Reduction of Defects in The Lapping Process of The

Engineering Management in Production and Services
Volume 11 • Issue 2 • 2019
received: 5 December 2018

accepted: 5 June 2019
Reduction of defects in the lapping pages: 87-105
process of the silicon wafer

manufacturing: the Six Sigma
application
Mithun Sharma, Sanjeev P. Sahni, Shilpi Sharma
ABSTRACT
Aiming to reduce flatness (Total Thickness Variation, TTV) defects in the lapping process
of the silicon wafer manufacturing, it is crucial to understand and eliminate the root
cause(s). Financial losses resulting from TTV defects make the lapping process
unsustainable. DMAIC (Define, Measure, Analyse, Improve and Control), which is a Six
Sigma methodology, was implemented to improve the quality of the silicon wafer
manufacturing process. The study design and the choice of procedures were contingent
on customer requirements and customised to ensure maximum satisfaction; which is
the underlying principle of the rigorous, statistical technique of Six Sigma. Previously
unknown causes of high TTV reject rates were identified, and a massive reduction in
the TTV reject rate was achieved (from 4.43% to 0.02%). Also, the lapping process
capability (Ppk) increased to 3.87 (beyond the required standard of 1.67), suggesting
sustainable long-term stability. Control procedures were also effectively implemented
using the techniques of poka yoke and control charts. This paper explores the utility of
Six Sigma, a quality management technique, to improve the quality of a process used Corresponding author:
in the semiconductor industry. The application of the Six Sigma methodology in the
current project provides an example of the root cause investigation methodology that Mithun Sharma
can be adopted for similar processes or industries. Some of the statistical tools and
techniques were used for the first time in this project, thereby providing new analysis O.P. Jindal Global University, India
and quality improvement platform for the future. The article offers a deeper e-mail: mithun.sharma@gmail.com
understanding of the factors that impact on the silicon wafer flatness in the lapping
process. It also highlights the benefits of using a structured problem-solving Sanjeev P. Sahni
methodology like Six Sigma.
O.P. Jindal Global University, Indiea
e-mail: drspsahni@jgu.edu.in
KEY WORDS
Six Sigma, quality, silicon wafer, lapping, quality control, total thickness variation,
wafer manufacturing Shilpi Sharma
DOI: 10.2478/emj-2019-0013 O.P. Jindal Global University, India

e-mail: shilpi@jgu.edu.in
Introduction
business and $5000 billion in services, representing
The semiconductor industry consists of compa- close to 10% of the global GDP, thus gaining the rec-
nies engaged in the design and fabrication of semi- ognition for its critical role in the supply chain of the
conductor devices, which form the foundation of electronics industry (Kazmierski, 2012).
modern electronics. The industry began in the 1960s The silicon wafer manufacturing involves several
and currently accounts for about 0.5% of the global stages. The semiconductor industry consists of three
GDP ($299.5 billion). It also enables the generation of key sectors: silicon ingot growing, the silicon wafer
approximately $1200 billion in the electronic systems manufacturing and the fabrication of integrated cir-
87
Tab. 1. Four key stages of Silicon wafer manufacturing proces
Tab. 1. Four key stages of silicon wafer manufacturing process
PROCESS FLOW DESCRIPTION PROCESS OBJECTIVE PROCESS DEMERITS
Slicing Silicon ingot is mounted on the slicing machine. A
To generate wafer slice structure. Poor flatness High
web of meatl wire along with cutting slurry passes
To achieve correct wafer surface damage High
through the ingot slowly to provide disc shaped
thickness, orientation and warp contamination
silicon wafers
Lapping Sliced wafers held in carriers are placed in between
two metal plates along with abrasive slurry falling Reduce slicing damage. Establish Susceptible to surface
on wafers. Rotation of plates and carriers causes optimum wafer flatness and flatness rejects
mechnical removal of slicon therby reducing wafer
Chemical Etching Lapped wafers are subjeted to abrasive chemicals
which disolves silicon into chemical thereby thinning Reduce Previous Process Damage Degrades flatness
wafers Reduce surface contamination wafer staining
Polishing Etched wafers are mounted on plates. With

Establish optimum flatness.
pressure applied on these plates, the wafers are High rejection rate Re-
Eliminate surface damage.
rubbed against fine abrasive polishing pads along work cost
Minimise contamination level
with polishing slurry to give a mirror like finish
cuit
Tab. 2(IC) chips.
Overview Thequality
of key silicon ingot growing
management conceptsprocess is 1. Literature review
contingent
Concept on many factors, such
Origin Aim as size, specifica- Focus Methodology and Criticisms
tions and quality, and so the ingot growing time can Tools
range from a weekJapan
Total Productive
to a(the
1950s)
month. Increase
The next process capability Preventive
stage is the and isGap
Quality an analysis
elusiveofand abstract
historical records,
Skilled concept
workers (Hos-
required to implement,
Maintenance by reduction of unplanned predictive
wafer production, whereby a failures, fully-grown
accidentssilicon
and
sain, Tasnim &cause-effect
maintenance of
Hasan, 2017; Wright,
analysis 1997).
resource In a study
demanding
ingot is sliced into wafers of different defects thicknesses. processesby Evans and Dean (2002), managersand from
long-term 86 firms in
This sliced wafer is then subjected to the flattening the United States were asked to define “quality”. They
process, and
Total Quality thenJapan
Control undergoes
(the the fabrication process To reduce found severalMethodology:
definitions Plan ranging from waste elimi-
Do Vague and difficult to
producing IC chips. The focus ofTothis
1960s) research
coordinate project rework
quality nation,
and conformity to Tools:
Study Act customer requirements,
coordinate policy
maintenance and compliance,
achieve Statistical techniques
consistency in output, getting it right the
is on the silicon wafer manufacturing.
improvement from all maximum
first time and customer satisfaction. Broadly, quality
The silicon wafer manufacturing groups consists
to achieveof thefour
most customer
key value-adding processes: slicing, lapping,
economical can be understood along four dimensions: excellence,
chemical satisfaction
process
Total Customer
value, conformance to specifications
Plan Do Vague andandthe extent to
etching andQuality Japan The
polishing. (the primaryImprove
aimthe of quality and
the slicing Methodology:
Management (TQM) 1990s) consistency of processes satisfaction Study Act Tools:
which expectations are met (Yong & Wilkinson,inconsistent
process is to define the crystal structure of the wafer, Statistical techniques conceptualisation,
finding the best possible shape. The wafer is subjected 2002). Quality as excellence is believed excessive toresource
be immea-
to high-pressure cutting, to achieve correct wafer surable, and control can only be exerted
consumption,through the
investment of maximum efforts unsatisfactory
and best skill sets.
thickness, but on the downside, it causes high surface resu lts
damage and contamination. The second stage is lap- This approach has been criticised of being of little
Zero Defects
ping, which removes DenvertheDivision
surfaceTo damage
enhance the quality by
caused practical value
of a Defect toattention
Extra an organisation
and (Garvin, 1988).
Expensive
of the Martin process outcome through elimination Quality as a care devoted
value to each focuses on external
definition
slicing. It is also critical
Marietta
in defining the flatness of the
the elimination of any step of the production
wafer as a failure Corporation
to achieve(the thedefects
optimum in thewafer flat-
production effectiveness (meaning costs)
process ensuring noand internal efficiency.
1960s)
ness can lead to complete wafer process
rejection. Wafers are This definition, once
mistakes again, has been argued to lack
then exposed to abrasive chemicals during the practical and measurable parameters for organisa-
chemical
Reliability etching stage that helps to remove impuri-
To reduce failure modes tional use (Yong
Longevity and
& Wilkinson, 2002).
Reliability-Centred
A more quanti-
Technical and requires
engineering
ties from the wafer surface. Finally, wafers that have fiable
dependabilityand
of objective
Maintenance, definition
failure skilledofstaff.quality
Expensiveis
Shewart during
been manufactured to the required flatness standards, conformance
parts, products and to specifications;
modes and effects, which
and was
demands dominant
long term
the 1920s and systems root cause analysis, commitment
with no surface damage, are polished on one side to in the 20th century. It means
condition- based
that an outcome must
the 1930s: cited
give a smooth, mirror-like
in Kapur andfinish, on which IC chips not deviate from the specifications set by an organisa-
maintenance
can be fabricated.Lamberson
This process flow is further illus- tion, and any deviations are considered as lowering
(1977) the quality (Reeves & Bednar, 1994). Such confor-
trated in Tab. 1.
The lapping process helps to achieve the maxi- mance measures have been argued to increase the
mum wafer flatness. Wafer flatness measured as Total internal efficiency of the process and sale prices over
Thickness Variation (TTV) in microns is one of criti- time (Topalovic, 2015). Later, as the focus shifted
cal-to-quality (CTQ) requirements for a silicon wafer. from manufacturers to customers, a new definition of
This project aimed to reduce the number of TTV “quality” emerged, i.e. meeting or exceeding customer
rejects in the lapping process. expectations. Anything that does not satisfy the cus-
tomer was considered to be of low quality (Yong &
88
Wilkinson, 2002). However, customer requirements tions have documented unsuccessful implementation
are variable and subjective, which makes them diffi- stories of the TQM concept in the manufacturing
cult to satisfy. These intricacies associated with the industry (e.g., Brown et al., 1994; Eskildson, 1994;
concept of quality, highlight the need to have a man- Cao et al., 2000; Nwabueze, 2001). Based on the evi-
agement system that ensures the manufacturing of dence from independent publications by consulting
good-quality products resulting in monetary profits firms, it can be argued that two-thirds of TQM imple-
and customer satisfaction. This leads to the concept mentation efforts failed to produce any significant
of quality management. improvements in the overall quality of the product,
The revolution in quality management began in financial gains or the company’s competitive situation
Japan during the 1950s and gradually gained momen- in the industry (Jimoh et al., 2018). This is partly due
tum in the rest of the world during the 1970s and the to the ever-changing and evolving definitions of
1980s (Foley, 2004). One of the most influential qual- TQM, which can mean different things to different
ity movements during this era was the total quality people, making its implementation insufficiently
management (TQM), which again owes its origin and consistent and reliable (Andersson et al., 2006;
conceptual developments to Japan (Cole, 1998; Esaki, Boaden, 1997; Talapatra, Uddin & Rahman, 2018).
2016; Juran, 1995). TQM eliminated the weaknesses Two other branches of quality management were
of previous quality improvement techniques, which introduced during the 1960s: reliability engineering
makes it an efficient quality management tool. TQM and zero defects. The reliability engineering tech-
is an integrated approach that stresses the top-down nique has roots in the disciplines of pure probability
approach, staff engagement in the process, evidence- and statistics. It was mostly used in the USA and
based decision-making and the consideration of cus- aimed to apply principles of probability to reduce
tomer requirements (Tobin, 1990). Despite its defect rates of durable products (Dimitri, 1991). The
influential and successful reputation, many publica- zero defects strategy, which originated in the USA
Tab. 2 Overview of key quality management concepts
METHODOLOGY AND
CONCEPT ORIGIN AIM FOCUS CRITICISMS
TOOLS
Increase process capability Preventive and Skilled workers

Gap analysis of
Total Productive Japan (the by reduction of unplanned predictive required to implement,
historical records,
Maintenance 1950s) failures, accidents and maintenance of resource demanding
cause‐effect analysis
defects processes and long‐term
To reduce
To coordinate quality
rework and
maintenance and Methodology: Plan Do
Japan (the achieve Vague and difficult to
Total Quality Control improvement from all Study Act Tools:
1960s) maximum coordinate
groups to achieve the most Statistical techniques
customer
economical process
satisfaction
Vague and
inconsistent
Customer Methodology: Plan Do conceptualisation,
Total Quality Japan (the Improve the quality and
Study Act Tools: excessive resource
Management (TQM) 1990s) consistency of processes satisfaction
Statistical techniques consumption,
unsatisfactory
resu lts
Denver Division To enhance the quality of a Extra attention and

of the Martin process outcome through Defect care devoted to each
Zero Defects Marietta the elimination of any step of the production Expensive
elimination
Corporation (the defects in the production process ensuring no
1960s) process mistakes
Shewart during Reliability‐Centred
the 1920s and Longevity and Maintenance, failure Technical and requires
Reliability the 1930s: cited dependability of modes and effects, skilled staff. Expensive
To reduce failure modes
engineering in Kapur and parts, products and root cause analysis, and demands long term
Lamberson systems condition‐ based commitment
(1977) maintenance

89
during the 1960s, was considered to be the most yearly savings (Su & Chou, 2008; Yang & Hsieh,
optimistic approach in the quality management field 2009).
as it aimed to achieve the complete elimination of Besides, in the last decade or so, there has been
defects and process failures (Crosby, 1979). Both a massive uptake and implementation of the Six
concepts, however, received criticism for being Sigma technique as a process change, management
impractical and expensive (Crosby, 1984). It can, and improvement strategy by global industries, which
thus, be argued that the different quality management include the manufacturing process (Al-Aomar, 2006;
concepts differ in their origin, aim, definitions, meth- Gangidi, 2019; Valles et al., 2009), financial organisa-
odology and focus, thereby confusing rather than tions (Brewer & Eighme, 2005), engineering firms
informing the readers. Furthermore, each of the (Bunce et al., 2008), hospitals and intervention clinics
techniques cited above has limitations that reduce (Craven et al., 2006), banking, hospitality, pharma-
their applicability and anticipated benefits (Kedar et ceutical companies (Cupryk et al., 2007), chemical
al., 2008). Many organisations have reported difficul- industries (Doble, 2005), educational institutions
ties in the implementation of quality management (Bandyopadhyay & Lichtman, 2007), software indus-
programmes (Brown et al., 1994; Eskildson, 1994; try (Arul & Kohli, 2004), call centres (Schmidt
Harari, 1997; Nwabueze, 2001); they suggested the & Aschkenase, 2004), utility service providers (Agar-
lack of inherent connectivity between parts and also wal & Bajaj, 2008), the automobile sector (Gerhorst et
reported some missing information about some of al., 2006), information technology (Edgeman et al.,
the relevant sections, as shown in Tab. 2. 2005), human resources departments (Wyper & Har-
None of the quality management tools discussed rison, 2000), military administration units (Chappell
so far had global success, and quality managers were & Peck, 2006) and even government departments
still in search of a complete quality management (Furterer & Elshennawy, 2005). A summary is pre-
programme when Six Sigma arrived. Six Sigma is sented Tab. 3.
a systematic set of guidelines that aims to significantly The origins of Six Sigma can be found in statis-
improve the quality of a manufacturing process and tics. The term Six Sigma owes its origin to the termi-
reduce costs by minimising process variation and nology employed in the statistical modelling of
reducing defects. It utilises statistical tools that can manufacturing processes. A Six Sigma process is the
either be applied to facilitate a new product develop- one that produces 3.4 defects or non-conformances
ment or strategic process improvement (Breyfogle et per million opportunities (DPMO). A defect is any-
al., 2001). Six Sigma is the edge that helps win the thing that does not conform to the manufacturer’s
market competition as it provides financial, business guidelines or customer’s specifications, and an oppor-
and personal benefits; financial — by optimal and tunity is any chance for this defect to happen. The
efficient use of resources; business — through ensur- sigma level, also known as a Z-value, is used as
ing maximum customer satisfaction; and personal a capability index for the process, which indicates
— enhancing skills of an individual and, thereby, how well that process can meet the customer’s
increasing their employability. In the last decade or requirements (Bothe, 2001; Da Silva et al., 2019).
so, there has been a rapid uptake of the Six Sigma Each sigma level corresponds to a certain number of
technique as a process change, management and defects/non-conformances that are associated with
improvement strategy by global industries. This the process, as shown in Fig. 1.
helped them beat market competition and maximise
Sigma (σ) Defects / Non-Conformances per

Level Million Oppurtunity (DPMO)
2σ 308,537
3σ 66,807
4σ 6,210
5σ 233
6σ 3.4
Fig. 1. Sigma levels depending on DPMO
90
Tab. 3. Selected success stories of the Six Sigma implementation in industries
AUTHORS NAME OF AN ORGANISATION BENEFITS OF IMPLEMENTING THE SIX SIGMA TECHNOLOGY

Financial sector
Citibank group Numerous benefits have been reported across different organisations of this
group. They successfully halved their credit processing time and reduced
internal call‐back time by 80% and external by 85%. Reduced the time between
a customer first placing an order till the actual service delivery and the credit
Rucker decision cycle from 3 days to just 1 day. The time taken to process a statement
(2000) was also decreased from 28 days to 15 days only
JP Morgan Chase (Global Investment Improved customer experience in using bank's services such as account
Banking) opening, balance enquiry, making transfers and payments via online or cheque
mode; leading to increased customer satisfaction and a reduction in process
cycle time by more than 30%
Antony British Telecom wholesale Financial benefits of over $100 million, greater customer satisfaction, error
(2006) reduction
Bank of America 24% reduction in customer complaints and a 10.4% increase in customer
Roberts
satisfaction
(2004)
Sun Trust Banks Significant improvement in customer satisfaction
American Express Improved the external vendor related processes and reduced the numbers of
Bolt et al. (2000) non‐received renewal cards
Manufacturing sector
Motorola (1992 and 1999) 1992: Achieved dramatic reduction in the defect levels of their process by about
150 times
1999: Huge financial gains of about $15 billion over 11 years
Honeywell Profit of $1.2 billion
Antony
Texas Instruments Achieved a financial gain of over $ 600 million
(2006)
Johnson and Johnson The financial gain of about $500 million
Telefonica de Espana (2001) Whooping increase in revenue by about 30 million in the first 10 months and
gain in savings too
Dow chemical/rail delivery project Reported substantial savings in capital expenditures: of over $2.4 million
AlliedSignal/ Bendix IQ brake pads The cycle time of their production‐shipment process decreased by 10 months
(18 to 8 months)
AlliedSignal/ Laminates plant in South Reaped a range of benefits: their capacity almost doubled, and punctuality in
Carolina delivering goods reached 100% threshold level. Their cycle time and inventory
had a reduction of 50% each
DuPont/Yerkes plant in New York Increase in yearly savings of over $25 million
McClusky
(2000)
(2000)
Seagate Technology Gained financial profits of about £132 million in just 2 years
General Electric Increase in yearly financial savings by about $2 billion
Hughes Aircraft's Missiles Systems
Group / Wave soldering op. The quality of their yields improved by about 1000% and productivity by 500%
Raytheon/ Aircraft Integration Achieved a significant reduction (approx. 88%) in the inspection time spent on
Systems the depot maintenance process
GE/ Railcar leasing business Achieved a 62% reduction in the time spent at repair stops
McClusky
(2000)
Healthcare sector
Radiology film library, Anderson Service quality improved
cancer centre Preparation and waiting times for patients reduced from 45 min to 5 min
Benedetto
Dramatic increase by 45% in daily numbers of examinations without an increase
(2003)
Outpatient CT exam lab at the in workforce/ equipment
University of Texas
Engineering and Construction sector
General Electric 1997: made a profit of $320 million which was more than double their goal of
Byrne(1998) $150 million, further in 1999, annual savings of 2 billion
Magnusson et Volvo cars (Sweden) Profit of over 55 million euro in the years 2000 and 2002

al. (2003)
Business Unit of Transmission and Savings of over 200 million euro between the years 1997‐2003
Anderson et al.
Transportation Networks at Ericsson
(2006)

91
Each manufacturing process has set specification outcome. This gives FED an edge over the OFAT
limits for a process and product quality. If six stan- technique, wherein only one factor can be evaluated
dard deviations can be managed between the statisti- at one time.
cal mean of a process and its nearest specification Many studies have compared OFAT and FED —
limits, then all aspects of that process would meet the the two problem-solving techniques — and FED is
specification criteria. This distance between the pro- considered to be more effective than OFAT for the
cess mean and the specification limit is measured in following reasons (Czitrom, 1999):
sigma units and is known as the process capability. • In FED, investment of comparatively fewer
The process capability measurement index is the resources (time, money and material) results in
process performance index (Ppk). Once a process has greater and more accurate information. This
been brought under statistical control through the makes it extremely useful in industries, where
implementation of a Six Sigma project, Ppk estimates time and financial costs of running a process are
how stable these improvements would be in the long- extremely high;
term and how closely they would meet customer • Interactions between factors cannot be identified
expectations. The larger the Ppk value, the less is the using OFAT technique, which uses trial and error
process variability and the higher the long-term sta- method; whereas the FED technique provides
bility. To satisfy customers, the Ppk value should be a systematic procedure for estimating interac-
greater than 1.67 (Kotz, 1993; Raman & Basavaraj, tions between several factors;
2019). • Each observation carried out during a factorial
The literature indicated that even if a process experiment considers all the factors and their
achieved a high sigma level over short-term, its per- interactions, which estimate the effect of factors
formance could have declined over long-term; and much more precisely. In contrast, OFAT typically
a common research finding is that it might fall from uses only two observations to measure the effect
Six Sigma to the 4.5 sigma level (Alexander, Antony of one factor; these estimations are subject to
& Rodgers, 2015; Pandey, 2007). This could happen greater variability.
because the process may ‘drift’ over time. Such shift- The current study used the FED technique to
ing can further lead the process mean to move away address the problem of high TTV rejects in the lap-
from the target, thus reducing the number of standard ping process. At a broader level, the Six Sigma meth-
deviations that can fit between the process mean and odology was applied for improving the output quality
the closest specification limit. This is commonly of the lapping process. Six Sigma emphasises the need
known as a 1.5 sigma shift. So, the standardised defi- to identify and clearly define customer requirements
nition of the Six Sigma quality considers this shift and and internal industrial factors for setting goals. The
guarantees that a six-sigma process will produce no data-driven rigour component of the Six Sigma
more than 3.4 DPMO (Antony, Snee & Hoerl, 2017; approach delineates objective decision-making
Harry, 1988). purely guided by the statistical analysis of data to
Some previous attempts to resolve the problem of determine process strengths and weaknesses. It is
high TTV failed. The problem of obtaining high TTV crucial to this approach that a solution is not offered
rejects had been a recurrent issue at the current until the problem has been clearly and completely
organisation, for a significantly long time. Earlier, the defined (Ishikawa, 1985; Kume, 1985, 1995; Hoerl,
traditional problem-solving technique called One 1998; Sreedharan et al., 2019).
Factor At a Time (OFAT) had been applied in an
attempt to reduce TTV rejects. OFAT is an experi-
mental technique that evaluates the impact of poten-
2. Research methods
tial factors on the process outcome, one at a time
while keeping other factors constant. However, these The Six Sigma methodology was applied to
attempts failed to identify the root cause of the cur- resolve the issue of high TTV rejects in the wafer lap-
rent problem. Further expert consultations and sys- ping process, of the semiconductor manufacturing
tematic research review indicated a better potential of industry. The problem of the current project was for-
the technique called Factorial Experimental Design mulated as follows: the mandatory replacement of
(FED), as compared to the OFAT strategy. FED evalu- slurry in the lapping process results in poor wafer
ates the effect of more than one factor together with flatness causing TTV rejects to increase from 0.1% to
their interactions, simultaneously on the process 4.43% or a loss of £58k/month.
92
The overall aim of this research was to evaluate according to customer’s preferences, these are
the effectiveness of Six Sigma aiming to improve the known as Critical to Quality (CTQ) metrics;
quality output of the lapping process in the silicon • Measure: at this stage, in accordance with the set
wafer manufacturing industry. goal and CTQ specifications, further measure-
Specific research objectives were to test the utility ments of key elements for the concerned process
of the Six Sigma methodology in: are carried out;
• Identifying the factors responsible for high TTV • Analyse: this stage includes an in-depth and sci-
defect rates; entific investigation of collected data, statistical
• Implementing sustainable long-term process tools are utilised to identify and assess the statis-
improvements that will reduce the defect rate to tical significance of associations between various
<0.1%; aspects of a process. Any findings should be
• Increasing the Lapping Process Performance objective, valid and justifiable by means of data.
index (Ppk) to >1.67; Analyses results help reveal the root cause of the
• Delivering at least a £25k saving to the company defect in the manufacturing process. Results
by the end of the project. from the Analyse stage may also be treated as
The rationale for the project selection was based pilot results which are documented to facilitate
on the experience of one of the authors of this paper, replication of the same process design in the
who worked as a process engineer in the world’s lead- future;
ing semiconductor wafer manufacturing company • Improve: the results from the analyses were
and received black belt level in-service training in the applied to eliminate the root cause of defects in
Six Sigma methodology. The issue of high TTV a process, thereby improving the overall quality
rejects had been causing huge financial losses (> of the outcomes;
£50k/month) to the organisation and was topping the • Control: the methodology and results of each
priority list of the senior management. It was, thus, stage are clearly spelt out in sufficient detail to
considered necessary to apply the statistically vali- allow the replication in future processes to iden-
dated, well-known, effective strategy of Six Sigma to tify and correct early errors and prevent a finan-
address this problem. A team of the appropriately cial loss due to defects in the yield. At this stage,
qualified technical staff was delegated to undertake Six Sigma tools of poka yoke, statistical process
this task, under the leadership of a trained Six Sigma and quality control charts, and control plan were
black belt specialist, the first author of this article. used.
DMAIC, which is a Six Sigma process, was
employed for achieving the above stated objectives.
Six Sigma, inspired by Deming’s Plan-Do-Check-
3. Research results
Act cycle, has two popular methodologies, namely,
DMAIC and DFSS. The DMAIC methodology is uti- Stage 1 — Define: the key aim of this phase was to
lised for improving an existent process whereas DFSS mutually agree on a clear and concise problem state-
— for the development of a new product. The current ment, gain a fuller understanding of the process and
investigation followed the DMAIC principles, follow- identify the manageable focus area.
ing the five stages: To clearly define and communicate the issue at
• Define is the most important and critical stage of hand, the following problem statement was developed
the Six Sigma process. First and foremost, project based on IS-IS Not Analysis: the mandatory replace-
scoping and mapping is carried out that helps ment of slurry in the lapping process is resulting in
explain the basic problem to all the team mem- poor wafer flatness causing TTV rejects to increase
bers. Then, current process defects are defined from 0.1% to 4.43% or a loss of £58k/month.
Tab. 4. Critical to Quality metrics of the lapping process
TYPE NAME VOC CTQ
External Polishing No sharp roll‐off at the edge of wafer For Loose Spec TTV < 3.5 µm
For Tight Spec TTV < 2.0 µm
Internal Lapping Comparable yield to old active agent TTV reject < 0.10 %
Good Flatness For Tight Spec TTV < 2.0 µm

Tab. 5. Data collection plan
ITEM NO. WHAT WHY WHEN HOW WHERE WHO
1 Wafer Traceability To correlate poor TTV Every Lot Data capture Central Database ‐ Lapping Op 93

with different application Browser
Lapping ADE, Data
2 TTV KPOV, CTQ Every wafer Central Database CW Insp. Op
Process mapping: all the essential components All causes identified during the making of the
and steps of lapping were outlined in detail, such as C&E diagram were combined in broad categories to
material flow, operational activities, resources and form clusters. These clusters (five in the current sce-
material required; and the desired standards of the nario) formed the highest level in the Y=f(x) cascade,
outcome were established. The aim is to provide all which was drilled into lower levels till it reached
the team members with the understanding of the a manageable scope, as shown in Fig 2. Key input
internal requirements of the process to allow for variables to be investigated are highlighted in blue
effective and quicker reviews in the case of any errors. and green.
Critical to quality metrics: in Six Sigma, customer It should be noted that Six Sigma is an iterative
requirements are expressed through the Voice of process, and it must continue running until the
Customer (VOC), which is converted into quantifi- desired results are achieved. So, in this case, if the
able Critical to Quality (CTQ) metrics. Tab. 4 displays chosen experimental variables did not result in any
the results on VOC and CTQ characteristics for both improvement, remaining inputs were selected for the
internal and external customers in the context of the next iterative cycle.
current project. As lapping supplies two different Stage 2 — Measure: from the define stage, four-
products to polishing, there were two different CTQs teen input variables were identified, for which the
for the same customer requirement (VOC). These data collection plan using the Kipling’s checklist was
CTQs were used throughout the project to assess developed, as shown in Tab. 5.
improvements made to the lapping process. MSA — the measurement systems analysis —
Cause and Effect (C&E) Diagram is also known was conducted on all the measurement devices used
as the fishbone, 6M or Ishikawa diagram (Ishikawa, for data collection, which aimed to identify sources of
1968). It is a tool to facilitate brainstorming, identify- variations, induced due to the measurement process.
ing the causes for an effect under six broad categories It included checks on measurement devices, person-
of Measurements, Material, Man, Environment, nel engaged in the data collection process, their skill
Methods and Machines. In the current project, poor sets, adequacy and accuracy of specifications, raw
Sigma (σ) Defects / Non-Conformances per
flatness or TTV reject (CTQs) was the effect and material and the measurement procedure. In the cur-
Level Million Oppurtunity (DPMO)
a potential factor responsible for poor flatness. The rent study, two MSA tools of Gage R & R and Gage
2 σ scoping identified over one hundred
initial project 308,537of Bias & Linearity Study were used. Results from both
potential causes for TTV rejects. It would have been studies suggested that all measurement tools were fit
3σ 66,807
excessively time consuming to analyse all these fac- for purpose.
4 σ important to prioritise potential
tors. So, it was 6,210
causes Establish the Baseline DPMO: first and foremost,
and reduce the project scope to a manageable extent, the starting line for the process was established with
5σ 233
which was achieved by using the Y=f(x) cascade tool the help of DPMO and process capability (Ppk). Key
6 σ in Fig. 2.
as shown below 3.4 CTQ was TTV based on loose and tight specifica-
tions. The process capability for both types of materi-
Fig. 1. Sigma levels depending on DPMO
Fig. 2. Y = f (x) cascade
Process Capability of TTV for Tight Spec Material Process Capability of TTV for Loose Spec Material
94
LB USL LB USL
P rocess D ata Within P rocess D ata Within
LB 0 Ov erall LB 0 Ov erall
Level
External Polishing Million Oppurtunity (DPMO)For Loose Spec TTV < 3.5 µm
No sharp roll‐off at the edge of wafer
For Tight Spec TTV < 2.0 µm
Internal
2 σLapping 308,537
Comparable yield to old active agent Engineering Management in Production and Services
TTV reject < 0.10 %
3σ Good Flatness 66,807 For Tight Spec TTV < 2.0 µm

4σ
Tab. 5. Data collection plan 6,210
ITEM NO. WHAT WHY WHEN HOW WHERE WHO
5σ To correlate poor TTV 233 Data capture Central Database ‐ Lapping Op
1 Wafer Traceability Every Lot
with different application Browser
2
6σ
TTV KPOV, CTQ
3.4
Every wafer
Lapping ADE, Data
Central Database CW Insp. Op
upload
Every time fresh
3 Fig. 1.Slurry Mixing Time
Sigma levels depending
KPIV on DPMO Machine Setting (SOP) Slurry Sheet Lapping Op
slurry prepared
4 Slurry Density KPIV Every time fresh Manual Measurement Control Charts Lapping Op
slurry prepared
5 Machine flowrate KPIV Every time loop Tab. 4. Critical to Quality metrics of the lapping process
Manual Measurement Control Charts Lapping Op
was changed
KPIV (can affect wafer Every 40 hr of T YPE NAME VOC
6 Plate Shape Using dial gauge Plate Shape Sheet Lapping Op
shape) Opertion Time External Polishing No sharp roll‐off at the edge of wafer For Lo
7 Recycle Slurry To evaluate impact of Every 5 mins Data capture BMS Database PSE Op For Tig
Status different slurry application
Internal Lapping Comparable yield to old active agent TTV re
8 Active Agent KPIV (can affect slurry Every time fresh Slurry Sheet Lapping Op
Machine Setting (SOP) Good Flatness For Tig
Volume viscosity) slurry prepared
KPIV (can affect wafer Any time changed
9 Sun Gear Ratio Machine Setting (SOP) QA Records Engineer
rotation) by engineer Tab. 5. Data collection plan
ITEM NO. WHAT WHY WHEN
10 KPIV (can affect wafer Any time changed
Bottom Plate Speed Machine Setting (SOP) QA Records Engineer
rotation) by engineer To correlate poor TTV Data ca
1 Wafer Traceability Every Lot
11 Exhaust Timer KPIV (can affect wafer Any time changed Machine Setting (SOP) QA Records with different Engineer applica
rotation) by engineer Lapping
2 TTV KPOV, CTQ Every wafer
12 Acceleration Timer KPIV (can affect wafer Any time changed Machine Setting (SOP) QA Records Engineer upload
rotation) by engineer Every time fresh
Slurry 3 Slurry Mixing Time KPIV Machin
KPIV (can affect slurry Every time fresh slurry prepared
13 Machine Setting (SOP) Slurry Sheet Lapping Op
Temperature viscosity) slurry prepared Every time fresh
Fig. 2. Y = f (x) cascade 4 Slurry Density KPIV Manual
KPIV (can affect slurry Lot Processing slurry prepared
14 Plate Temperature At the start of shift Digital thermometer Lapping Op
viscosity) Sheet Every time loop Manual
5 Machine flowrate KPIV
was changed
KPIV (can affect wafer Every 40 hr of
Process Capability of TTV for Tight Spec Material
Tab. 6. Key results from process capability charts 6 Process Plate Shape
Capability of TTV for Loose Spec Material Using d
shape) Opertion Time
LB
TIGHT FLATNESS LOOSE FLATNESS
USL
7 D ata Recycle Slurry
LB USL To evaluate impact of Every 5 mins Data ca
LB PARAMETERS COMBINED
P rocess D ata Within P rocess Within
0
SPEC SPEC Ov erall LB 0 Status different slurry Ov erall applica
Target * Target *
8 2.11004Active Agent KPIV (can affect slurry Every time fresh

P otential (Within) C apability P otential (Within) C apability
Machin
USL 2 USL 3.5
S ample M ean 1.21547 Cp * S ample M ean Cp *
5,623 28,120 33,743 28120 Volume viscosity) C P L slurry prepared

Wafer Qty
S ample N
S tD ev (Within)
5623
1.11571
C PL
C P U 0.23
* S ample N
S tD ev (Within) 0.78286 C P U 0.59
*
C pk 0.23
9 Sun Gear Ratio KPIV (can affect wafer Any time changed
C pk 0.59
9.82 3.35 O v erall C apability4.43
S tD ev (O v erall) 1.22691
Machin
TTV Reject % rotation)
O v erall C apability
Pp * P p by engineer
*
DPMO 98,168 33,499 PPL

44,282
*
KPIV (can affect wafer P P U Any time changed
PPL *
P P U 0.21
10 Bottom Plate Speed 0.52
Machin
Process Capability rotation) C pmby engineer
P pk 0.21 P pk 0.52
0.21 0.52 C pm *
‐NA‐ *
(Ppk) 11 Exhaust Timer KPIV (can affect wafer Any time changed

Machin
rotation) by engineer
Tab. 7. Response Table for Means -0.0 1.4 2.8 4.2 5.6 7.0 8.4 9.8 0.0 1.3 2.6 3.9 5.2
KPIV (can affect wafer Any time changed
6.5 7.8 9.1
O bserv ed 12
P erformance Acceleration Timer Machin
P P M < rotation) by engineer
O bserv ed P erformance E xp. Within P erformance E xp. O v erall P erformance E xp. Within P erformance E xp. O v erall P erformance
P P M < LB 0.00
P P M > U S L 98168.24SUN
P P M < LB
P P MG >EAR
*
BOTTOM
P P M < LB
PPM > P U SLATE
*
E XHAUST A CCELERATION P
P P M < LB
LATE 0.00
RECYCLE
P P M < LB
P P M >S
Slurry ULURRY
* LB
AP PCTIVE AGENT
*
SLURRY SLURRY
KPIV (can affect slurry Every time fresh
U S L 240975.32 L 261269.05 P P M > U S L 33499.29 S L 37907.98 M > U S L 59120.69
PPML
EVEL98168.24
TEMP13 Machin
Total P P M Total 240975.32 P P M Total 261269.05 P P M Total 33499.29 P P M Total 37907.98 P P M Total 59120.69
RATIO SPEED TIMER TIMER . STATUS

Temperature CONCENTRATION viscosity) TEMP. M IXING TIME
slurry prepared
Fig.1 3. Process capability KPIV (can affect slurry
1.2893 of TTV for1.0338 Tight and1.0053 Loose Spec Material 1.0092 1.1997 14 Plate Temperature
0.9940 1.0120 0.9957 At the start of shift Digital t1.0062
viscosity)
2 0.7123 0.9678 0.9963 0.9925 0.8020 1.0077 0.9897 1.0060 0.9955
Delta 0.5770 0.0660 0.0090 0.0167 0.3977 0.0137 0.0223 0.0103 0.0107
als was calculated using Minitab Software, which is Tab. 6. Key results from process capability charts
Rank 1 3 9 5 2 6 4 8 7
shown

in Fig. 3 and Tab. 6. TIGHT FLATNESS LOOSE FLATNESS
PARAMETERS COMBINED
As shown in Tab. 6, the combined TTV reject
Tab. 8. Displaying the impact of improvements on TTV rejects SPEC SPEC

was very high at 4.43%, with TTV rejects for tight B EFORE I MPROVEMENT A FTER I MPROVEMENT
Wafer Qty 5,623 28,120 33,743
TIGHT FLATNESS LOOSE TIGHT FLATNESS LOOSE
flatness specification PARAMETERS being higher than that of loose COMBINED
TTV Reject % 9.82 3.35 COMBINED 4.43
SPEC FLATNESS SPEC SPEC FLATNESS SPEC
flatnessWafer Qty
specification. For a Six Sigma project
5,623
to be 28,120 DPMO 33,743 98,168
19,677 33,499
50,666 44,282
70,343
successful,
TTV Reject % it should be able to reduce9.82 DPMO to 3.35 Process Capability 4.43 0.02
0.21 0.01
0.52 ‐NA‐ 0.01
1/10th, DPMO which means the combined 98,168 TTV DPMO 33,499 (Ppk) 44,282 152 118 127
Process Capability (Ppk) 0.52 ‐NA‐ 3.87 1.30 ‐NA‐
should be less than 4428. 0.21 Tab. 7. Response Table for Means

Stage 3 — Analyse: during this phase, the col- Although there SUN GEAR are different
BOTTOM PLATE kindsEXHAUST of control
ACCELERATION charts, PLATE REC

lected data was analysed through a systematic appli- LEVEL
in the current RATIO study,
Xbar-SSPEED and I-MR TIMER charts TIMER were TEMP.
cation of statistical and graphical tools. used together 1.2893
with boxplots. The further investiga-
1 1.0338 1.0053 1.0092 1.1997
Control charts, also known as Shewhart charts, tion into out-of-control lots gave more clues, from
2 0.7123 0.9678 0.9963 0.9925 0.8020
were used to identify key trends and generate clues. which
Delta a number of Multi-Vari
0.5770 charts0.0090
0.0660 were generated
0.0167 0.3977
Rank 1 3 9 5 2

95
Tab. 8. Displaying the impact of improvements on TTV rejects
BEFORE IMPROVEMENT
but only two charts displayed a significant trend (Fig. • Generally, the wafers with Line Saw Mark (LSM)
4). reject have a higher Avg and Std Dev TTV than
Avg TTV refers to the mean of TTV for a lot; and the wafers without LSM reject. This is an impor-
it should be noted that a lot can have a wafer quantity tant finding as it indicates that due to the poor
from 100 to 320. Then, this mean of Avg TTV is fur- quality of an incoming wafer, the wafer was una-
ther split by three factors: Lapping Machines, lots ble to rotate freely at lapping. Wafer rotation
with Line Saw Mark (LSM) Rejects and Run Order of could be affected by factors like sun gear ratio,
lots during the shift. The right-hand side graph is plate speed, acceleration time and exhaust time;
exactly the same, except for Std Dev of TTV on the • Generally, lapping machines have no significant
Y-Axis. In summary, the following inferences could impact on Avg and Std Dev TTV. It means that
be extracted from the Multi-Vari charts (Fig. 4): the problem is global and related to something
• The green trend line in Fig. 4 shows that gener- that was common to all lapping machines, like
ally, Avg TTV for the first lot of the shift was slurry composition, slurry type etc.
comparatively higher than the rest of the lots Stage 4 — the Improve phase: based on enhanced
processed during the same shift. It means that learning gained from the Analyse phase, a list of fac-
something at the start of the shift was not correct, tors that can influence TTV, was generated by the
which was resulting in high Avg TTV. Factors team using the brainstorming technique. To charac-
that were different at the start of the shift and got terise the impact of these input variables on TTV,
stabilised during the shift were the slurry tem- Design of Experiments were used. As part of DoE
perature, slurry mixing time and plate tempera- planning, a CNX diagram was generated, as shown in
ture; Fig. 5.
Multi-Vari Chart for Avg TTV by LSM Reject - Run Order Multi-Vari Chart for Std Dev TTV by LSM Reject - Run Order
LA P -001 LA P -002 LA P -003 LA P -004
LA P -001 LA P -002 LA P -003 LA P -004
1st lot 2nd lot 3rd lot LSM
Multi-Vari Chart for Avg TTV by LSM Reject - Run Order
3.25 Multi-Vari Chart for 1.4
Reject Std Dev TTV by LSM Reject - Run Order Reject
No
LA P -001 LA P -002 LA P -003 LA P -004 No
3.00 Yes LA P -001 LA P -002 LA P -003 LA P -004
1.2 Yes
3.25 Reject 1.4
2.75 Reject
Std Dev TTV
No 1.0 No
3.00 Yes
Avg TTV
1.2 Yes
2.50
0.8
2.75
Std Dev TTV
2.25 1.0
0.6
Avg TTV
2.50
2.00 0.8
0.4
2.25
1.75 0.6
0.2
2.00
LA P -001 LA P -002 LA P -003 LA P -004 LA P -001 LA P -002 LA P -003 LA P -004
1.50 0.4
Machine
1.75
Machine
0.2
1.50 Panel variable: Run Order Panel
LA P -001 LA P -002 LA P -003 LA P -004 variable: Run Order LA P -001 LA P -002 LA P -003 LA P -004

Machine
Machine
Panel variable: Run Order Fig. 4. Multi-Vari charts for Avg and Std Dev TTV byvariable:
Panel LSM reject
Run Order— Run order
Fig. 4. Multi-Vari charts for Avg and Std Dev TTV by LSM reject — Run order
Fig. 5. CNX Diagram for DOE
Fig. 5. CNX Diagram for DOE

96
Fig.5.5.CNX
Fig. CNXDiagram
Diagramfor
forDOE
DOE
5 Machine flowrate KPIV Every time loop Manual Measurement Control Charts Lapping Op
was changed
6 Plate Shape
KPIV (can affect wafer Every 40 hr of Engineering Management
Using dial gauge in Production
Plate Shape Sheet and Services
Lapping Op
Volume 11 • Issue 2 • 2019 shape) Opertion Time
7 Recycle Slurry To evaluate impact of Every 5 mins Data capture BMS Database PSE Op
Status different slurry application
8 Active Agent KPIV (can affect slurry Every time fresh
MainEffects
Main EffectsPlot PlotMachine Setting (SOP)
forMeans
for Means Slurry Sheet Lapping Op
Volume viscosity) slurry prepared
DataMeans
Data Means
9 Sun Gear Ratio KPIV (can affect wafer Any time changed QA Records Engineer
Machine Setting (SOP)
rotation)
SSun
unGGear
earRatio
Ratio by engineer BottomPPlate
Bottom lateSSpeed
peed EExhaust
xhaustTimer
Timer
10 1.2 KPIV (can affect wafer Any time changed

1.2
Bottom Plate Speed Machine Setting (SOP) QA Records Engineer
1.0
1.0
11 Exhaust Timer KPIV (can affect wafer Any time changed
Machine Setting (SOP) QA Records Engineer
0.8
0.8 rotation) by engineer
12 Acceleration Timer KPIV (can affect wafer
0.43
0.43 0.70
0.70
Any time changed
25
25 Machine Setting (SOP)
35
35 11
QA Records 55 Engineer
Means
of Means
Acceleration
ccelerationTimer
Timer Plate
lateTemperature
Temperature Recycle
cleSSlurry
lurrySStatus
tatus
Slurry A P
KPIV (can affect slurry Every time fresh Recy
13 Machine Setting (SOP) Slurry Sheet Lapping Op
Temperature 1.2
1.2 viscosity) slurry prepared
KPIV (can affect slurry Lot Processing
Mean of
14 Plate Temperature
1.0
1.0 At the start of shift Digital thermometer Lapping Op
viscosity) Sheet
Mean
0.8
0.8

11 33 18
18 3030 OOnn OOffff
Tab. 6. Key results from process capability charts
A ctiv e A gent C oncentration
A ctiv e A gent C oncentration SSlurry
lurryTemperature
Temperature SSlurry
lurryMMixing
ixingTime
Time
1.2FLATNESS LOOSE FLATNESS
TIGHT
1.2
PARAMETERS COMBINED
SPEC SPEC
1.0
1.0
Wafer Qty 0.8
0.8 5,623 28,120 33,743
TTV Reject % 9.82
0.36
3.35
1.00
4.43 25 32
0.36 1.00 25 32 11 10
10
DPMO 98,168 33,499 44,282
Process Capability
Fig.6.6.Main
Fig. 0.21
Maineffect
effect plotsfor
plots 0.52
formeans
means ‐NA‐
(Ppk)

Tab. 7. Response Table for Means
SUN GEAR BOTTOM PLATE EXHAUST ACCELERATION PLATE RECYCLE SLURRY ACTIVE AGENT SLURRY SLURRY
LEVEL
RATIO SPEED TIMER TIMER TEMP. STATUS CONCENTRATION TEMP. MIXING TIME
1 1.2893 1.0338 1.0053 1.0092 1.1997 0.9940 1.0120 0.9957 1.0062

2 0.7123 0.9678 0.9963 0.9925 0.8020 1.0077 0.9897 1.0060 0.9955
Delta 0.5770 0.0660 0.0090 0.0167 0.3977 0.0137 0.0223 0.0103 0.0107
Rank 1 3 9 5 2 6 7 4 8

The list of
experimental variablesBEFORE was Istill too
MPROVEMENT
was followed by factorAFTER B and an interaction effect
IMPROVEMENT
long, and only screening DOE was feasible.
TIGHT FLATNESS Taguchi
LOOSE between factors A and B.
TIGHT FLATNESS The LOOSE chart was also
Pareto
PARAMETERS COMBINED COMBINED
L12 design was used to rank theSPEC factors
in Fthe order
LATNESS SPEC used to determine statistically
SPEC significant
FLATNESS SPEC factors and
ofWafer Qty
their impact on TTV. Taguchi L125,623 implied that by 28,120 the effect of their interactions
33,743 19,677 on the output. The
50,666 fac-
70,343
TTV Reject %
performing 12 lapping batches, nine experimental 9.82 3.35 tors above the red line were significant, and the0.01
4.43 0.02 0.01 fac-
DPMO 98,168 33,499 44,282 152 118 127
factors at two different values were evaluated for their0.52 tors below
Process Capability (Ppk) ‐NA‐
were non-significant.
3.87 1.30 ‐NA‐
0.21
impact on TTV, as shown in Fig. 6 and Tab. 7. T main effects plot (Fig. 8) indicated that an
From Taguchi DOE, the top two ranked factors increase in the sun gear ratio, plate temperature and
(the sun gear ratio and plate temperature) seemed to the bottom plate speed resulted in reduced TTV.
have a significant impact on TTV. Although the fac- Further, as centre points (marked red ) for all the
tor ranked third (the bottom plate speed) did not three factors were not on the line, the impact of these
seem significant, it was still chosen for further analy- factors on TTV was not linear. In simple terms, the
sis. As Taguchi is only a screening DOE, it is always increase in the plate temperature to 24 °C did not
recommended to perform a more comprehensive result in any reduction in TTV; however, a plate
DOE to confirm the results of any screening DOE. So, temperature closer to 30 °C was most likely to reduce
three factors of the second level of the full factorial the TTV reject rate.
design of experiment (DOE) were designed with To make things simple, Minitab’s response opti-
three centre points to characterise and optimise the miser function could be used, which calculated values
three experimental factors. for input variables to achieve the optimum output.
DoE results in the Pareto chart represent the rela- Results are shown in Fig. 9, suggesting that to achieve
tive impact of differences between different factors on the minimum TTV (0.462), inputs should be set to
the process outcome. As shown in Fig. 7, factor A the following parameters (highlighted in red):
(sun gear ratio) had the biggest impact on TTV. This • Sun Gear Ratio = 0.7;
97
Normal Plot of the Standardized

Normal Plot
Effects
of the Standardized Effects Pareto Chart of the Standardized
Pareto Chart
Effects
of the Standardized Effects
(response is Avg TTV, Alpha =(response
0.05) is Avg TTV, Alpha = 0.05) (response is Avg TTV, Alpha =(response
0.05) is Avg TTV, Alpha = 0.05)
99 99 2.20 2.20
Effect Ty pe Effect Ty pe
F actor N ame F actor N ame
Not Significant Not Significant A S un G ear Ratio A S un G ear Ratio
95 95 Significant ASignificant A B P late Temperature B P late Temperature
F actor N ame F actor N ame C Bottom P late S peed C Bottom P late S peed
90 90 AB AB
A S un G ear Ratio A S un G ear Ratio
80 80 B P late Temperature B B P late Temperature B
AC C Bottom
A C P late S peed C Bottom P late S peed
70 70
AB AB
Percent
Percent
60 60
Term
Term
50 50
40 40 C C
30 30 C C
B B
20 20
A A
AC AC
10 10
5 5
BC BC
1 1
-60 -50 -40 -30 -20 -60
-10 -500 -40
10 -30
20 -20
30 -10 0 10 20 30 0 10 20 30 0 4010 5020 6030 40 50 60
Standardized Effect Standardized Effect Standardized Effect Standardized Effect
Fig. 7. Normal and standardised

Fig. 7. Normal
effects
and standardised
for Pareto charts
effects for Pareto charts
Main Effects Plot for Avg

Main
TTVEffects Plot for Avg TTV Interaction Plot for AvgInteraction
TTV Plot for Avg TTV
Data Means Data Means Data Means Data Means
Sun Gear Ratio Sun
Plate
Gear
Temperature
Ratio Plate Temperature
Point Ty pe Point Ty pe18 24 30 25 30 18 35 24 30 25 30 35
1.08 1.08 1.5 1.5
Corner Corner Sun Gear Sun Gear
Center Center Ratio Point Ty pe Ratio Point Ty pe
0.96 0.96
0.430 Corner 0.430 Corner
1.0 1.0
Sun Gear Ratio Sun Gear Ratio 0.565 Center 0.565 Center
0.84 0.84
0.700 Corner 0.700 Corner
0.72 0.72
0.5 0.5
1.5 1.5
0.60 0.60 Plate Plate
Mean
Mean
Temperature Point Ty pe Temperature Point Ty pe

0.430 0.565 0.700 0.430
18 0.56524 0.700
30 18 24 30 18 Corner 18 Corner
Bottom Plate Speed Bottom Plate Speed 1.0 1.0
P late T emper atur e P late T emper atur e 24 Center 24 Center
1.08 1.08 30 Corner 30 Corner
0.96 0.96 0.5 0.5
0.84 0.84
0.72 0.72
Bottom P late Speed Bottom P late Speed
0.60 0.60
25 30 35 25 30 35
Fig. 8. Main effects and

Fig.interaction
8. Main effects
plotsand
for interaction
avg TTV plots for avg TTV
Optimal Sun Gear

Optimal Sun Gear
Plate Te Plate Bottom
Te P Bottom P
D High D0.70 High 0.70 30.0 30.0 35.0 35.0
Cur 0.70 Cur 0.70 30.0 30.0 35.0 35.0
0.76000 Low 0.76000
0.430 Low 0.430 18.0 18.0 25.0 25.0
Composite Composite
Desirability Desirability
0.76000 0.76000
Avg TTV Avg TTV

Minimum Minimum
y = 0.4620 y = 0.4620
d = 0.76000 d = 0.76000
Fig. 9. Response optimiser

Fig. 9. result
Response optimiser result
• Plate Temperature = 30 °C; Using the recommended input values, Response

• Bottom Plate Speed = 35 rpm. Optimiser predicted Avg TTV to be around 0.462
To validate the theoretical predictive DOE model, μm, which is exactly the same as the confirmation
a confirmation run should be conducted using the run result of 0.460 μm (see Fig. 10). The confirmation
recommended input values. A confirmation run was test, thus, confirmed the validity of the theoretical
carried out with six lapping batches, and the results model.
are presented in Fig. 10.
98
Summary for Confimation Test TTV
A nderson-Darling N ormality Test
A -S quared 0.50
Summary for Confimation
P -V alue Test0.207
TTV
M ean 0.46084 A nderson-Darling N ormality Test
S tDev 0.10751
A -S quared 0.50
V ariance 0.01156
P -V alue 0.207
S kew ness -0.011959
Kurtosis -0.157437 M ean 0.46084
N 210 S tDev 0.10751
V ariance 0.01156
M inimum 0.17696
S kew ness -0.011959
1st Q uartile 0.38360
Kurtosis -0.157437
M edian 0.46912
N 210
3rd Q uartile 0.53777
0.2 0.3 0.4 0.5 0.6 0.7 M aximum M inimum 0.17696
0.74813
95% C onfidence Interv al for M ean 1st Q uartile 0.38360
M edian 0.46912
0.44622 0.47547
3rd Q uartile 0.53777
0.2 0.3 0.4 0.5 0.695% C onfidence
0.7 Interv al for M edian M aximum 0.74813
0.43928 0.49011 95% C onfidence Interv al for M ean
95% C onfidence Interv al for S tDev 0.44622 0.47547
9 5 % C onfidence Inter vals
0.09812 0.11891 95% C onfidence Interv al for M edian
Mean 0.43928 0.49011
95% C onfidence Interv al for S tDev
Median 9 5 % C onfidence Inter vals
0.09812 0.11891
0.44 0.45 Mean 0.46 0.47 0.48 0.49
Median
Fig. 10. Graphical Summary0.44

of Confirmation
0.45
Test Run0.47Results0.48
0.46 0.49
Fig. 10. Graphical Summary of Confirmation Test Run Results
Boxplot of TTV for Tight Flatness Spec by Improvement Stages

8
Before Improvement PilotTight
Boxplot of TTV for Stage -Flatness
After Improvement
Spec by Improvement Stages
7 8
Before Improvement Pilot Stage - After Improvement
6 7
5 6
4
TTV
5
3 4
TTV
2 3 2
1 2 2
0 1
6 5 1 6 0 4 3 2 9 7 8 5 5 4 4 6 6 5 0 1 8 2 5 5 9 3 3 0 1 9 1 5 6 3 1 7 1 2 1 3 1 6 5 3 1 7 9 1
: 5 : 1 :5 :2 :5 :1 :3 : 3 : 1 : 4 : 2 :5 :4 :17:29:1 6: 19: 30: 09:28:37:40:48:2 7: 56: 10: 47: 04:1 :0 :2 :4 :2 : 4 : 5 : 2 : 0 :2 :3 :4 :32:1 8: 08: 18: 44:07:49:14:39:4 0: 28: 46: 16: 55:41:4 :3 :3 :0 : 2 : 1 : 2 :1 :2 :4 :0 :0 : 50: 39: 04: 37:18:50:49:58:2 1: 50: 08: 58: 29:51:09:1 :2 :5 : 2 : 4 : 1 :2 :4 :0 :1 :0
03 19 20 2 2 23 01 02 04 06 04 06 09 1 2 4 0 3 2 2 5 5 6 9 1 3 3 3 3 0 04 0 5 23 00 03 03 04 23 00 02 0 2 12 5 7 0 0 3 5 6 7 1 3 4 5 7 2 3 0 0 02 23 02 03 04 21 23 0 0 02 03 04 4 3 0 2 4 4 3 3 3 3 5 7 3 3 3 04 04 23 03 03 09 12 1 3 17 02 00
2 2 2 2 2 2 2 2 2 2 2 2 2 2 12 22 22 02 02 02 02 02 02 12 22 22 02 02 02 2 2 2 2 2 2 2 2 2 2 2 2 12 12 02 02 02 02 12 12 22 22 12 12 12 22 22 2 2 2 2 2 2 2 2 2 2 2 2 02 22 02 02 02 02 22 22 02 02 02 12 22 22 02 2 2 2 2 2 2 2 2 2 2
1
0
01 0 1 01 01 01 01 01 01 01 01 0 1 01 01 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 01 01 01 01 01 01 01 01 0 1 01 01 01 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 01 01 01 01 01 01 01 0 1 01 01 01 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 01 01 01 01 01 0 1 01 01 01 01 01
2 2 2 2 2 2 2 2 2 2 2 2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 02 2 2 2 2 2 2 2 2 2 2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 02 2 2 2 2 2 2 2 2 2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 2 2 2 2 2 2
0/12/12/12/12/13/13/13 /1 3/14/14/14/14/14/124/124/125 /125/215/215/215/215/125/023/023 /024/204/206/206/06/07/08/08/08 /0 8/09/10/10/10/13/13/123 /125/125/215/215/215/125/125/125/128 /128/218/218/218/219/19/23/24/24 /2 4/25/25/26/26/26/26/226 /226/227/227/227/227/320/320/321/321 /321/231/207/207/208/028/028/12 /1 3/1 3/17/17/17/17/20/22/
3/ 3/ 3/ 3/ 3 / 3/ 3/ 3/ 3/ 3/ 3/ 3/ 3/ 3 / 3/ 3/ 3/ 3/ 3/ 3/ 3/ 3 / 3/ 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 /5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 6/ 6/ 6 / 6 / 6/ 6/ 6/ 6/ 6/ 6/ 6/ 6 / 6/ 6/
6 5 1 6 0 4 3 2 9 7 8 5 5 4 4 6 6 5 0 1 8 2 5 5 9 3 3 0 1 9 1 5 6 3 1 7 1 2 1 3 1 6 5 3 1 7 9 1
: 5 : 1 :5 :2 :5 :1 :3 : 3 : 1 : 4 : 2 :5 :4 :17:29:1 6: 19: 30: 09:28:37:40:48:2 7: 56: 10: 47: 04:1 :0 :2 :4 :2 : 4 : 5 : 2 : 0 :2 :3 :4 :32:1 8: 08: 18: 44:07:49:14:39:4 0: 28: 46: 16: 55:41:4 :3 :3 :0 : 2 : 1 : 2 :1 :2 :4 :0 :0 : 50: 39: 04: 37:18:50:49:58:2 1: 50: 08: 58: 29:51:09:1 :2 :5 : 2 : 4 : 1 :2 :4 :0 :1 :0
Lapping Date Out
03 19 20 2 2 23 01 02 04 06 04 06 09 1 2 4 0 3 2 2 5 5 6 9 1 3 3 3 3 0 04 0 5 23 00 03 03 04 23 00 02 0 2 12 5 7 0 0 3 5 6 7 1 3 4 5 7 2 3 0 0 02 23 02 03 04 21 23 0 0 02 03 04 4 3 0 2 4 4 3 3 3 3 5 7 3 3 3 04 04 23 03 03 09 12 1 3 17 02 00
2 2 2 2 2 2 2 2 2 2 2 2 2 2 12 22 22 02 02 02 02 02 02 12 22 22 02 02 02 2 2 2 2 2 2 2 2 2 2 2 2 12 12 02 02 02 02 12 12 22 22 12 12 12 22 22 2 2 2 2 2 2 2 2 2 2 2 2 02 22 02 02 02 02 22 22 02 02 02 12 22 22 02 2 2 2 2 2 2 2 2 2 2
01 0 1 01 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 01 0 1 01 01 01 01 01 01 01 0 1 01 01 01 01 01
0 /22/22/22/ 22/ 23/ 23/ 23 /23/24/24/24/24/ 24/ 24/ 24/ 25 /25/25/25/25/25/ 25/ 23/ 23 /24/24/26/26/26/ 27/ 28/ 28/ 28 /28/29/20/20/20/ 23/ 23/ 23 /25/25/25/25/25/ 25/ 25/ 25/ 28 /28/28/28/28/29/ 29/ 23/ 24/ 24 /24/25/25/26/26/ 26/ 26/ 26 /26/27/27/27/27/ 20/ 20/ 21/ 21 /21/21/27/27/28/ 28/ 28/ 22 /23/23/27/27/27/ 27/ 20/ 22/ 2
1 1 1 1 1 1 1 1 1 1 1 1 11 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 11 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 22 2 2 2 2 2 2 3 3 3 3 3 3 0 0 0 0 0 1 1 1 1 1 1 1 2 2
3/ 3/ 3/ 3/ 3 / 3/ 3/ 3/ 3/ 3/ 3/ 3/ 3/ 3 / 3/ 3/ 3/ 3/ 3/ 3/ 3/ 3 / 3/ 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 /5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5/ 5 / 5/ 5/ 5/ 5/ 5/ 6/ 6/ 6 / 6 / 6/ 6/ 6/ 6/ 6/ 6/ 6/ 6 / 6/ 6/
Lapping Date Out

Fig. 11. Boxplot of TTV for tight flatness spec by improvement stages
Pilot testing was usedFig. for11.the Boxplot of TTV for

validation oftight
theflatnessTTVspechadby improvement stages and few wafers were above
become stable;
results found during the improvement
Process Capability stage Improvement
of TTV for Tight Spec_After by the spec limit. This
Process improvement
Capability of TTV for Loosewas statistically
Spec_After Improvement
implementing it on larger sample size. The incremen-

P rocess D ata
LB significant (W=1367172420.0 and p< 0.01). This data
USL LB USL
Within P rocess D ata Within
Process Capability of TTV for Tight Spec_After Improvement Process Capability of TTV for Loose Spec_After Impro
tal implementation with a continuous review of
LB
Target
0
* provided sufficient evidence in support of the Ov erall
P otential (Within) C apability
LB
Target
0
*
Ov erall
P otential (Within) C apability

USL 2 USL 3.5
results was conducted to reduce the risk of large

S ample M ean
S ample N
0.351407
19677
improved condition for loose spec material. These
P rocess D ata
LB USL Cp
C PL
*
*
S ample M ean
S ample N
1.68514
Within 50666 P rocess D ata
LB Cp
C PL
*
*
USL
LB 0 C P U 4.39 Ov erall 0.371802 LB 0 C P U 1.63
rejects. Loose TTV spec materials were run on one conditions were then transferred to tight spec mate-
S tD ev (Within) 0.125247 S tD ev (Within)
S tD ev (O v erall) 0.141981 Target * C pk 4.39 S tD ev (O v erall) 0.464149 Target * C pk 1.63
USL 2 P otential (Within) C apability USL 3.5 P oten
O v erall C apability O v erall C apability
Cp *
lapping machine for a limited period of one week, rial, and the improvement in TTV was again statisti-
S ample M ean 0.351407 S ample M ean 1.68514
Pp * C PL * Pp *
S ample N 19677 S ample N 50666
PPL * C P U 4.39 PPL *
S tD ev (Within) 0.125247 S tD ev (Within) 0.371802
PPU 3.87 PPU 1.30
and then, the results were awaited. Since results from cally significant according to the Mann Whitney test
S tD ev (O v erall) 0.141981 C pk 4.39 S tD ev (O v erall) 0.464149
P pk 3.87 P pk 1.30
C pm * C pm *
Pp *
this initial phase were as expected, this optimum (W=124139220.0 and p= 0.00). PPL
PPU
*
3.87
condition was expanded-0.00 to 0.65

other1.30 machines, one4.55 at a The Establishing0.00Process
0.55 1.10 capability and3.85DPMO
P pk 3.87
C pm *
1.95 2.60 3.25 3.90 1.65 2.20 2.75 3.30
time. Boxplot results from the pilot stage for loose

O bserv ed P erformance
P P M < LB 0.00 P P M < LB were the next step, aiming to quantify and validate
E xp. Within P erformance
*
E xp. O v erall P erformance
P P M < LB *
O bserv ed P erformance
P P M < LB 0.00
P P M > U S L 118.42
E xp. Within P erformance
P P M < LB
P P M > U S L 0.53
*
E xp. O v erall P erformance
P P M < LB
P P M > U S L 46.13
*
P P M > U S L 152.46 P P M > U S L 0.00 P P M > U S L 0.00
spec material are shown in Fig. 11.

P P M Total 152.46 P P M Total
the new improved results. At the start
-0.00 0.65 1.30 1.95 2.60 3.25 3.90 4.55
0.00 P P M Total 0.00
of the process,
0.00 0.55 1.10 1.65 2.20 2.75 3.30 3.85 P P M Total 118.42 P P M Total 0.53 P P M Total 46.13
O bserv ed P erformance E xp. Within P erformance E xp. O v erall P erformance O bserv ed P erformance E xp. Within P erformance E xp. O v erall P erformance
There were rather a few wafers with TTV > 3.5 the
Fig. 12. Process capability of TTV for tight and loose spec after improvement
baseline
P P M < LB
for
0.00
P P M > U S L 152.46
the lapping process was
P P M < LB
established
P P M > U S L 0.00
* P P M < LB
P P M > U S L 0.00
* P P M < LB 0.00
P P M > U S L 118.42
P P M < LB
P P M > U S L 0.53
* P P M < LB
P P M > U S L 46.13
*
P P M Total 152.46 P P M Total 0.00 P P M Total 0.00 P P M Total 118.42 P P M Total 0.53 P P M Total 46.13
microns (spec limit) before improvement. However, using DPMO, which was now repeated for the
a noticeable difference wasFig. visible after capability
12. Process the introduc- improved
of TTV for tight and looseprocess.
spec afterThe process capability charts are
improvement
tion of the improved condition. It was found that
99

Tab. 6. Key results from process capability charts
Fig. 11. Boxplot of TTV for tight flatness spec by improvement stages
TIGHT FLATNESS LOOSE FLATNESS Volume 11 • Issue 2 • 2019
PARAMETERS COMBINED
SPEC SPEC
Wafer Qty Process Capability of5,623 28,120 Improvement

TTV for Tight Spec_After 33,743 Process Capability of TTV for Loose Spec_After Improvement
TTV Reject % LB 9.82
USL 3.35 4.43 LB USL
P rocess D ata Within P rocess D ata Within
DPMO LBTarget 0
*
98,168 33,499 44,282
Ov erall LB
Target
0
*
Ov erall
P otential (Within) C apability P otential (Within) C apability

Process Capability
USL 2 USL 3.5
0.21 0.52 ‐NA‐

S ample M ean 0.351407 Cp * S ample M ean 1.68514 Cp *
(Ppk) SS ample N 19677 C PL * S ample N 50666 C PL *

tD ev (Within) 0.125247 C P U 4.39 S tD ev (Within) 0.371802 C P U 1.63
C pk 4.39 C pk 1.63
Tab. 7. Response Table for Means Pp
PPL
*
*
Pp
PPL
*
*
PPU 3.87 PPU 1.30
SUN GEAR BOTTOM PLATE EXHAUST ACCELERATION P pk 3.87

PLATE RECYCLE SLURRY ACTIVE AGENT SLURRY SLURRY
P pk 1.30
LEVEL C pm * C pm *
RATIO SPEED TIMER TIMER TEMP. STATUS CONCENTRATION TEMP. MIXING TIME
1 1.2893
-0.00 0.65 1.0338
1.30 1.95 2.60 3.25 3.90 4.55 1.0053 1.0092 1.1997 0.9940
0.00 0.55 1.10 1.65 2.20 2.75 3.30 3.851.0120 0.9957 1.0062
O bserv ed P erformance E xp. Within P erformance E xp. O v erall P erformance O bserv ed P erformance E xp. Within P erformance E xp. O v erall P erformance
2 P P M < LB 0.00
P P M > U S L 152.46
0.7123
P P M < LB
P P M > U S L 0.00
* 0.9678
P P M < LB
P P M > U S L 0.00
*0.9963 0.9925 0.8020 P P M < LB 0.00
P P M > U S L 118.42
1.0077
P P M < LB
P P M > U S L 0.53
* 0.9897 1.0060
P P M < LB
P P M > U S L 46.13
* 0.9955
Delta 0.5770 0.0660 0.0090 0.0167 0.3977 0.0137 0.0223 0.0103 0.0107
P P M Total 152.46 P P M Total 0.00 P P M Total 0.00 P P M Total 118.42 P P M Total 0.53 P P M Total 46.13
Rank 1
Fig. 12. Process capability 9
of TTV for3 tight and loose 5 2
spec after improvement 7 6 4 8

BEFORE IMPROVEMENT AFTER IMPROVEMENT
TIGHT FLATNESS LOOSE TIGHT FLATNESS LOOSE
PARAMETERS COMBINED COMBINED
SPEC FLATNESS SPEC SPEC FLATNESS SPEC
Wafer Qty 5,623 28,120 33,743 19,677 50,666 70,343
TTV Reject % 9.82 3.35 4.43 0.02 0.01 0.01
DPMO 98,168 33,499 44,282 152 118 127
Process Capability (Ppk) 0.21 0.52 ‐NA‐ 3.87 1.30 ‐NA‐

shown in Fig. 12. The key results from Fig. 12 and method was exercised on the lapping machine, which
from the start of the project are summarised in Tab. 8. would automatically prevent the machine from run-
As shown in Tab. 8, there was a massive reduc- ning if any of these two parameters were set incor-
tion in TTV rejects from 9.82% to 0.02% for tight rectly. This was achieved using continuous, automatic
flatness spec; from 3.35% to 0.01% for loose flatness measurement of these parameters by the machine,
spec; and from 4.43% to 0.01% for the combined and as soon as incorrect values were identified, an
materials. This is also reflected in the DPMO values. alarm went-off and the machine stopped running.
The initial target of the project was to reduce TTV The Statistical Process Control (SPC) Chart: the
rejects to less than 0.1%; and after the improvements, underlying principle of Six Sigma is to reduce the
TTV reject for both specs was found to be less than process output variation by controlling key process
0.03%. Process capability charts clearly demonstrated input variables (KPIV’s), which is done with the help
that the mean and variation for loose spec material of SPC charts. As discussed earlier, the second most
was comparatively higher than that of the tight spec critical parameter for good TTV is the plate tempera-
material. Loose spec material went through other ture, where poka yoke cannot be applied. However, it
manufacturing processes that increased its TTV is imperative to control it using SPC charts. The most
value, whereas tight spec material went straight to the appropriate SPC chart for this purpose is the Indi-
TTV measurement stage. Hence, a true reflection of vidual – Moving Range (I-MR) control chart.
the lapping process capability could be evaluated by Statistical Quality Control Chart is used to con-
tight flatness spec material only. The process capabil- trol the key process output variables (KPOV). In this
ity index increased from 0.21 to 3.87, suggesting project, the KPOV was TTV, which could be con-
a significant improvement in the output quality of the trolled using the Mean and Std. Deviation (Xbar-S)
process, as well as sustainable long-term stability of chart.
the lapping process, in terms of controlling TTV of Control Plan is the master document, which lists
wafers. all the parameters of the process that should be
Stage 5 — the Control phase: poka yoke. This monitored and controlled to achieve the desired
term, commonly referred to as “mistake proofing”, results. It is of utmost importance to communicate
originated from Japanese words “poka” meaning the control procedure amongst various stakeholders.
“avoid” and “yokeru” meaning “mistakes”. In the cur- Maintaining timely updates to the lapping process
rent scenario, the sun gear ratio and the bottom plate control plan was also important. Proper control and
speed were identified as two key input variables for vigilance mechanisms for the key input parameters
TTV control in the lapping process. So, the poka yoke impacting the identified output and placing adequate
100
controls for achieving quantifiable improvements ambiguity or confusion. Following which, a detailed
were crucial. breakdown of the components of the problem facili-
tated a productive brainstorming session, resulting in
a list of possible causes and solutions. For example,
4. Discussion of the results initially, a long list of over 60 potential causes was
generated. It was practically impossible to evaluate all
The current implementation of the Six Sigma of them at the same time. However, it did not take
methodology to reduce flatness defects in lapping very long for the Six Sigma experts in the team to
was successful. The silicon wafer flatness was found to suggest solutions for dealing with such situations,
be dependent on several and their interaction. This which was only possible due to a clear and detailed
project presented an interesting revelation regarding written statement of the problem. Six Sigma tools of
the key input variables that can affect lapping TTV. project framing, cause and effect diagram and Y=f(x)
Based on the controlled design of experiments, it can cascade helped to shorten the long list of identified
be concluded that the sun gear ratio and the plate potential causes, thus making it easier to reduce the
temperature were the two most critical input variables scope of the project to a manageable extent and pro-
affecting TTV. The interaction between these two viding the necessary focus to the evaluation.
factors was also statistically significant. The sun gear A seamless, systematic roadmap of statistical
ratio of 0.7 and the plate temperature of 30 ºC were thinking was ensured. Six Sigma is a step-by-step
the optimum setting to minimise TTV. However, care procedure for problem solving or process improve-
must be taken as these were optimum settings for one ment. Previously, at this organisation, all the efforts to
manufacturing plant location and were completely identify the root cause for high TTV rejects had
different from other locations using a similar manu- failed. Whereas, with the help of Six Sigma methodol-
facturing set-up. ogy, wherein every step was clearly laid out in suffi-
The bottom plate speed was a statistically signifi- cient detail, the problem-solving process was much
cant input factor. However, its impact on TTV was more straightforward. Whenever the project met
not as high as the other two significant factors (the a roadblock, a range of statistical tools was available
sun gear ratio and the plate temperature). It was to facilitate the problem-solving process. For example,
found that the bottom plate speed should be opti- after using several statistical and graphical analysis
mised at 35 rpm. The wafer rotation was critical for techniques, such as Multi-Vari charts, there were still
achieving the optimum TTV: Lapping is a closed nine potential factors whose impact on wafer flatness
process, which makes it difficult to see what is going was unknown. At that stage, Taguchi screening DOE
inside the machine. As per the results of the test, helped reveal that causal relationship. The learning
which were discussed earlier, the increasing sun gear point here is that since there are a wide number of
ratio and the bottom plate speed helped with wafer tools and techniques listed at every step of the five
rotation. Further, the increasing plate temperature phases of the Six Sigma process, at least one of them
reduced the viscosity of lapping slurry, thereby aiding is likely to work for the concerned process.
the wafer rotation as well. This suggested that if sili- Statistical tools were used to validate the engi-
con wafer does not receive enough rotation in the neering models. Statistical knowledge forms the core
lapping carrier, it can result in uneven lapping of the Six Sigma methodology, which is yet another
removal, thereby high TTV. important learning point gained from the current
Detailed scoping of the problem at early stages project. The implementation of the Six Sigma meth-
helped to generate the following solutions. As part of odology in the current project can be argued to be
the Six Sigma approach, a clear statement of the a creative interplay between three components of
problem was drafted at the start of the project. Con- managerial objectives, the expansion of engineering
sistent with evidence of previous research (Mishra, knowledge and the application of statistical tech-
2018; Pyzdek, 2003), the scoping exercise largely niques. Statistical analyses were used to validate the
consisted of ensuring that team members understand theoretical engineering models.
the nature of the problem and what is expected of The full factorial DOE technique gives the Six
them. A clear and shared understanding of the prob- Sigma methodology an edge over traditional prob-
lem and the process at the early stages enabled work- lem-solving techniques. Full factorial DOE tool,
ing towards a common goal and channelling of team which also belongs to Six-Sigma, was particularly
members’ efforts in the right direction, avoiding any useful in identifying key factors that were responsible
101
for causing TTV rejects. Previous studies have also original state. Nevertheless, another measure of long-
shown that the application of Taguchi and full facto- term stability of an improved process achieved
rial DOE technique helps to identify sources of varia- through the implementation of Six Sigma methodolo-
tion in the process (Ghosh & Rao, 1996). Previous gies is the process capability (Ppk) index. The Ppk
attempts at this organisation involved the use of One index of the current process was found to be 3.87,
Factor at a Time (OFAT) technique. This technique which indicates that a sustainable solution to the
had repeatedly failed to identify the underlying cause problem has been identified and implemented. It also
in the current scenario. In hindsight, it is now logical ensures that correct factors responsible for poor wafer
to understand why traditional problem-solving tech- flatness have been identified and that adequate con-
niques had failed. In the current project, an interac- trol has been exerted to maintain a long-term process
tion between more than one factors (plate temperature stability. Any decline in the process capability or
and sun gear ratio) was responsible for high TTV quality over long term should, therefore, be unlikely.
rejects. Such an evaluation of the interaction between
more than one input factor would have been impos-
sible to perform through traditional problem-solving
Conclusions
techniques, such as OFAT. The use of Full factorial
DOE technique revealed not only the impact of all This study aimed to reduce flatness rejects in the
three significant factors but also imparted new lapping process of silicon wafer manufacturing using
insights on the interaction between these factors. It the Six Sigma methodology. It is a novel application
was found that unless these factors are set at their of Six Sigma as previously it has been implemented in
respective optimum levels simultaneously, inputting several industries, such as finance, service, manufac-
of individual optimum values of these factors will still turing and non-manufacturing but not for the lapping
result in poor TTV values. process of silicon wafer manufacturing in the semi-
Implementation and control were also carefully conductor industry. A significant reduction in rejects
considered. It had been argued in previous research from 4.83% to 0.02% was achieved through the
(Sreedharan et al., 2019; Raman & Basavaraj, 2019) implementation of the Six Sigma methodology,
that after the desired results through the Six Sigma resulting in the savings of £57.5k/month. A signifi-
methodology are achieved, if appropriate monitoring cant increase in the capability index (Ppk) of the lap-
and control procedures are not established, then most ping process also occurred, which is indicative of the
likely the process would regress back to its original, enhanced product quality and efficiency, thereby
flawed state (Antony & Coronado, 2002; Muraliraj et increasing customer satisfaction. In the current proj-
al., 2018). A similar phenomenon had been observed ect, three factors (the sun gear ratio, the plate tem-
in this organisation, where an effective solution had perature and the bottom plate speed) and the
temporarily fixed a problem, but a lack of appropriate interaction between the sun gear ratio and the plate
control systems caused the re-appearance of the same temperature were statistically significant. The optimi-
issue. For such reasons, in the current project, much sation and tighter control of these variables were key
emphasis was put on the control stage of the Six for the successful reduction in TTV rejects.
Sigma methodology, which was a new experience for On a large scale, dealing with such complex pro-
the entire team and, probably, the organisation too. cesses as lapping, it is often difficult to agree on
a starting point and contain the scope to a manage-
able extent. The Six Sigma techniques of define and
Limitations measure provided a much-needed direction and
structure to the problem-solving process, at an early
How stable are the improvements in the lapping stage. Previous attempts using traditional problem-
process? It has been shown that despite noticeable solving techniques had failed since several potential
improvements achieved by a project in the short- factors and their complex interactions were respon-
term, its sigma level and performance can decline sible for high TTV rejects. On the other hand, the Six
over time by 1.5 sigma. The standardised six-sigma Sigma technique could quantify and rank order sev-
level accounts for such variations in the long-term. eral factors with their interactions in terms of their
Since the current process did not achieve the full six- relative effect on the process outcome. Other Six
sigma level, it is likely that in case of a 1.5 sigma shift Sigma pointers, such as the commitment of the top
occurring, the process capability might regress to its management, the interim publication of success sto-
102
ries and open communication, also helped to main- Bothe, D. (2001). Statistical reason for the 1.5s shift. Quality Engi-
neering, 14(3), 479-488.
tain the motivation and focus throughout the project.
Brewer, P., & Bagranoff, N. A. (2004). Near zero-defect accounting
Overall, Six Sigma was shown to be an effective with six-sigma. Journal of Corporate Accounting & Finance
problem solving and quality improvement statistical (Wiley), 15(2), 67-43.
technique for the semiconductor industry, which lead Breyfogle, F. W. III. (1999). Implementing six-sigma: smarter so-
to a series of other benefits (such as the increase in the lutions using statistical methods. New York, United States:
John Wiley & Sons.
process efficiency, the reduction in time and financial
Breyfogle, F. W., Cupello, J. M., & Meadows, B. (2001). Managing
costs, the enhanced satisfaction of customers and six-sigma: a practical guide to understanding, assessing, and
employees as well as the provision of innovative team implementing the strategy that yields bottom-line success.
building and networking opportunities), in addition New York, United States: Wiley.
to the defect rate reduction in the lapping process. Brown, M. G., Hitchcock, D. E., & Willard, M. A. (1994). Why
TQM fails and what to do about it. Boston, United States:
The relevance of Six Sigma in the manufacturing Irwin Professional.
industry is irrefutable, but there may be a need to Bunce, M., Wang, L., & Bidanda, B. (2008). Leveraging six-sigma
conduct further rigorous research on its utility across with industrial engineering tools in crateless retort produc-
non-conventional sectors and the practicalities in tion. International Journal of Production Research, 46(23),
6701-6719.
terms of incurring costs and resources.
Byrne, J. A. (1998). How Jack Welch runs GE. Business Week,
Charles, 8 June 1997, 90-104.
Literature Cao, G., Clarke, S., & Lehaney, B. (2000). A systematic view of or-
ganisational change and TQM. The TQM Magazine, 12(3),
186-193.
Agarwal, R., & Bajaj, N. (2008). Managing outsourcing process: Chappell, A., & Peck, H. (2006). Risk management in military sup-
applying six-sigma. Business Process Management Journal, ply chains: is there a role for six-sigma? International Jour-
14(6), 829-837. nal of Logistics, 9(3), 253-67.
Al-Aomar, R. (2006). A simulation-based DFSS for a lean ser- Cole, R. (1998). Learning from the quality movement. California
vice system. International Journal of Product Development, Management Review, 41, 43-73.
3(3/4), 349-368. Craven, E., Clark, J., Cramer, M., Corwin, S., & Cooper, M. (2006).
Alexander, P., Antony, J., & Rodgers, B. (2019). Lean Six Sigma New York-Presbyterian hospital uses six-sigma to build
for small- and medium-sized manufacturing enterpris- a culture of quality and innovation. Journal of Organiza-
es: a systematic review. International Journal of Quality tional Excellence, 25(4), 11-19.
& Reliability Management, 36(3), 378-397. doi: 10.1108/ Crosby, P. B. (1979). Quality is free. New York, United States:
IJQRM-03-2018-0074 McGraw-Hill.
Andersson, R., Eriksson, H., & Torstensson, H. (2006) Similari- Crosby, P. B. (1984). Quality without tears. New York, United Sta-
ties and differences between TQM, six-sigma and lean. The tes: McGraw-Hill.
TQM Magazine, 18(3), 282-296. Cupryk, M., Takahata, D., & Morusca, D. (2007). Crashing the
Antony, J. (2004). Some pros and cons of six-sigma: an academic schedule’ in DCS validation pharmaceutical projects with
perspective. The TQM Magazine, 16(4), 303-306. lean Six-sigma and project management techniques: case
Antony, J. (2006). Six-sigma for service processes. Business Process study and discussion. Journal of Validation Technology,
Management Journal, 12(2), 234-248. 13(3), 222-233.
Antony, J., & Coronado, R. B. (2002). Design for six-sigma. IEE Da Silva, I. B., Filho, M. G., Luiz, A. O., & Lima junior, O. F. (2019).
Manufacturing Engineer, 24-26. A new Lean Six Sigma framework for improving competi-
tiveness. Acta Scientiarum Technology, Maringa, 41, e37327.
Antony, J., Snee, R., & Hoerl, R. (2017). Lean Six Sigma: yester- doi:10.4025/actascitechnol.v41i2.37327
day, today and tomorrow. International Journal of Quality
& Reliability Management 34(7),1073-1093. doi: 10.1108/ Dimitri, K. (1991). Reliability engineering handbook. New Jersey,
IJQRM-03-2016-0035 United States: Prentice-Hall, Englewood Cliffs.
Arul, K., & Kohli, H. (2004). Six-sigma for software application of Doble, M. (2005). Six-sigma and chemical process safety. Interna-
hypothesis tests to software Data. Software Quality Journal, tional Journal of Six-sigma and Competitive Advantage, 1(2),
12(1), 29-42. 229-244.
Bandyopadhyay, J., & Lichtman, R. (2007). Six-sigma approach to Edgeman, R., Bigio, D., & Ferleman, T. (2005). Six-sigma and busi-
quality and productivity improvement in an institution for ness excellence: strategic and tactical examination of IT ser-
higher education in the United States. International Journal vice level management at the office of the chief technology
of Management, 24(4), 802-808. officer of Washington, DC. Quality and Reliability Engineer-
ing International, 21(3), 257-273.
Benedetto, A. (2003). Adapting manufacturing-based six-sigma
methodology to the service environment of a radiology Esaki, K. (2016). Common Management Process Model of new
film library. Journal of Healthcare Management, 48, 263-80. TQM based on the situation analysis. Intelligent Informa-
tion Management, 8(6), 181-193.
Boaden, R. J. (1997). What is total quality management… and does
it matter? Total Quality Management, 8(4), 153-171. Eskildson, L. (1994). Improving the odds of TQM’s success. Quality
Progress, 27(4), 61-63.
Bolt, C., Keim, E., Kim, S., & Palser, L. (2000). Service quality six-
th
sigma case studies. ASQ’s 54 Annual Quality Congress Pro- Evans, J. R., & Dean, J. W. Jr. (2002). Total Quality: Management,
ceedings, Anais, 225-231. organisation, strategy. Mason, United States: South Western
103
Publishing. Kume, H. (1995). Management by quality. Tokyo, Japan: 3A Cor-

Foley, K. (2004). Five essays on quality management. Sydney, Aus- poration.
tralia: SAI Global Ltd. Kume, H. (Ed.) (1985). Statistical methods for quality improve-
Furterer, S., & Elshennawy, A. (2005). Implementation of TQM ment. Tokyo, Japan: The Association for Overseas Technical
and lean six-sigma tools in local government: a framework Scholarship.
and a case study. Total Quality Management and Business Magnusson, K., Kroslid, D., & Bergman, B. (2003). Six-sigma. The
Excellence, 16(10), 1179-1191. pragmatic approach. 2nd ed. Lund, Sweden: Student litera-
Gangidi, P. R. (2019). Application of Six Sigma in Semiconductor ture.
Manufacturing: A Case Study in Yield Improvement, Appli- McClusky, R. (2000). The rise, fall and revival of six-sigma. Mea-
cations of Design for Manufacturing and Assembly. Ancuţa suring Business Excellence, 4(2), 6-17.
Păcurar, IntechOpen. doi: 10.5772/intechopen.81058. Mishra, M. N. (2018). Identify critical success factors to
Retrieved from https://www.intechopen.com/books/ap- implement integrated green and Lean Six Sigma. In-
plications-of-design-for-manufacturing-and-assembly/ap- ternational Journal of Lean Six Sigma. doi: 10.1108/
plication-of-six-sigma-in-semiconductor-manufacturing- IJLSS-07-2017-0076
a-case-study-in-yield-improvement
Muraliraj, J., Zailani, S., Kuppuswamy, S., & Santha, C.
Garvin, D. A. (1988). Managing Quality: The strategic and competi- (2018). Annotated methodological review of Lean
tive edge. New York, United States: The Free Press. Six Sigma. International Journal of Lean Six Sigma, 9(1),
Gerhorst, F., Grömping, U., Lloyd-Thomas, D., & Khalaf, F. (2006). 2-49. doi: 10.1108/IJLSS-04-2017-0028
Design for six-sigma in product development at Ford Mo- Nwabueze, U. (2001). An industry betrayed: the case of total qual-
tor Company in a case study on robust exhaust manifold ity management in manufacturing. The TQM Magazine,
design. International Journal of Product Development, 13(6), 400-408.
3(3/4), 278-91.
Padhy, R. K., & Sahu, S. (2011). A Real Option based Six-sigma
Ghosh, S., & Rao, C. R. (Eds.) (1996). Design and analysis of experi- project evaluation and selection model. International Jour-
ments. Handbook of Statistics. 13. North-Holland. nal of Project Management, 29, 1091-1102.
Harari, O. (1997). Ten reasons why TQM doesn’t work. Manage- Pandey, A. (2007). Strategically focused training in six-sigma way:
ment Review, 86(1), 38-44. A case study. Journal of European Industrial Training, 31(2),
Harry, M. J. (1988). The nature of six-sigma quality. Rolling Me- 145-162.
adows, United States: Motorola University Press. Pyzdek, T. (2019). The six-sigma handbook: A complete guide for
Hoerl, R. W. (1998). Six-sigma and the future of the quality profes- black belts, green belts and managers at all levels (5th edn.).
sion. Quality Progress, 31(6), 35-42. New York, United States: McGraw-Hill.
Hossain, M. Z., Tasnim, M., & Hasan, M. R. (2017). Is Quality En- Raman, R., & Basavaraj, Y. (2019). Defect reduction in a capacitor
suring to Get Competitive Advantages in Auto Manufactur- manufacturing process through Six Sigma concept: A case
ing Industries? – A Study of Volvo Group. American Journal study. Management Science Letters, 9(2), 253-260.
of Industrial and Business Management, 7(1), 48-68. Reeves, C. A., & Bednar, D. A. (1994). Defining quality: alterna-
Ishikawa, K. (1968). Guide to quality control (Japanese): Gemba No tives and implications. Academy of Management Review, 13,
QC Shuho. Tokyo, Japan: JUSE Press Ltd. 419-445.
Ishikawa, K. (1985). What is Total Quality Control? The Japanese Roberts, C. M. (2004). Six-sigma signals. Credit Union Magazine,
Way. New Jersey, United States: Englewood Cliffs, Prentice- 70(1), 40-43.
Hall. Rucker, R. (2000). Citibank increases customer loyalty with defect-
Jimoh, R., Oyewobi, L., Isa, R., & Waziri, I. (2018). Total Quality free processes. The Journal for Quality and Participation, 23,
Management Practices and organisational performance: the 32-36.
mediating role of strategies for continuous improvements. Schmidt, M., & Aschkenase, S. (2004). The building blocks of ser-
International Journal of Construction Management, 19(2), vice. Supply Chain Management Review, 07-08, 34-40.
162-177.
Sreedharan, R. V., Raju, R., Vijaya Sunder, M., & Jiju, A. (2019).
Juran, J. (1995b). Summary, trends, and prognosis. In Juran, J. Assessment of Lean Six Sigma Readiness (LESIRE) for man-
M. (Ed.). A History of managing for quality: The evolution, ufacturing industries using fuzzy logic. International Jour-
trends and future directions of managing for quality (pp. 603- nal of Quality & Reliability Management, 36(2), 137-
657). Milwaukee, United States: ASQC Quality Press. 161. doi: https://doi.org/10.1108/IJQRM-09-2017-0181
Kapur, K. C., & Lamberson, L. R. (1977). Reliability in engineering Su, C. T., & Chou, C. J. (2008). A systematic methodology for the
design. New York, United States: John Wiley & Sons. creation of six-sigma projects: A case study of semiconduc-
Kazmierski, C. (2012). Semiconductor Industry Posts Record- tor foundry. Expert Systems with Applications, 34, 2693-
Breaking Revenues Despite 2011 Challenges. Retrieved from 2703.
http://www.sia-online.org/news/2012/02/06/global-sales- Talapatra, S., Uddin, M. K., Rahman, M. H. (2018). Development
report-2012/semiconductor-industry-posts-record-break- of an Implementation Framework for Integrated Manage-
ing-revenues-despite-2011-challenges/ ment System Based on the Philosophy of Total Quality
Kedar, A. P., Lakhe, R. R., Washimkar, V. S., & Wakhare, M. V. Management. American Journal of Industrial Business Man-
(2008). A comparative review of TQM, TPM and related or- agement, 8(6), 1507-1516.
ganisational performance improvement programs. ICETET Tobin, L. M. (1990). The new quality landscape: Total Quality
‘08: First International Conference on Emerging Trends in Management. Journal of System Management, 41, 10-14.
Engineering and Technology, India.
Topalovic, S. (2015). The Implementation of Total Quality Man-
Kotz, S. (1993). Process capability indices. New York, United States: agement in order to improve production performance and
Chapman and Hall. enhancing the level of customer satisfaction. Procedia Tech-
nology, 19, 1016-1022.
104
Valles, A., Sanchez, J., Morales, Noriega S., & Gomez Nunez, B.
(2009). Implementation of Six Sigma in a Manufacturing
Process: A Case Study. The International Journal of Indus-
trial Engineering: Theory, Applications and Practice, 16, 171-
181.
Wright, A. (1997). Public service quality: Lessons not learned.
Quality Management, 8(5), 313-318.
Wyper, B., & Harrison, A. (2000). Deployment of six-sigma meth-
odology in human resource function: a case study. Total
Quality Management, 11(4-6), S720-S727.
Yang, T., & Hsieh, C. H. (2009). Six-sigma project selection using
national quality award criteria and Delphi fuzzy multiple
decision-making method. Expert Systems with Applications,
36, 7594-7603.
Yong, J., & Wilkinson, A. (2002). The long and winding road. The
evolution of quality management. Total Quality Manage-
ment, 13(1), 101-121.
105

Reduction of Defects in The Lapping Process of The

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Reduction of Defects in The Lapping Process of The

Uploaded by

Copyright:

Available Formats

Engineering Management in Production and Services

Volume 11 • Issue 2 • 2019

received: 5 December 2018

Reduction of defects in the lapping pages: 87-105

process of the silicon wafer

Mithun Sharma, Sanjeev P. Sahni, Shilpi Sharma

DOI: 10.2478/emj-2019-0013 O.P. Jindal Global University, India

Polishing Etched wafers are mounted on plates. With

Increase process capability Preventive and Skilled workers

Denver Division To enhance the quality of a Extra attention and

Sigma (σ) Defects / Non-Conformances per

AUTHORS NAME OF AN ORGANISATION BENEFITS OF IMPLEMENTING THE SIX SIGMA TECHNOLOGY

Magnusson et Volvo cars (Sweden) Profit of over 55 million euro in the years 2000 and 2002

1 Wafer Traceability To correlate poor TTV Every Lot Data capture Central Database ‐ Lapping Op 93

Fig. 2. Y = f (x) cascade

8 2.11004Active Agent KPIV (can affect slurry Every time fresh

5,623 28,120 33,743 28120 Volume viscosity) C P L slurry prepared

DPMO 98,168 33,499 PPL

(Ppk) 11 Exhaust Timer KPIV (can affect wafer Any time changed

RATIO SPEED TIMER TIMER . STATUS

LA P -001 LA P -002 LA P -003 LA P -004 LA P -001 LA P -002 LA P -003 LA P -004

Fig. 5. CNX Diagram for DOE

Fig. 5. CNX Diagram for DOE

10 1.2 KPIV (can affect wafer Any time changed

1 1.2893 1.0338 1.0053 1.0092 1.1997 0.9940 1.0120 0.9957 1.0062

Normal Plot of the Standardized

Fig. 7. Normal and standardised

Main Effects Plot for Avg

Temperature Point Ty pe Temperature Point Ty pe

0.96 0.96 0.5 0.5

Fig. 8. Main effects and

Optimal Sun Gear

Avg TTV Avg TTV

Fig. 9. Response optimiser

• Plate Temperature = 30 °C; Using the recommended input values, Response

Fig. 10. Graphical Summary0.44

Fig. 10. Graphical Summary of Confirmation Test Run Results

Boxplot of TTV for Tight Flatness Spec by Improvement Stages

Lapping Date Out

Pilot testing was usedFig. for11.the Boxplot of TTV for

implementing it on larger sample size. The incremen-

P otential (Within) C apability

results was conducted to reduce the risk of large

LB 0 C P U 4.39 Ov erall 0.371802 LB 0 C P U 1.63

condition was expanded-0.00 to 0.65

time. Boxplot results from the pilot stage for loose

spec material are shown in Fig. 11.

Wafer Qty Process Capability of5,623 28,120 Improvement

P otential (Within) C apability P otential (Within) C apability

0.21 0.52 ‐NA‐

(Ppk) SS ample N 19677 C PL * S ample N 50666 C PL *

SUN GEAR BOTTOM PLATE EXHAUST ACCELERATION P pk 3.87

RATIO SPEED TIMER TIMER TEMP. STATUS CONCENTRATION TEMP. MIXING TIME

Publishing. Kume, H. (1995). Management by quality. Tokyo, Japan: 3A Cor-

You might also like