An Analysis of the Economics
of Photomask Manufacturing
Part – 1: The Economic
Environment
Charles M. Weber – Portland State University
C. Neil Berglund – Portland State University
ISMT Mask Automation Workshop
February 9, 2005
Problem Statement
• The cost of producing photomasks
– has increased by an order of magnitude over the last
decade,
– and is likely to continue to escalate for the 90-nm
technology node and beyond.
– Ultimately the cost of photomasks may rise to the point
where it threatens the profit potential of most low
volume designs.
• Suppliers of photomasks are thus looking for ways
to reduce the cost of producing photomasks,
• and semiconductor manufacturers are looking for
creative ways to amortize them.
ISMT Project LITG323
“Mask Industry Assessment”
• Conducted by
– Portland State University’s (PSU)
– Maseeh College of Engineering and Computer Science’s
(MCECS)
– Department of Engineering and Technology Management
(ETM)
• Principal Investigators:
– C. Neil Berglund – Portland State University
– Charles M. Weber – Portland State University
• Purpose:
– Investigate the economics of photomask manufacturing
– Provide input into annual mask industry survey
Research Question
• What can a supplier of photomasks do to
enhance the profitability of its customers, in
addition to augmenting his/her own
profitability?
Research Methods
• Interviews with two dozen experts
– Photomasks and related disciplines
• Data have been transcribed, coded and analyzed.
• Numerical model of the semiconductor lifecycle
• Leads to stylized facts concerning
– the economics of photomask fabrication,
– and profitability in semiconductor manufacturing.
• Please see reference (Weber, 2003, Ch. 3) for
details.
The Mask Shop’s Perspective
• Finite demand.
– Global market for photomasks <$3 billion
(Kimmel, 2003@ BACUS)
– Unit revenues for masks erode over time.
• Capital-intensive industry. Investment levels…..
– ~$40M for ‘conventional’ (180-nm node or above)
– >$100M for ‘advanced’ (130-nm node and beyond)
– Photomask manufacturers fund making advanced
masks from making conventional masks.
• Few players can afford to be in business.
State of Affairs
(Industry Survey by Kurt Kimmel @ BACUS, 2003)
• Mask business growth rate is 6.8%, as compared
to 9% in semiconductors
• Currently <180-nm masks drive 37% of revenue
(15% of mask shop volume).
• Small, but growing portion of masks require
advanced resolution enhancement techniques.
• Hard defects are the primary yield problem!
• Severe data management issues
• 4 merchants control 65% of market
• Things will get worse as features shrink.
Expected Mask Costs
for Future Technology Nodes
(Source: International Sematech)
• $0.5M per mask set for 130-nm node
• $1M per mask set for 90-nm node
• $2M per mask set for 65-nm node
(Weber, Berglund, Gabella, 2004)
• >50% yield is critical
for profitability.
• Amortization of
fixed costs complete
after 20,000 masks.
• Maximize equipment
utilization.
250
Average Total Cost (k$/mask)
Mask Shop
Experience Curve:
Advanced Masks
Av e ra g e T o ta l C o s t o f
P h o to m asks
Y ld = 0 .1
200
Y ld = 0 .2
Y ld = 0 .3
Y ld = 0 .4
150
Y ld = 0 .5
Y ld = 0 .6
100
Y ld = 0 .7
Y ld = 0 .8
Y ld = 0 .9
50
Y ld = 1 .0
0
0
10
20
30
C u m u la tiv e N u m b e r o f
M a s k s P ro d u c e d ('0 0 0 )
Average Total Cost of Critical Layer Photomasks
Assumptions: variable costs equal $15k per mask;
fixed costs of $100 million need to be amortized.
Initial Recommendations
• Improve yield rapidly.
–
–
–
–
–
Mask operations are similar to fab operations.
Transfer knowledge from customer.
However, yield knowledge is tacit (Polanyi, 1966).
Personnel transfer and extensive training is required.
How many yield engineers can a mask shop afford?
• Centralize mask making to enhance capital
productivity.
The Chipmakers’ Perspective
• Amortization challenges
• Limited effects of yield learning for
foundries and ASIC manufacturers
• The Designer’s Learning Cycle
• Rapid turn-around time
• Shorter product lifetimes
• Early versus mature masks
Price Reduction or Market Shrinkage
(Weber, Berglund Gabella, 2004)
(Trybula, Intl. Sematech)
Mask Cost as a Fraction of Total
Semiconductor Investment
Pressure to reduce
number of designs
0.10
0.08
1k Des.
300 Des.
100 Des.
30 Des.
10 Des.
3 Des.
1 Des.
0.06
“Pain Threshold”
0.04
Historical Level
0.02
Pressure
to reduce
mask cost
$10M
$3M
$1M
$300k
$100k
0.00
$30k
Fraction of Total Investment
• If mask costs cannot be
reduced,
• Then pressure to
reduce the number of
designs increases.
• Fewer designs means
market shrinkage.
• Foundries and ASIC
vendors are affected
disproportionately.
Cost per Mask Set
Model assumes a billion chips over the lifetime of a factory, a total investment of $3billion excluding photomasks.
Amortization Challenges for
Foundries and ASIC Vendors
•
•
•
•
Only best sellers can be printed.
10,000 new designs in 2000
Only 3500 new designs in 2003
Drop cannot be completely explained by the
economic downturn.
Mask Cost and Wafer Cost
Mask Cost in Relation to Wafer Cost
Mask Cost/Wafer Cost (%)
10.0%
8.0%
$200k/MS
$500k/MS
6.0%
$1M/MS
$2M/MS
4.0%
$5M/MS
$200k/MS
2.0%
$500k/MS
$1M/MS
0.0%
$2M/MS
$5M/MS
0
5
10
15
20
25
30
35
40
Cumulative Number of Wafers Produced ('000s)
Calculations assume 400 fully functional chips per wafer.
• >20,000 300-mm wafers need to be produced
for a 65-nm mask set to be amortized.
$3k/wafer $8k/wafer
Cost per
Mask set
Scale versus Scope
in Semiconductor Manufacturing
Decreasing cost per function
(Moore’s Law)
Decreasing unit cost
Increasing product
functionality and complexity
(more devices per product)
Increasing scale
(300-mm fabs)
Market forces replace
capacity constraint
(Bohn & Terwiesch)
Increased process complexity
Increased development costs
Increased fixed cost per product type
(e.g. cost of photomasks)
Increased cost of
manufacturing
multiple products
Conflict
Shorter market windows
Increased need for
manufacturing
multiple products
Scale versus Scope
• A firm enjoys economies of scale when it can
double its output at less than twice the cost.
(Pindyck & Rubinfeld, p. 223)
• Economies of scope are present when the
joint output of a single firm is greater than the
output that could be achieved by two different
firms each producing a single product (with
equivalent production inputs allocated between
the two firms). (Pindyck & Rubinfeld, p. 227)
Logic Life Cycle Drives the Industry
$1,000
486DX
$900
P5-66
PII-300
PIII-800
$700
$600
8 years
$500
$400
386/16
$300
$200
$100
68EC100
286
386/25
6-9
Mos.
$0
85
Q
1
85
Q
3
86
Q
1
86
Q
3
87
Q
1
87
Q
3
88
Q
1
88
Q
3
89
Q
1
89
Q
3
90
Q
1
90
Q
3
91
Q
1
91
Q
3
92
Q
1
92
Q
3
93
Q
1
93
Q
3
94
Q
1
94
Q
3
95
Q
1
95
Q
3
96
Q
1
96
Q
3
97
Q
1
97
Q
3
98
Q
1
98
Q
3
99
Q
1
99
Q
3
00
Q
1
00
Q
3
Average Sales Price ($)
$800
Life cycle is decreasing rapidly. (Source: Walt Trybula, Intl. Sematech)
Limits Effects of Yield Learning
in ASIC Manufacturing (Weber, 2002)
NPV (M$) @ 20% Discount Rate
T h e Effect o f Accelerated/D ecelerated L earn in g
6000
Not an option
for ASIC vendors
Designers are
not ready.
5000
4000
3000
6 months ahead
on schedule
6 months behind
12 months behind
18 months behind
2000
1000
0
-1000
-2000
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
T ime in Years
• ASIC manufacturer profitability in a 200-mm wafer fab.
• When venture is on schedule, one minute on the critical
path amounts to $2500.
The Designer’s Learning Cycle
Test
(1 week)
Design
(1 year)
Dev. Mask Set
Fab
Analysis
(1 month)
(3 months) (1 week)
• Full learning cycle can take more than 15 months.
• Market fluctuates much faster. (Sematech GEM)
• Only end user firms with long planning horizon
can afford to design chips.
Rapid Turn-Around
• Fab problems can cost $10,000 per minute.
(Weber & von Hippel, 2000)
• Inter-continental delivery of masks is
unacceptable to many fabs.
• Mask maker must build fabs close to
customers. Centralization is very difficult.
• Capital productivity of mask shops drops.
Early Masks vs. Production Masks
NPV (M$) @ 20% Discount Rate
T h e Effect o f Accelerated/D ecelerated L earn in g
6000
Early Masks
5000
4000
6 months ahead
on schedule
6 months behind
12 months behind
18 months behind
Production
Masks
3000
2000
1000
0
-1000
-2000
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
T ime in Years
• Early masks: low volume; low yield; long turnaround; very
high cost; very high price.
• Production masks: moderate volume; higher yield; short turn
around; relatively lower cost; much lower price.
• A mask shop that cannot make early masks won’t get business
downstream.
Summary
• For mask shops …..
– Reducing defects is not enough.
– Rapid yield improvement efforts may not be
affordable.
– Rapid turnaround prevents full utilization of
equipment.
– Cost reduction or market shrinkage
– Enormous consolidation pressure
– Market segmentation: advanced versus
conventional.
Recommendations for Mask Shops
• ITRS: Move the mask problem onto the near-term
problem list.
• Coordinated development:
– Align mask development with leading customers
(chipmakers) and complementors (e.g. stepper
manufacturers, resist suppliers)
• Develop pricing strategies with customers.
• Joint ventures with customers.
• Merge!
Currently in Tech Transfer:
Strategic Threat Analysis for the
Photomask Industry
Entry
Economies of scale
Proprietary product
differences
Brand identity
Access to distribution
Government policies
Expected retaliation
Suppliers
Suppliers
Supplier concentration
Supplier volume
Supplier switching costs
Brand identity
Impact on cost
Impact on differentiation
Supplier profile
Entry
Rivals
Substitution
Substitution
Substitute Products
Rivalry
Industry growth
Product differences
Brand identity
Exit barriers
Diversity of competitors
Buyers
Buyers
Buyer concentration
Buyer volume
Buyer switching costs
Product differences
Brand identity
Impact on quality
Impact on performance
Buyer profile
• Based
on
Porter
Model
Currently under Investigation:
Best Practices for Mask Industry
• A list of best practices suggested by
analysis of economic environment is in tech
transfer.
• Additional questions regarding best
practices may be included in 2005 mask
survey.
Currently under Investigation:
Scenarios for Evolution of the
Semiconductor Industry
1. Technical Substitutions for Photomasks
•
Maskless Lithography
2. Operational Solutions
•
Fewer critical masks per mask set
3. Vertical Integration of Advanced Mask Making
4. Bifurcation of Moore’s Law
•
•
DRAM and Microprocessors on accelerated schedule
ASIC and perhaps foundries on slower schedule
Interim Conclusions
• A combination of scenarios 1-4 is likely to occur.
• Severe consolidation pressure for all four
scenarios.
• The patterning problem may cause a restructuring
of the semiconductor industry.
• Factory automation and operational efficiency are
becoming critically important.
Factory Automation and Operational
Efficiency in Fabs and Mask Shops
R.C. Leachman (UC Berkeley); C. Weber, C. N. Berglund (Portland State U.)
Problem Statement:
• Capital investments in the billions
(100s millions) of dollars
• Profitability depends on operational
efficiency.
• Processes, manufacturing
environment and economic
environment are highly complex.
• Highly dynamic manufacturing
environment.
• Critical decisions must be made in
real time.
• Real-time decisions may have to be
based on simulations of fab and
economic environment.
Limits of Many Approaches:
• Wrong evaluation metric Æ wrong
answer!
– Wafer cost or even die cost
are insufficient!
• Models that do not cover the
complete investment horizon give
the wrong answer!
• Models that do not assess risk are
of little value!
• Insufficient number of managerial
variables
– Management responses tend to
be multidimensional.
– Need to make Apples-tooranges comparisons!
• Models do not run in real time.
The Value of Ownership Model
R.C. Leachman (UC Berkeley); C. Weber, C. N. Berglund (PSU)
Configurable, Real-time, dynamic
factory simulation
• Predicts factory behavior
• Covers complete economic
environment
• Spans required investment horizon
• Comprehends variability.
• Gets the right answer!
• Drives decisions.
Project Implementation
• Collaborative development of emanufacturing infrastructure
• Scalable models
• Three-year project horizon
• Intended start date: Fall 2005
Value of Ownership Approach
• Gives real-time assessment of
impact on bottom line
• Latest data and latest expectations
are entered into simulation.
• Simulation
– generates potential
scenarios for action;
– assigns $ value to each
scenario;
– assesses risk for each
scenario.
• Managers choose the best
scenario.
• Ultimate MC Benefits: rapid,
data-driven decision making
capability based on $ value.