Beruflich Dokumente
Kultur Dokumente
com
Expert Systems
with Applications
Expert Systems with Applications 34 (2008) 2210–2220
www.elsevier.com/locate/eswa
a
Department of Information Management, Yuanpei University, 306 Yuanpei Street, Hsin Chu 30015, Taiwan, ROC
b
Department of Finance, Yuanpei University, 306 Yuanpei Street, Hsin Chu 30015, Taiwan, ROC
c
Graduate Institute of Business and Management, Yuanpei University, 306 Yuanpei Street, Hsin Chu 30015, Taiwan, ROC
Abstract
Silicon wafer slicing is an increasingly complex manufacturing process. This involves high purity levels, crystallographic perfection
and precise mechanical tolerances, thus 12 in. wafer slicing is the most difficult in terms of semiconductor manufacturing yield. As silicon
wafer slicing directly impacts production costs, semiconductor manufacturers are especially concerned with increasing and maintaining
the yield, as well as identifying why yields decline. The criteria for establishing the proposed algorithm are derived from literature review
and modified Delphi method in semiconductor manufacturing. The main objective of this paper is to propose a new approach within the
AHP framework for tackling the uncertainty and imprecision of silicon wafer slicing evaluations during manufacturing process stages,
where the decision-maker’s comparison judgments are represented as fuzzy triangular numbers. Additionally, the proposed algorithm
can select the evaluation outcomes to identify the worst machine of precision. Finally, results of EWMA control chart demonstrate
the feasibility of the proposed fuzzy AHP-based algorithm in effectively selecting the evaluation outcomes and evaluating the precision
of the worst performing machines. So, through collect data (the quality and quantity) to judge the result by fuzzy AHP, it will the key to
help the engineer can find out the manufacturing process yield quickly effectively.
Published by Elsevier Ltd.
Keywords: Silicon wafer slicing; Modified Delphi method; Fuzzy analytical hierarchy process; EWMA control chart
increasing and maintaining wafer yield, as well as under- thinness and brittleness (Lin, Chang, & Chen, 2005). More-
standing factors contributing to declining yields are of pri- over, slicing is a cutting procedure that has difficulty in
ority concern among semiconductor manufacturers. yielding the required precision. Wafer slicing depends on
Previous studies on product quality in semiconductor variables of machine-related, human-related, manage-
manufacturing have largely adopted statistical methods ment-related and measurement accuracy-related factors to
to examine either wafer yield or how process engineers ensure quality of manufacturing operations, errors in which
select the process parameters of wafer yield based on their would destabilize the slicing process (Lin, Chang, & Chen,
subjective experiences. Those results are then analyzed 2004).
using statistical or experimental design methods. However, This study presents an evaluation decision model that
semiconductor manufacturing includes up to thousands of assesses the yield quality of 12 in. wafer slicing in semicon-
process parameters that influence each other, making it ductor manufacturing. A literature review is performed,
extremely difficult to determine those factors that influence along with the modified Delphi method and a novel
them. Cunningham, Spanos, and Voros (1995) indicated AHP-based method for resolving the uncertainty and
that, although conventional statistical and experimental imprecision of silicon wafer slicing evaluations during
design methods have enhanced wafer yield, statistical manufacturing process stages, where the decision-maker’s
methods have many limitations with respect to complex comparison judgments are represented as fuzzy triangular
mutual influence and the non-linear problem. Additionally, numbers. In order to deal with the uncertainty and vague-
Braha and Shmilovici (2002) found that under a large num- ness from subjective perception and experience of one’s in
ber of parameters used in semiconductor manufacturing, decision process, we propose a revised fuzzy AHP based
statistical methods couldn’t analyze useful decision infor- on linguistic variable weight method. We utilize the sym-
mation efficiently. metric triangular fuzzy numbers to obtain the fuzzy judg-
Silicon wafer manufacturing processes include crystal ment matrices via pair-wise comparison. And then the
growing, pulling, slicing, lapping, etching, polishing, and fuzzy interval arithmetic and confidence index a with
cleaning. Lin, Chen, and Chang (2002) proposed this mea- interval mean approach are used to estimate the fuzzy
surement items of process quality and the parameters of eigenvectors. Also, aggregate decision-makers’ judgment
quality control are: (1) lack of precision in measuring thick- matrices to obtain the global weights ranking of relative
ness (THK) (ASTM F657, 1995), bow (ASTM F534, 1995), importance and these results can provide location selection
warp (Takeshi, 1998), total thickness variation (TTV) of insights. In fact, many real-world decision problems
(Takeshi, 1998), center thickness (CTRTHK) (Takeshi, involve multiple criteria in qualitative domains. As
1998), caused by unstable motion of the wire knife and expected, such problems will be increasingly modeled as
scrape mark during slicing process and (2) quality control multiple criteria decision-making problems, which involve
parameters, such as, electricity, resistivity and oxidation scoring on subjective/qualitative domains. This results in a
that relate to crystal pulling are not discussed in this paper. class of significant problems for which an evaluation
In polishing, quality characteristics in item (1) are incorpo- framework, which handles occurrences of seeming intran-
rated to enhance the polished wafer by using precision pol- sitivity and inconsistency, will be required. Another inter-
ishing machines. Silicon wafer slicing manufacturing esting issue of group decision-making analysis is how to
process exhibits several characteristics. They are: (1) the deal with disagreements between two or more different
product type is small batch production, (2) saw cutting rankings within an alternative set. These phenomena are
must be very precise, (3) the process run time is long, and likely to appear in qualitative/subjective domains where
(4) inspecting samples is difficult. Furthermore, the process the decision-making environment is ambiguous and vague.
involves several synchronously occurred multiple quality Therefore, this study proposes a fuzzy AHP model of
characteristics, such as thickness (THK), bow and warp, silicon wafer manufacturing processes that is sufficiently
which must be closely monitored and controlled. However, robust to permit conflict and imprecision. Numerical
the manufacturing cost of polishing is significantly higher examples demonstrate the effectiveness and applicability
than that of slicing. Hence, quality control focuses largely of the proposed models in deriving the most promising
on slicing. While adopting the EWMA control chart, Lin priority vector from a fuzzy AHP problem within a group
et al. (2002) verified that analysis bow is the worst quality decision-making environment. Additionally, the proposed
characteristic in wafers. From the perspective of quality algorithm can select the evaluation outcomes to identify
feature, the manufacturing wafer yield of silicon wafer slic- the worst machine of precision. Finally, the exponen-
ing that bow is of priority concern. tial weighted moving average (EWMA) control chart
The complex process and high variation in wafer fabri- demonstrates the effectiveness of the proposed fuzzy
cation make its production management very difficult AHP method in selecting the evaluation outcomes and
(Pai, Lee, & Su, 2004). Yield of the silicon wafer slicing is evaluating the precision of the worst performing machines.
the most difficult to control. Silicon wafer slicing is a com- The proposed evaluation decision model significantly con-
plex manufacturing process, complicating efforts to domi- tributes to efforts in silicon wafer slicing to establish a
nate process stability and quality control effectively. A standard operational procedure for ensuring quality yield
wafer can be easily broken during inspection owing to its in the semiconductor industry.
2212 C.-W. Chang et al. / Expert Systems with Applications 34 (2008) 2210–2220
more important’’, 7 represents ‘‘demonstrably more impor- solidate fragmented expert opinions. The triangular fuzzy
tant’’ and 9 denotes ‘‘absolutely more important’’. An n- numbers ~uij are established as follows:
by-n matrix A as follows:
~uij¼ ðLij ; M ij ; U ij Þ; Lij 6 M ij 6 U ij and
Lij ; M ij ; U ij 2 ½1=9; 1 [ ½1; 9; ð2Þ
Lij ¼ minðBijk Þ; ð3Þ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n
Yn
ð1Þ M ij ¼ Bijk ; ð4Þ
k¼1
U ij ¼ maxðBijk Þ; ð5Þ
where Laij ¼ ðM ij Lij Þ a þ Lij , represents the left-end va- where w denotes the eigenvector of (Aa)k, 0 6 k 6 1,
lue of a-cut for aij, U aij ¼ U ij ðU ij M ij Þ a, represents 0 6 a 6 1. Comparing Eqs. (1) and (9), the traditional
the right-end value of a-cut for aij AHP only uses a specific figure geometric mean to repre-
sent the expert opinions for the pair-ware comparison ma-
ðaaji Þk ¼ 1=ðaaij Þk ; 0 6 k 6 1; 0 6 a 6 1; i > j: ð8Þ
trix. However, the triangular fuzzy numbers are used to
The single pair-ware comparison matrix is expressed in present the fuzzy opinions and expert consensus. Mean-
Eq. (9) while, both approaches use the eigenvector method for
weight calculation.
(V) Consistency test: The essential idea of the AHP is
that a matrix A of rank n is only consistent if it has one
positive eigenvalue n = kmax while all other eigenvalues
are zero. Further, Saaty developed the consistency index
(CI) to measure the deviation from a consistent matrix:
CI ¼ ðkmax nÞ=ðn 1Þ: ð12Þ
Step 1: Define the evaluative criteria and sub-criteria used to select the
Modified Delphi Method
worst performing machine in terms of precision.
Step 8: Compute the relative weight of the elements for each level.
Step 9: Calculate the overall level hierarchy weight to select the worst
performing machine in terms of precision.
Step 10: Perform EWMA verifying analysis of previous Fuzzy AHP step EWMA
results.
Fig. 2. The manufacturing quality of silicon wafer slicing decision model to select worst machine of precision processes.
C.-W. Chang et al. / Expert Systems with Applications 34 (2008) 2210–2220 2215
n P 5 it should be less than 0.10 to get a sufficient consis- ufacturing process. Also, a previous study adopted EWMA
tent matrix. Otherwise the matrix should be revised (Saaty, to verify the analysis results of bow (Lin et al., 2002). By
1994). adjusting the drift of the wire knife to enhance the quality
(VI) Compute the overall hierarchy weight: After the of discussions among experts on slicing, the EWMA con-
weights for various hierarchies and elements are computed, trol chart demonstrates the effectiveness of the proposed
computation results for the overall hierarchy weight are com- AHP-based algorithm.
piled. Finally, the most appropriate strategy is determined.
Table 1
Operational type for defining criteria and sub-criteria factors
Code name The operating type defining
Criteria
Machine-related (C1) The defective rate is owing to the machine
Human-related (C2) Human factors result in the reason of the slice defective rate
Management (C3) Implement the goal to ensure the process yield of the wafer slice
Measurement (C4) Balance the approaches of wafer slicing using a relevant measurement procedure
Sub-criteria
Wire knife life cycle (CS1) The wire knife life has serious influence in its processing capability. As the wire knife is still used under
the state of scrap item, it tends to produce injured knife and cause chip defective rate
Machine precision (CS2) Because of using machine for a long time, the accuracy of machine becomes worse that will influence the
chip quality and the yield after processing
Parameters setting (CS3) A bad setting of the machine parameter would influence the process capability
Establish adjusting standard (CS4) Establishing a suit of managements of the standard process would make the staff deal with the problem s
procedures of the process in order
Engineer’s experience (CS5) An engineer has to accept the whole in-service training before working and has to possess the related
technique knowledge so that operating the machine practically and raising the product yield
Adjusting time (CS6) Proofreading regularly could guarantee the process capability
Color management (CS7) Use color management to enable not only staff but the raw material and the control of the poor yield
could be conducted
Online education (CS8) On-line training could enhance professional knowledge and engineering expertise, as well as reduce the
errors of the artificial importation and increase productivity
Multi-response (CS9) Adequate control is available for the multiple quality characteristics, which could effectively determine
the optimum factor-level combinations and raise the proficiency of the wafer slicing process
Method to check (CS10) A verification method to achieve the most reliable measurement would effectively promote the ability and
proficiency of the process
Measure characteristic (CS11) Formulating the measurement quality characteristics could reduce engineering errors; otherwise, the
characteristics of errors could be easily identified
2216 C.-W. Chang et al. / Expert Systems with Applications 34 (2008) 2210–2220
Multi-response
A-WD-300
Method to check
Measurement
Measure characteristic
Fig. 3. Hierarchical structure to select the manufacturing quality of silicon wafer slicing.
Table 2 Table 3
Fuzzy aggregate pair-wise comparison matrix for criteria of level 2 Aggregate pair-wise comparison matrix for criteria of level 2
Goal C1 C2 C3 C4 Goal C1 C2 C3 C4
C1 1, 1, 1 1/2, 1.909, 3 1/3, 0.951, 3 1/2, 1.147, 3 C1 1 1.829 1.309 1.449
C2 — 1, 1, 1 1, 1.582, 3 1, 1.622, 3 C2 0.547 1 1.791 1.811
C3 — — 1, 1, 1 1/4, 1.026, 3 C3 0.764 0.558 1 1.325
C4 — — — 1, 1, 1 C4 0.690 0.552 0.755 1
kmax = 4.103071; CI = 0.034357; RI = 0.90; CR = 0.038175 6 0.1
based on four evaluation criteria, 11 evaluation
sub-criteria and, finally, the alternatives (Fig. 3). Step 5: Calculate the eigenvalue and eigenvector. Using the
Step 3: Establish the triangular fuzzy numbers. According comparison matrix (such as in Table 3), the eigen-
to Table 2, administer the AHP questionnaire to vectors were calculated by Eqs. (10) and (11).
sample 16 respondents comprised of administra- Table 4 summarizes the results of the eigenvectors
tors and engineers from wafer factories, with each for criteria, sub-criteria and three diamond cut-
respondent making a pair-wise comparison of the ting machines.
decision elements and then assigning those rela- Step 6: Calculate the a-cut. Experts can determine a-cut
tive scores. subjectively, depending on environmental uncer-
Step 4: Establish each fuzzy positive reciprocal matrix. tainty for evaluating the objectives of the manu-
Perform defuzzification using Eqs. (7) and (8) facturing wafer yield of silicon wafer slicing.
and establish the fuzzy pair-wise comparison Notably, a higher a value can be selected when
matrix. For instance, the main criteria as the sam- the decision-making environment is stable and
ple, such as in Table 3. information is readily available. Conversely, when
According to Table 3, when a and c = 0.5, defuzz- the decision marking is ambiguous, and informa-
ification is performed as follows: tion is lacking, a lower a value can be used to
more accurately reflect reality. Furthermore, the
L0:5
12 ¼ ð1:909 1=2Þ 0:5 þ 1=2 ¼ 1:204;
evaluator can be based on their own judgment
and, furthermore, adopt a conservative or opti-
U 0:5
12 ¼ 3 ð3 1:909Þ 0:5 ¼ 2:454;
mistic attitude when determining k value. Where
0:5
ða0:5
12 Þ ¼ ½0:5 1:204 þ ð1 0:5Þ 2:454 ¼ 1:829; k = 0 represents the most optimistic scenario,
ða0:5
0:5
¼ 1=½ða0:5
0:5 while k = 1 is the most pessimistic scenario. Fol-
21 Þ 12 Þ ¼ 1=1:829 ¼ 0:547:
lowing modified Delphi method discussion and a
C.-W. Chang et al. / Expert Systems with Applications 34 (2008) 2210–2220 2217
Table 4
Weights of the criteria, sub-criteria and three diamond cutting machines
Criteria Weights of criteria Sub-criteria Weights of sub-criteria Global priority DFD660 DFD670 A-WD-300
C1 0.331 CS1 0.330 0.109 0.310 0.452 0.238
CS2 0.330 0.109 0.394 0.334 0.272
CS3 0.340 0.113 0.341 0.371 0.288
Global priority 0.348 0.386 0.266
C2 0.280 CS4 0.333 0.093 0.394 0.320 0.286
CS5 0.380 0.106 0.366 0.345 0.289
CS6 0.287 0.081 0.431 0.301 0.268
Global priority 0.394 0.324 0.282
C3 0.211 CS7 0.273 0.058 0.365 0.414 0.221
CS8 0.342 0.072 0.249 0.400 0.351
CS9 0.385 0.081 0.438 0.296 0.266
Global priority 0.353 0.364 0.283
C4 0.178 CS10 0.402 0.072 0.355 0.428 0.217
CS11 0.598 0.106 0.317 0.430 0.253
Global priority 0.332 0.429 0.239
Table 5
Eigenvectors (weights) of criteria under different a-cuts when k = 0.5
a-Cuts
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
C1 0.352 0.347 0.342 0.337 0.331 0.325 0.319 0.312 0.305 0.298
C2 0.289 0.287 0.285 0.283 0.280 0.278 0.275 0.272 0.270 0.267
C3 0.201 0.203 0.205 0.208 0.211 0.213 0.216 0.219 0.222 0.225
C4 0.158 0.162 0.167 0.172 0.178 0.184 0.190 0.196 0.203 0.210
consensus of opinion among the 16 experts, 0.5 of the five experts and 11 users are all <0.1, indi-
was assigned as the value of k. To understand cating ‘‘consistency’’. Furthermore, the CR of the
how criteria weights impact various evaluation aggregate matrix is also <0.1, also indicating
environments, a of 0.1–1 were attempted to more ‘‘consistency’’.
thoroughly understand the change of ranking in Step 8: Compute the relative weight of the elements for
the evaluation criteria. In the following analysis, each level. Aggregated scores provided by all deci-
when a = 1, the fuzzy number becomes concrete, sion makers are aggregated as showen in Table 4.
thus confirming the evaluation results of AHP Table 4 summarizes the relative weight of the ele-
where a < 1 falls under the fuzzy concept results. ments for each level.
Table 5 summarizes the results of eigenvectors Step 9: Calculate the overall level hierarchy weight to select
for the four criteria under various a-cut when the worst performing machine in terms of precision.
k = 0.5. The composite priorities of the alternatives are
Table 5 summarizes different a-cuts, machine- then determined by aggregating the weights
related, human-related, management and throughout the hierarchy. The composite priori-
measurement according to the magnitude of envi- ties of the alternatives are showed Table 6.
ronmental changes. When experts confer that
Table 6
uncertainty is increasingly higher, the significance Selection of the worst performing machine in terms of precision in silicon
of management and measurement becomes more wafer slicing
apparent. Conversely, when experts believe uncer- Criteria Weights DFD660 DFD670 A-WD-300
tainty is increasingly lower, the significance of Global Global Global
machine-related and human-related rises. How- priority priority priority
ever, in this study, a = 0.5 is used to express that C1 0.331 0.348 0.386 0.266
environmental uncertainty is steady. C2 0.280 0.394 0.324 0.282
Step 7: Perform the consistency test. According to Eqs. C3 0.211 0.353 0.364 0.283
(12) and (13) the criteria comparison matrix of C4 0.178 0.332 0.429 0.239
consistency for each criterion is calculated, as Result Aggregate 0.359 0.371 0.269
shown in Table 2. Results of the consistency test score
Rank 2 1 3
and the CR of the comparison matrix from each
2218 C.-W. Chang et al. / Expert Systems with Applications 34 (2008) 2210–2220
According to Table 6, ‘‘A-WD-300’’ is used to Step 10: Perform EWMA verifying analysis of previous
select the evaluation outcomes and evaluate the fuzzy AHP step results. The process standard
worst performing machine in terms of precision. deviation, r, is estimated using the X chart and,
25
Group 1 Group 4 Group 5
Group 3
Group 2 47
20 20
12 33 61
UCL
15
X
10 LCL
0
1
11
16
21
26
31
36
41
46
51
56
61
66
represents out of control Subgroup
25
Group 2 Group 3
20 Group 1 59
40
19 UCL
15
X
10 LCL
0
1
11
21
31
41
51
61
6
16
26
36
46
56
66
25 Group 3
Group 1 Group 4 Group 6
30
20 7 14 Group 2 Group 5
38 49 64
UCL
15
X
10 LCL
0
41
1
11
16
21
26
31
36
51
61
6
46
56
66
then, k = 0.3 and n = 2 are set to monitor algorithm can assist semiconductor manufacturers in simi-
and inspect the bow of three diamond cutting lar muilt-criteria questions by offering an objective and sys-
machines (DFD660, DFD670 and A-WD-300). tematic means of selecting the worst performing machine in
In this chart, 133 samples are generated while terms of precision and increasing the quality yield of silicon
the process is controlled. The upper and lower wafer slicing. Final, the proposed procedure allows engi-
control limits for the EWMA statistics are used neers to rapidly adjust a manufacturing system to eliminate
to calculate the bow of three diamond cutting problematic phenomena and increase slicing quality and
machines. In Fig. 4, the out-of-control conditions process capability.
appear at the 12th, 20th, 33th, 47th and 61th sig-
nals. In Fig. 5, the process is out-of-control at the References
19th, 40th and 59th signals. In Fig. 6, the process
is out-of-control at the 7th, 14th, 30th 38th, 49th, Alejandro, H. L., Douglas, C. M., George, R., Connie, M. B., & Richard,
and 64th signals. Unusual operating conditions of I. P. (2001). Performance of customized control charts to detect
process disturbances. Quality and Reliability Engineering International,
the manufacturing process appear in A-WD-300
17(3), 205–218.
diamond cutting machines the most. Such identi- ASTM F534 (1995). Annual Book of ASTM Standards.
fication significantly enhances the quality yield ASTM F657 (1995). Annual Book of ASTM Standards.
of silicon wafer slicing manufacturing. Conse- Braha, D., & Shmilovici, A. (2002). Data mining for improving a cleaning
quently, the quality of the EWMA chart for the process in the semiconductor industry. IEEE Transactions on Semi-
bow verifies the analysis results, in which the conductor Manufacturing, 15(1), 91–101.
Buckley, J. J. (1985). Fuzzy hierarchical analysis. Fuzzy Sets and Systems,
FAHP results are the same. Based on the evalua- 17, 233–247.
tion outcomes, ‘‘A-WD-300’’ is determined to be Butler, S. W., & Stefani, J. A. (1994). Supervisory run-to-run control of a
the worst performing machine in terms of polysilicon gate etch using in situ elliposometry. IEEE Transactions on
precision. Semiconductor Manufacturing, 7(2), 193–201.
Cunningham, S., Spanos, C. J., & Voros, K. (1995). Semiconductor yield
improvement: Results and best practices. IEEE Transactions Semicon-
ductor Manufacturing, 8(2), 103–109.
4. Conclusions Ho, C. T. (2004). Strategic evaluation of emerging technologies in the
semiconductor foundry industry (special case: Semiconductor foundry
In most of the real-world problems, some of the decision industry in Taiwan) (pp. 251–278). Portland State University.
Hsu, T. H., & Nian, S. H. (1997). Interactive fuzzy decision aided systems-
data can be precisely assessed while others cannot. Real
a case on public transportation system operations. Journal of Trans-
numbers are used to represent data which can be precisely portation Taiwan, 10(4), 79–96.
measured. For those data which cannot be precisely Hsu, T. H., & Yang, T. H. (2000). Application of fuzzy analytic hierarch
assessed, fuzzy sets can be used to denote them. The use process in the selection of advertising media. Journal of Management
of fuzzy set theory allows us to incorporate unquantifiable and Systems, Taiwan, 7(1), 19–39.
information, incomplete information, non-obtainable Huang, L. C., & Wu, R. Y. H. (2005). Applying fuzzy analytic hierarchy
process in the managerial talent assessment model – an empirical study
information and partially ignorant facts into the decision in Taiwan’s semiconductor industry. International Journal of Technol-
model. When decision data are precisely known, they ogy Management, 30(1–2), 105–130.
should not be forced into a fuzzy format in the decision Kang, H. Y., & Lee, A. H. I. (2007). Priority mix planning for
analysis. Crisp MADM methods solve problems in which semiconductor fabrication by fuzzy AHP ranking. Expert Systems
with Applications, 32(2), 560–570.
all decision data are assumed to be known and must be rep-
Lin, C. T., Chang, C. W., & Chen, C. B. (2004). Focus groups: Impact of
resented by crisp numbers. The methods are to effectively quality and process capability factors on silicon wafer slicing process.
aggregate performance scores. Fuzzy MADM methods International Journal of Manufacturing Technology and Management,
have difficulty in judging the preferred alternatives because 2(2), 171–184.
all aggregated scores are fuzzy data. Therefore, selecting Lin, C. T., Chang, C. W., & Chen, C. B. (2005). Relative control
philosophy- balance and continual change for forecasting abnormal
silicon wafer slicing manufacturing quality system is an
quality characteristics in a silicon wafer slicing process. International
extremely complex issue, often relying on the subjective Journal of Advanced Manufacturing Technology, 26(9–10), 1109–1114.
assessment of decision makers. In particular, administra- Lin, C. T., Chen, C. B., & Chang, C. W. (2002). Screening synchronously
tors and engineers in semiconductor manufacturing lack occurred multiple abnormal quality characteristics screening in a
objective decision-making procedures and clearly defined silicon wafer slicing process. The Asian Journal on Quality, 3(1), 48–60.
evaluating criteria. Therefore, this study presents a fuzzy Liu, T. S., & Wang, M. J. (1992). Ranking fuzzy numbers with integral
value. Fuzzy Sets and Systems, 49(3), 247–255.
AHP-based algorithm to determine manufacturing quality Loon, C. T., & Su, E. T. (1999). Computing process capability indices for
in silicon wafer in order to identify the worst performing non-normal data: a review and comparative study. Quality and
machines in terms of precision. The EWMA control chart Reliability Engineering International, 15, 339–353.
is also adopted to identify the worst performing diamond Lucas, J. M., & Saccussi, M. S. (1992). Exponentially weighted moving
average control charts: properties and enhancements. Technometrics,
cutting machine in terms of precision as the A-WD-300
32, 1–12.
one. The proposed fuzzy AHP-based algorithm signifi- Murry, J. W., & Hammons, J. O. (1995). Delphi: A versatile methodology
cantly contributes to efforts to upgrade manufacturing for conducting qualitative research. Review of Higher Education, 18(4),
quality in silicon wafer slicing. Specifically, the proposed 423–436.
2220 C.-W. Chang et al. / Expert Systems with Applications 34 (2008) 2210–2220
Pai, P. F., Lee, C. E., & Su, T. H. (2004). A daily production model for Sung, W. C. (2001). Application of Delphi method, a qualitative and
wafer fabrication. International Journal of Manufacturing Technology quantitative analysis, to the healthcare management. Journal of
and Management, 23(1–2), 58–63. Healthcare Management, 2(2), 11–19.
Robbins, S. P. (1994). Management. New Jersey: Prentice Hall. Takeshi, H. (1998). Ultraclean surface processing of silicon wafers: Secrets
Saaty, T. L. (1980). The analytic hierarchy process. New York: McGraw of VLSI manufacturing. Springer.
Hill. Tzeng, G. H., Chiang, C. H., & Li, C. W. (2007). Evaluating intertwined
Saaty, T. L. (1990). How to mark a decision: The analytic hierarchy effects in e-learning programs: a novel hybrid MCDM model based on
process. European Journal of Operational Research, 48(1), 9–26. factor analysis and DEMATEL. Expert Systems with Applications,
Saaty, T. L. (1994). The fundamentals of decision making and priority 32(4), 1028–1044.
theory with the analytic hierarchy process. AHP series (Vol. VI). Yang, T., Su, C. T., & Hsu, Y. R. (2000). Systematic layout planning: A
Pittsburgh, PA: RWS Publication, pp. 1–527. study on semiconductor wafer fabrication facilities. International
Saaty, T. L. (2000). Fundamentals of decision making and priority Journal of Operations and Production Management, 20(11), 1360–1372.
theory with the analytic hierarchy process. Pittsburgh: RWS Publica- Yurdakul, M. (2004). AHP as a strategic decision-making tool to justify
tions, pp. 1–478. machine tool selection. Journal of Materials Processing Technology,
Sachs, S. E., & Montgomery, D. C. (1996–1997). Process capability indices 146(3), 365–376.
and non-normal distributions. Quality Engineering, 7, 305–316. Zadeh, L. (1965). Fuzzy sets. Information and Control, 8, 338–353.