the destination are considered the essential criterion in the demand-driven dispatching approach. A new classifi-cation of dispatching systems presented by Le-Anh and de Koster (2006) were decentralised and centra-lised, and the general objectives of dispatching included minimising waiting time, maximising throughput, minimising queue length, and guarantee-ing a certain service level were introduced.
For the AMHS in fab, Lin et al. (2001) outlined the dispatching system in a double loop interbay with the three decision points, namely, the loop selection, cassette-initiated rule, and vehicle-initiated rule. Mean-while, Wang and Liao (2003, 2004) developed the policies, which are classified as preemptive highest priority job first (PHP) and differentiated preemptive dispatching (DPD) to enhance the service of hot lot by reducing the frequent blockage of the normal lots transport. Lin et al. (2006) introduced a hybrid push-pull rule for a photobay, and the numbers of input/ retrieve in the stocker can be decreased. In the connecting AMHS, Lin et al. (2003a) classified the vehicle into four types to service the designated area; tool-to-tool delivery can then be accomplished with the combination of the different vehicle types. Further-more, the mixture of the different combination of vehicle types was presented by Lin et al. (2003b) and the optimum mixture percentage could be obtained through the application of the response surface methodology. Vehicle types can also be changed for various task requests according to Lin et al. (2004) to make the dispatch more flexible. Conversely, the multi-attribute metric of vehicle reassignment was proposed by Kim et al. (2007), and the empty trip distance as well as the waiting time of the load would be taken into account in determining the delivery task. Sha et al. (2008) addressed the search range assignment for dispatching, indirectly to limit the distance of a vehicle’s empty trip which will eventually make the vehicle work effectively.
In the literature, much effort has been put into enhancing the performance of TD or VD separately. However, the dynamic of traffic has been ignored in TD studies. Also, tool status like blocking or starvation has not been considered in VD studies, or has been set as a constant time delay. Researchers believe that these two issues should be discussed simultaneously in fully automated manufacturing owing to the close interac-tions between production and transport. In addition, most VDs in the existing literature are based on the status of the tasks or tools where the tasks originate. This type of dispatching is inflexible in a fully automated wafer fab, as shown in section 2. Therefore, this study
attempts to propose strategies in VD (named transport strategy) which will determine the task’s transport priority based on the status of the tool, either the tasks originate or the task’s destination, in tool and vehicle dispatching integration (TVDI) architecture. Likewise, the performance was evaluated using a simulation built in the object-oriented software eM-PlantTM.
The study is presented in the following sections. Specifically, the problem statement is addressed in section 2 while the three transport strategies are presented in section 3. Also, the interactions between TD and VD are presented in section 4 and the simulation experiment is reported on section 5. Lastly, conclusions are presented in the final section, along with suggestions for further research.
2. Problem statement
The dispatching procedure of the object fab shown in Figure 2 illustrates that the transport priority of the FOUP (defined as a process unit with 25 pieces of wafer in it) is based on the longest waiting time first, as the source-driven dispatching. It might appear that a FOUP is being moved to a tool, but in reality, many FOUPs are ready for process in the tool’s port, which means that the tool does not need more work in progress (WIP). The same circumstance occurs if the FOUP is transported to its downstream tool, however, this FOUP keeps circulating on the vehicle because the tool is blocked and there are no available ports to load it on. At the same time, other starving tools remain idle, and the tool’s
capacity might be lost because the FOUP requested by the starving tools is waiting for transport – this FOUP is not the longest waiting. Also, FOUP might not obtain the highest priority (the longest waiting time) to be moved for gathering in the stocker before the furnace process. This means that there will be a larger variance time period for batch collection. Unfortunately, these situations are caused by omitting a tool’s capability when executing VD and the vehicles transporting the wrongFOUP to the wrong tool. This also implies that there is a misallocation of vehicles.
Thus, it is necessary to identify the interactions between TD and VD, and to develop a transport strategy that will adjust the FOUP’s transport priority according to ‘special properties’ for better vehicle allocation.
3. Transport strategies
The demand-driven transport strategies, considering the tool states like blocking, starvation, and batch processing, are proposed. Hence, the functions of transport are not only to provide service to the production request as described in section 1, but also to carry out some activities that will make the production more efficient like obviating production obstacles or avoiding capacity loss.
3.1. Avoid blocking
To avoid tool blocking, there is a need to detect the states of a tool’s port. This will determine if there is
Figure 1. Configuration of a 300 mm wafer fab.
tool blocking occurrence. Specifically, there are five states of a tool’s port identified in the study. They are notated as Pa, Pr, Pp, Pn, and Po.
(1) Pa: port available for assignment (avail.-port). (2) Occupied port:
(a) Pr: the port is reserved by a FOUP on the way to this tool (res.-port).
(b) Pp: the port is occupied by a FOUP which the wafers in it are processing now (proc.-port)
(c) Pn: the port is occupied by a FOUP which is waiting for processing (in-port).
(d) Po: the port is occupied by a FOUP which is waiting to be moved (out-port).
The state of the port changes dynamically and is in only one, of the five states. nPa, nPr, nPp, nPn, and
nPoare the ports numbered Pa, Pr, Pp, Pn, and Po of a tool respectively. To determine the number of available ports, the equation is nPa¼ nP – nPr – nPp – nPn – nPo, where nP is the total number of a tool’s port and nPa2 {0,1, . . . , nP}, nPr 2 {0,1, . . . , nP}, nPp2 {0,1}, nPn 2 {0,1, . . . , nP-1}, nPo 2 {0,1, . . . , nP}. Likewise, blocking occurs if nPa¼ 0, nPo4 0, and if there is a FOUP scheduled to be loaded to this tool. The FOUP in Po is considered as a blocking FOUP.
For example, tool A has four ports in Figure 3, and FOUPs f5and f6are scheduled for loading to the tool
but are stored somewhere. Meanwhile, f2is processed
and f1 is scheduled for processing after f2. f3 have
finished the step and f4is being moved to port c by v1.
The states of ports a, b, c, and d at this time are Pn, Pp, Pr, and Po respectively. nPn, nPp, nPr, and nPo are the same, and have no available port. In this case, blocking
Figure 2. Dispatching decision-making procedure.
occurs because nPa¼ 0, nPo ¼ 1 4 0, f5 and f6 are
queued to be processed. The f3 is a blocking FOUP,
and f5 or f6 cannot be loaded in the tool until f3 is
moved out.
The avoid blocking strategy will enable a higher priority of blocking FOUP to be transported, which leads to the release of the port’s capacity. Furthermore, the tool’s blocking status is removed and the produc-tion obstacle is obviated.
3.2. Avoid starvation
The avoid starvation strategy will enable a higher transport priority to the FOUP (named starved FOUP), which is requested by the most starving tool group, the tool group which has the Max (UDi). The
following notations are used to illustrate:
(1) i: tool group where i¼ 1, 2, . . . , n. n: number of tool groups.
(2) j: tool, where j¼ 1, 2, . . . , ni. ni: number of
tools which belong to i, j2 i.
(3) UPi: planned utilisation of i, which is calculated
using the static capacity analysis under the planned wafer out per month.
(4) UAi,j: average actual utilisation of j, j2 i.
(5) UAi: average actual utilisation of i, where
UAi¼
Pni
j¼1 UAi;j= ni, j2 i.
(6) UDi: average difference utilisation, where UDi¼
UPi– UAi.
The UAi,j, UAi, and UDi are calculated every 12
hours (2 shifts/day). This strategy indicates that in order to reach the planned wafer out, the UAi has to
achieve the planned UPi. This also implies that the
tool’s move (the volume of process complete) does not meet the plan, in fact, the tool is idle more than the expected time, making many FOUPs pile up some-where. Thus, if a tool group has the highest UDi, it is
considered as the most starving tool group in the pipeline. Hence, the starved FOUP is given a higher priority using the avoid starvation strategy, which will in turn avoid the capacity loss and smooth out the production.
3.3. Accelerate batch preparation
Moreover, the production cycle of the FOUP consists of three elements, namely, processing, transporting, and waiting, where waiting time is adjustible to reductions depending on dispatching decisions. Owing to the need to fill a batch of up to six FOUPs in the stocker before the furnace process, the waiting time for the batch preparation is considered a key element in increasing the cycle time. Figure 4 shows the time for the FOUP to prepare the batch process. It explains the shortened length of time for period 1 and period 3, i.e. the time the FOUP waits for VD in the tool’s out-port, the waiting time can be reduced and batch preparation can be accelerated.
Likewise, the accelerate batch preparation strategy will enable a higher priority of FOUP (named batch FOUP), which fulfills the status in both time periods 1 and 3 to be transported. Then, not only can the variance time period to prepare the batch be decreased, but also the capacity loss owing to the tool being idle for batch preparation can be reduced.
4. Dispatching interactions 4.1. Decision-making procedures
A five-level decision-making procedure is implemented in the TVDI architecture. The first level is the dispatching request, in which the request is triggered by a production or transport event: (e1) when a FOUP has just finished one process; (e2) when a FOUP is picked up from the tool’s port by a vehicle and the capacity of the tool’s port has just been released; (e3) when FOUPs required to form a batch are available in the stocker; (e4) when a movement request from FOUP is initiated by above three events; (e5) when a vehicle unloads a FOUP at a tool’s port, and the vehicle’s capacity has just been released. The events (e1), (e2) and (e3) are for production and (e4) and (e5) are for transport.
The second level is the resource checking, in which the status and capability of the resources, either tools or vehicles, will be assessed in order to determine whether the FOUP is to be transported to the next process, to be stored in the stocker, or to be kept waiting in the tool’s port. Once the resource is checked, the third level, which is the candidate selection, is executed. For instance, if a vehicle is determined to transport one FOUP through resource checking, then
Figure 3. States of a tool’s port.
FOUPs which have ‘special property’ will be selected as candidates for transport in this level.
Accordingly, the authors further classified the candidate selection into five categories, as the FOUP-selects-tool (FST), the FOUP-selects-stocker (FSS), the tool-selects-FOUP (TSF), the FOUP-selects-vehi-cle (FSV), and the vehiFOUP-selects-vehi-cle-selects-FOUP (VSF). FST, FSS, and TSF belong to TD, while FSV and VSF are parts of VD.
(1) FST deals with the selection of a specific tool from a set of available tools to process a FOUP’s next step.
(2) FSS deals with the selection of a stocker for temporarily storage due to FOUP’s next step tool which is blocking, or the selection of an appropriate stocker for batch collection. (3) TSF deals with the selection of a specific FOUP
from a set of waiting FOUP as a tool’s next task.
(4) FSV deals with the selection of a vehicle from a set of available vehicles to transport a FOUP which requests to move.
(5) VSF deals with the selection of a FOUP from a set of waiting FOUP as a vehicle’s next delivery task when the vehicle just completed a task. The VSF operation is the focal point in this study. The blocking FOUP, starved FOUP, and batch FOUP, which have ‘special properties’ will be selected as candidates first. This means that they have the higher transport priority.
The fourth level is the dispatching rules, in which a tool or a vehicle determines the next task from the candidate, or a FOUP requests a tool or vehicle from the candidate based on the defined rules. The candidate on this level is obtained from the previous level. Finally, the fifth level is the result execution, in which
the transport, production or storage will be executed after the four previous levels have been accomplished. The detailed interactions between TD and VD are elaborated in Figure 5, and in which the third and fourth levels are listed in Table 1.
4.2. Interactions remarks
Some dispatching interactions remarks are as follows: (1) (e1) and (e3) are the events that push the FOUP into the next process, while (e2) implies that a tool asks to pull the next task and (e5) indicates that a vehicle requests the next task.
(2) FSV is triggered after a series of production events (e1), (e2) or (e3) that request to transport the specific FOUP to the assigned location. (3) The blocking status of the current and
down-stream tool will be detected when (e1) occurs. (4) The available batch in (e2) means FOUPs have
formed the batch in the stocker and are ready for a tool to call for processing (TSF).
(5) The interactions between the candidate selection:
(5.1) a* following (e1) and c* following (e3) indicate that a FOUP has failed to be pushed to the next process (FST), and waits for a downstream tool to trigger TSF to pull (waits (e2) occurs).
(5.2) b* following (e2) indicates that a tool has failed to pull the next task (TSF), and waits for FOUPs to trigger FST to assign (waits (e1) occurs).
(5.3) d* following (e4) from (e1), (e2) and (e3) indicates that a FOUP has failed to call for a vehicle (FSV) and is waiting for a vehicle to trigger VSF to assign (waits (e5) occurs).
Figure 4. Time composing of batch process preparation.
(5.4) e* following (e5) indicates that a vehicle has failed to request the next task (VSF), and is idle until a FOUP triggers FSV to assign (waits (e4) occurs).
(6) Avoid blocking promotes the release of a port’s capacity while avoid starvation enables the starving tool to be fed as soon as possible. (7) Accelerate batch preparation attempts to reduce
the time required by FOUPs to collect the batch in the stocker, and then shortening the time period to (e3) occurs.
5. Simulation experiment 5.1. Capacity facilitated
The allocation of a production area and the track design of the material handling of a real fab in Taiwan are
abstracted in Figure 6. The capacity plan is based on static capacity analysis with one process flow, 0.13mm logic IC. The capacity plan also assumes that there are 6,000 pieces of wafer output per month, and keeps the tool utilisation below 90%. Accordingly, the 736 steps (without manual inspection) with 33 photo layers requires a net time of 352.41 h (approximately 14.7 days) to process (For the process flow information, please see Appendix 1). The 72 tool groups and 141 tools required for the above processing are facilitated and listed in Table 2.
The AMHS includes one interbay and ten intra-bays, and the types of track such as intrabay U-turn track, shortcut and bypass are facilitated. The U-turn track is located within the bay handling, while the shortcut track is located for a short transport distance. The bypass is installed for a vehicle to travel straight
Figure 5. Representation of dispatching interactions.
Table 1. Candidate selectionand dispatching rules in decision-making process. FST Candidate selection
p1 Identify the candidate tool set from all downstream tool j where tool’s nPa 4 0
p2 Identify the candidate tool set from all downstream tool j where tool’s nPa 4¼ batch size Rule: the lowest utilization first (LU)
FSS Candidate selection
p3 Identify the candidate STK set Y from all STKs which correspond to tool group i p4 Identify the set Y from all STKs which correspond to tool group i
Identify the set y2 Y where exists non-available batch (incomplete batch) If y6¼ Ø, candidate STK set is y. Else (y ¼ Ø), candidate STK set is Y Rule: the lowest WIP level first (LWL)
TSF Candidate selection
p5 Identify the candidate FOUP set from all FOUPs which are non-assigned to tool
p6 Identify the candidate FOUPs set from all FOUPs which are available batch and non-assigned to tool Rule: the first come first service (FCFS)
FSV Candidate selection
p7 Identify the candidate vehicle set from all vehicles which are idle Rule: the nearest vehicle first (NV)
VSF Candidate selection
p8 Identify the following sets:
(1) set W from FOUPs which are non-assigned to vehicle and queuing for transport; (2) set b2 W from FOUPs which are blocking FOUP
(3) set s2 W from FOUPs which are starvation FOUP (4) set f2 W from FOUPs which are batch FOUP
(5) set x1¼ {FOUP j (b \ s)}, set x2¼ {FOUP j (b \ f)}, set x3¼ {FOUP j (s \ f)},
set x4¼ {FOUP j (b [ s)}, set x5¼ {FOUP j (b [ f)}, set x6¼ {FOUP j (s [ f)},
set x7¼ {FOUP j (b \ s \ f)}, set x8¼ {FOUP j b [ s [ f)}
set x9¼ {FOUP j (b \ s) [ (b \ f) [ (s \ f)}
Then the candidate FOUP set is identified under different scenarios (see Section 5.5). Ø Scenario 1: A1B1C1
b¼ Ø s¼ Ø f¼ Ø set W
f6¼ Ø set f
s6¼ Ø f¼ Ø set s
f6¼ Ø If x36¼ Ø, set x3. Else (x3¼ Ø), set x6
b6¼ Ø s¼ Ø f¼ Ø set b
f6¼ Ø If x26¼ Ø, set x2. Else (x2¼ Ø), set x5
s6¼ Ø f¼ Ø If x16¼ Ø, set x1. Else (x1¼ Ø), set x4
f6¼ Ø If x76¼ Ø, set x7. Else if x16¼ Ø or x26¼ Ø or x36¼ Ø, set x9.
Else (x7¼ Ø & x9¼ Ø), set x8.
ØScenario 2: A1B1C2
b¼ Ø s¼ Ø set W
s6¼ Ø set s
b6¼ Ø s¼ Ø set b
s6¼ Ø If x16¼ Ø, set x1. Else (x1¼ Ø), set x4
ØScenario 3: A1B2C1
b¼ Ø f¼ Ø set W
f6¼ Ø set f
b6¼ Ø f¼ Ø set b
f6¼ Ø If x26¼ Ø, set x2. Else (x2¼ Ø), set x5
ØScenario 4: A1B2C2 b¼ Ø set W b6¼ Ø set b ØScenario 5: A2B1C1 s¼ Ø f¼ Ø set W f6¼ Ø set f s6¼ Ø f¼ Ø set s
f6¼ Ø If x36¼ Ø, set x3. Else (x3¼ Ø), set x6
ØScenario 6: A2B1C2 s¼ Ø set W s6¼ Ø set s ØScenario 7: A2B2C1 f¼ Ø set W f6¼ Ø set f ØScenario 8: A2B2C2 set W Rule:the longest waiting time first (LWT)
p1 * p8: path marks from Figure 5.
forward without the obstruction of another vehicle in front, which is assumed to be loading/unloading with a stocker. Thus, a vehicle travels and turns to the
direction of the stocker’s stopping point if it needs to input/retrieve a FOUP from the stocker. This track reduces traffic congestion, which might be caused by
Table 2. Capacity facilitated.
Production Area
TG number Tool number Critical TG
Prod. Insp. No. Prod. Insp. No. No. Util.%
WS 1 0 1 1 0 1 0 TF 15 1 16 30 1 31 1 84.01 ET 14 5 19 34 6 40 5 83.98 DF 9 2 11 13 2 15 1 82.12 PH 6 6 12 15 13 28 2 84.46 IMP 5 0 5 14 0 14 2 86.86 CMP 2 0 2 2 0 2 0 CU 2 2 4 6 2 8 0 QC 0 2 2 0 2 2 0 Total 54 18 72 115 26 141 11 84.29
TG: tool group; Prod.: production tools; Insp.: Inspection tools. Figure 6. Representative layout of a 300 mm wafer fab.
the delay of executing the loading/unloading process along the traffic arteries (interbay).
5.2. System behaviour
The following system behaviour is described: (1) only one product described above has been implemented due to process flow confidentiality; (2) the uniform loading (UL) (Glassey and Resende 1988) open-loop wafer release policy was adopted (200 pieces/day); (3) batch size of the furnace processing is six units, and the FOUP has to form the batch in stocker; (4) furnace tools are embedded with internal storage (twelve units) for batching; (5) four ports of furnace and photo tools, and three ports for other tools; (6) the direction of a port is bi-directional; (7) FOUP’s movement request from tool is sequential. That is, the next movement request from a tool can be initiated only when the present FOUP is moved out; (8) the traveling path is based on the shortest
distance; (9) the zone control (Garry 1987) is used to prevent traffic collision.
5.3. Model assumptions
The following assumptions are made: (1) process times are constant, with no set-up time, no reworking and no yield loss; (2) breakdown of tool and stocker are not considered; (3) acceleration and deceleration of the vehicle are ignored; (4) breakdown and battery recharge of the vehicle are not considered.
5.4. Performance indices
The performance indices are: (1) wafer out (WO, pieces/ month) – an average number of wafers output per month; (2) cycle time (CT, hrs) – the average time for the wafers to enter and then leave the system; (3) batch process prepared time (BPT, hrs) – the average time for wafers to prepare batch process. That is, the time from ‘start’ to ‘complete’
Table 3. AVOVA for the transport strategies experiment.
Source Sum of squares DF Mean square F-value Prob 4 F
(1) Response: TP Model 96764.3 7 13823.5 46.2 50.0001* significant A 4069.0 1 4069.0 13.6 0.0020* B 1881.5 1 1881.5 6.3 0.0234* C 2350.3 1 2350.3 7.8 0.0128* AB 49277.3 1 49277.3 164.5 50.0001* AC 1100.3 1 1100.3 3.7 0.0733 BC 16933.6 1 16933.6 56.5 50.0001* ABC 21152.3 1 21152.3 70.6 50.0001*
Pure error 4791.7 16 299.5 R-squared:0.9528
Cor total 101556.0 23 Adj R-squared:0.9322
(2) Response: CT Model 1534.1 7 219.2 41.6 50.0001* significant A 218.9 1 218.9 41.5 50.0001* B 28.1 1 28.1 5.3 0.0347* C 227.8 1 227.8 43.2 50.0001* AB 444.4 1 444.4 84.3 50.0001* AC 393.1 1 393.1 74.6 50.0001* BC 32.1 1 32.1 6.1 0.0252* ABC 189.6 1 189.6 36.0 50.0001*
Pure error 84.3 16 5.3 R-squared:0.9479
Cor total 1618.4 23 Adj R-squared:0.9251
(3) Response: BPT Model 3211.1 7 458.7 8128.8 50.0001* significant A 1283.0 1 1283.0 22734.8 50.0001* B 469.2 1 469.2 8314.0 50.0001* C 716.0 1 716.0 12687.1 50.0001* AB 362.3 1 362.3 6420.2 50.0001* AC 245.9 1 245.9 4358.2 50.0001* BC 69.7 1 69.7 1234.8 50.0001* ABC 65.0 1 65.0 1152.4 50.0001*
Pure error 0.9 16 0.1 R-Squared:0.9997
Cor total 3212.0 23 Adj R-Squared:0.9996
*significant at 95% confidence level.
the batch preparation in Figure 4. A total of 19 steps BPT among process flow (736 steps) will be summarised.
5.5. Design of experiment
Three strategies implemented during the candidate selection, VSF, were evaluated. The factors and levels were as follows:
(1) Factor A: blocking; two levels. Levels are A1¼ avoid blocking, A2¼ ignore blocking.
(2) Factor B: starvation; two levels. Levels are B1¼ avoid starvation, B2¼ ignore starvation.
(3) Factor C: batch preparation; two levels. Levels are C1¼ accelerate batch preparation, C2¼
ignore batch preparation.
A three-factor full-factorial with 23 designs was used. The number of scenarios is 2 (A1, A2) 6 2 (B1,
B2) 6 2 (C1, C2)¼ 8, in which A2B2C2means that the
FOUP is dispatched only through the traditional VD rule (level 4). In addition, the replication is set at 3 in determining the sum of squares due to error if the model includes all possible interactions (Montgomery 2001). Hence, the number of experiments performed is 24 [8 (scenarios) 6 3 (replications) ¼ 24]. Also, the simulation run is determined from the simulation time of four months at 24 hours a day, while the warm-up is set at two months, which is determined by a pre-simulation in which the stable trend of WIP can be obtained after two months.
5.6. Simulation results
The residual analysis of the indices measured by simulation satisfied the model assumptions (normality, independence of error term, and constant variance). The ANOVA analysis summarised in Table 3 indicates that A, B, C, and their interactions significantly affect the WO, CT, and BPT at the 95% confidence level, and stresses that addressing blocking, starvation, and batch preparation are critical to performance.
Further, it is necessary to examine any important interaction (Montgomery 2001), as well as the graphs of the highest-order significant interaction (ABC) to the indices. These data are shown in Figure 7. The slope in Figure 7(a) indicates that C2has little effect at A and B,
but C1has a large effect at A and B. This also implies that
C1 is important to WO. The better WO would be
obtained when A, B, and C are at A1, B1, and C1.
Analogical discussion points out that the best CT and BPT would be obtained when A, B, and C are at A1, B1,
and C1 respectively (Figures 7(b), 7(c)). In addition,
Figure 7(c) states that C1keeps the shorter BPT at both
A and B, and this result proves the idea in section 3.3.
In addition, the least significant difference (LSD) method was used for the post-hoc multiple comparisons to compare all pairs of the eight scenarios, as summarised in Table 4. The results show that the differences from the proposed strategies compared with ignoring the issues are statistically significant, and performances of A1B1C1
out-perform others. Besides, the percentage of BPT in the total process prepared time (includes serial and batch process) (BPT%) was also summarised. A large propor-tion of BPT, 49.5% * 65.9% indicates that it is necessary to reduce BPT by dispatching, and through this, will the adverse effects to the downstream of batch process be minimised.
Furthermore, a multiple-response method called desirability (Myers and Montgomery 1995) was used to integrate multiple indices into one. The method makes use of an objective function D(X), called the desirability function D ¼ dð 1 d2 . . . dnÞ 1 n¼ Y n i¼1 di !1 n
Figure 7. Graphs of the significant interaction.
where n is the number of responses, D(X) reflects the desirable range for each response (di), 0 ¼< di¼< 1, a
geometric mean of all transformed responses. The desirability of each scenario is shown at the right side of Table 4. We can see that A1B1C1is the optimal transport
strategy in response to the previous interactions analysis. The following points have to be emphasised: (1) Consideration of the tool status such as
blocking, starvation, and batch process in vehicle-initiated dispatching (VSF) is required in order to smooth out production.
(2) Avoid blocking (A1) and avoid starvation (B1)
are simultaneously required because the serious status that a tool’s starvation caused by its blockinghas to be obviated. Also, the accelerate batch preparation(C1) should be combined with
A1and B1to reduce the variances in cycle time,
and to improve wafers output.
6. Conclusion and further research
This paper has addressed the issue of transport strategies in a 300 mm wafer fab with fully-automated manufacturing and material handling. Three transport strategies involved in VD, namely, the avoid blocking, avoid starvation, and accelerate batch preparation, were developed and implemented in the tool and the vehicle dispatching integration (TVDI) architecture. Accord-ingly, there are five levels of decision-making in TVDI, namely, dispatching request, resource checking, candi-date selection, dispatching rules, and result execution. Particularly, candidate selection includes FST, FSS, and TSF which belong to TD, and FSV and VSF which belong to VD.
A simulation model abstracted from a fab in Taiwan and a three-factor full-factorial with 23designs were used to evaluate the transport strategies. The results show that the factors A: blocking, B: starvation, and C: batch preparation significantly affect the
performance of WO, CT, and BPT. Interaction analysis, LSD method, and desirability confirm that the combination of A1: avoid blocking, B1: avoid
starvation, and C1: accelerate batch preparation
(A1B1C1) has the best performance. The results also
prove that the function of transport is not only to provide service to production request but also fully to support production like obviating production obsta-cles and avoiding capacity loss.
Therefore, the topic does not only involve the integration of two dispatching issues that are relatively independent today, but also further provides the solution for practitioners involved in dispatching software development. For practical implementation, MES could maintain a list of prioritised moves and releasethe most important move to the AMHS upon request, and the priorities could be continuously updated by MES based on the changes in production status that the authors proposed. After determining the appropriate transport strategies, the other opera-tions in candidate selection can be further evaluated. Acknowledgements
The authors gratefully acknowledge the TSMC Corporation in Taiwan for providing the opportunity for this study, and the support from National Science (NSC) project: NSC 95-2221-E-468 -013.
References
Akcalt, E., Nemoto, K., and Uzsoy, R., 2001. Cycle-time improvements for photolithography process in semicon-ductor manufacturing. IEEE Transactions Semiconsemicon-ductor Manufacturing, 14, 48–56.
Bahri, N., Reiss, J., and Doherty, B., 2001. A comparison of unified vs. segregated automated material handling systems for 300 mm fabs. In: Proceedings of the Interna-tional Symposium on Semiconductor Manufacturing Con-ference, IEEE3–6.
Brain, M. and Abuzeid, S., 1994. Minienvironment techno-logy and automation systems for next generation IC fabs. International Symposium on Semiconductor Manufactur-ing, 173–178.
Table 4. LSD and desirability for the transport strategies experiment.
Order
Scenario WO (pieces) CT (h) BPT (h)
Desirability
A B C mean rank mean rank mean (BPT%) rank
1 A1 B1 C1 6004.2 A 704.4 A 51.3 (50.7%) A 0.9275 7 A2 B2 C1 5945.8 B 710.9 B 55.3 (53.3%) C 0.7332 8 A2 B2 C2 5933.3 B 712.3 BC 59.8 (55.7%) E 0.6621 6 A2 B1 C2 5866.7 C 715.4 C 58.8 (53.5%) D 0.5247 5 A2 B1 C1 5866.7 C 720.6 D 54.2 (49.5%) B 0.4808 2 A1 B1 C2 5858.3 C 726.6 EF 75.3 (61.5%) G 0.2373 3 A1 B2 C1 5783.3 D 723.1 DE 74.6 (62.4%) F 0.1385 4 A1 B2 C2 5862.5 C 729.4 F 85.3 (65.9%) H 0.0476
A: Blocking, B: Starvation, C: Batch form; mean: average value measured from the replications; rank: different alphabet means the effects were significant at the 95% confidence level; BPT%: percentage of BPT in the total time of process prepares.
Chern, C.C. and Liu, Y.L., 2003. Family-based scheduling rules of a sequence-dependent wafer fabrication system. IEEE T. Semiconductor Manufacturing, 16, 15–25. Dabbas, R.M. and Fowler, J.W., 2003. A new scheduling
approach using combined dispatching criteria in wafer fabs. IEEE Transactions Semiconductor Manufacturing, 16, 501–510.
Egbelu, P.J., 1987. Pull versus push strategy for automated guided vehicle load movement in a batch manufacturing system. Journal of Manufacturing Systems, 6, 271–280. Egbelu, P.J. and Tanchoco, J.M.A., 1984. Characterization
of automatic guided vehicle dispatching rules. Interna-tional Journal of Production Research, 22, 359–374. EM-Plant, 2001. Objects Manual. Stuttgart: Tecnomatix
Technologies.
Garry, A.K., 1987. Automatic guided vehicle systems: application, controls and planning. Material Flow, 4, 3–16.
Glassey, C.R. and Resende, M.G.C., 1988. Closed-loop job release control for VLSI circuit manufacturing. IEEE Transactions Semiconductor Manufacturing, 1, 36–46. Kaempf, U., 1997. Automated wafer transport in the wafer
Fab. In: Advanced Semiconductor Manufacturing Con-ference and Workshop, IEEE356–361.
Kim, Y.D., Lee, D.H., Kim, J.U., and Roh, H.K., 1998. A simulation study on lot release control, mask scheduling, and batch scheduling in semiconductor wafer fabrica-tion facilities. Journal of Manufacturing Systems, 17, 107– 117.
Kim, B.I., Oh, S., Shin, J., Jung, M., Chae, J., and Lee, S., 2007. Effectiveness of vehicle reassignment in a large-scale overhead hoist transport system. International Journal of Production Research, 45, 789–802.
Kurosaki, R., Nagao, N., Komada, H., Watanabe, Y., and Yano, H., 1997. AMHS for 300 mm wafer. In: Proceed-ings of the International Symposium on Semiconductor Manufacturing Conference, D13–D16.
Le-Anh, T. and De Koster, M.B.M., 2006. A review of design and control of automated guided vehicle systems. European Journal of Operational Research, 171, 1–23. Lin, J.T., Wang, F.K., and Chang, Y.M., 2006. A hybrid
push/pull-dispatching rule for a photobay in a 300mm wafer fab. Robotic CIM-International Manufacture, 22, 47–55.
Lin, J.T., Wang, F.K., and Wu, C.K., 2003a. Connecting transport AMHS in a wafer fab. International Journal of Production Research, 41, 529–544.
Lin, J.T., Wang, F.K., and Wu, C.K., 2003b. Simulation analysis of the connecting transport AMHS in a wafer fab. IEEE Transactions on Semiconductor Manufacturing, 16, 555–564.
Lin, J.T., Wang, F.K., and Yen, P.Y., 2001. Simulation analysis of dispatching rules for an automated interbay material handling system in wafer fab. International Journal of Production Research, 39, 1221–1238.
Lin, J.T., Wang, F.K., and Young, J.R., 2004. Virtual vehicle in the connecting transport automated material-handling system (AMHS). International Journal of Production Research, 42, 2599–2610.
Montgomery, D.C., 2001. Design and analysis of experiments, 2001. New York: Wiley.
Myers, R.H. and Montgomery, D.C., 1995. Response Surface Methodology. New York: Wiley.
Plata, J.J., 1997. 300 mm fab design- a total factory perspective. In: IEEE International Symposium on Semi-conductor Manufacturing Conference Proceedings, A5–8. Scholl Wolfgang and Domaschke Joerg, 2000.
Implementa-tion of modeling and simulaImplementa-tion in semiconductor wafer fabrication with time constrains between wet etch and furnace operations. IEEE Transactions on Semiconductor Manufacturing, 13, 273–277.
Sha, D.Y., Lin, J.T., and Yang, C.J., 2008. The evaluation of search range assignment in 300 mm automated material handling system (AMHS). International Journal of Advanced Manufacturing Technology, 35, 697–710. Stat Ease, 2001. Design Expert User’s Manual. Minneapolis:
StatEast.
Tyan, Jonah, C., Chen, James, C., and Wang, F.K., 2002. Development of a state-dependent dispatching rule using theory of constraints in near-real-word wafer fabrication. Production and Planning Control, 13, 253–261.
Wang, C.N. and Liao, D.Y., 2003. Effective OHT dispatch-ing for differentiated material handldispatch-ing services in 300mm wafer foundry. In: Proceedings of the 2003 IEEE International Conference on Robotics and Automa-tion, 1027–1032.
Wang, C.N. and Liao, D.Y., 2004. Differentiated automatic materials handling services in 300mm semiconductor foundry. In: Conference 2004 of Operations Research Society in Taiwan, 105–113.
Weng, W. Willie and Leachman, Robert C., 1993. An improved methodology for real-time production deci-sions at batch-process work stations. IEEE Transactions on Semiconductor Manufacturing, 6, 219–225.
Wu, M.C., Hung, Y.L., Chang, Y.C., and Yang, K.F., 2006. Dispatching for semiconductor fabs with machine-dedication features. International Journal of Advanced Manufacturing Technology, 28, 978–984.
Appendix 1. Process flow information.
Production functions (area) Layer s(t): step number (process time)
WS TF ET DF DF* PH IMP CMP CU QC Sub. 1 1(0.08) 7(5.45) 13(5.60) 3(1.12) 4(25.34) 14(2.83) 1(0.37) 43(40.78) 2 2(1.25) 10(4.80) 1(6.00) 10(2.00) 1(0.54) 1(1.03) 25(15.63) 3 2(1.10) 8(2.15) 5(2.13) 15(5.37) 4 2(1.10) 7(2.05) 5(1.99) 14(5.14) 5 2(1.10) 7(1.77) 5(1.98) 14(4.84) 6 1(0.94) 4(2.2) 2(1.85) 1(5.43) 7(2.05) 6(2.34) 21(14.80) 7 2(1.44) 3(1.20) 1(0.93) 2(11.02) 7(2.31) 15(16.89) 8 1(0.27) 8(5.15) 1(0.93) 13(2.13) 2(0.91) 25(9.38) 9 2(1.10) 7(1.21) 4(2.54) 13(4.85) 10 2(1.10) 4(1.33) 4(2.54) 10(4.97) 11 2(1.10) 7(1.49) 4(2.44) 13(5.03) 12 2(1.10) 4(1.33) 4(2.44) 10(4.87) 13 2(1.10) 7(1.21) 3(1.91) 12(4.22) 14 1(0.27) 8(4.55) 1(0.93) 3(16.33) 11(2.26) 2(1.78) 26(26.10) 15 3(2.10) 7(1.21) 5(4.61) 15(7.91) 16 1(0.50) 3(1.20) 1(0.93) 8(1.54) 4(3.37) 17(7.54) 17 7(4.88) 6(2.45) 12(2.84) 1(0.54) 1(1.03) 27(11.75) 18 11(4.95) 3(2.49) 14(1.67) 1(1.51) 1(0.10) 30(10.72) 19 11(5.00) 3(1.28) 1(4.22) 19(1.95) 4(2.21) 38(14.66) 20 1(0.29) 6(2.91) 10(1.52) 17(4.72) 21 11(5.27) 7(2.60) 1(4.22) 20(2.00) 4(2.21) 43(16.31) 22 1(0.29) 6(2.91) 10(1.52) 17(4.72) 23 11(5.27) 7(2.60) 1(4.22) 20(2.00) 4(2.21) 43(16.31) 24 1(0.29) 6(2.91) 10(1.52) 17(4.72) 25 11(5.00) 7(2.60) 1(4.22) 20(2.00) 4(2.21) 43(16.04) 26 1(0.29) 6(2.91) 10(1.52) 17(4.72) 27 11(5.24) 7(2.60) 1(4.22) 20(2.00) 4(2.21) 43(16.28) 28 1(0.29) 6(2.55) 10(1.52) 17(4.36) 29 12(5.86) 8(3.09) 1(4.22) 17(1.84) 4(2.21) 42(17.22) 30 4(2.16) 8(2.30) 1(4.22) 8(1.21) 21(9.90) 31 5(3.97) 4(1.99) 9(5.96) 32 1(1.07) 9(3.88) 2(1.18) 12(6.12) 33 2(0.57) 2(0.49) 1(5.50) 5(2.87) 2(0.17) 12(9.60) Sub. 1 112 170 9 19 339 56 3 25 2 736 (0.08) (56.81) (80.15) (6.67) (99.17) (59.99) (32.43) (3.57) (13.38) (0.17) (352.41) % 0.14 15.22 23.1 1.22 2.58 46.06 7.61 0.41 3.4 0.27 100 (0.02) (16.12) (22.74) (1.89) (28.14) (17.02) (9.2) (1.01) (3.8) (0.05) (100)
s: number of process step of each layer under different production functions; (t): total net process time (by hour) of each layer under different production functions; DF: serial step in diffusion process; DF*: batch step in diffusion process.
