Mask Costs Jumping (and the effects on CYMI....)
Merchant, captive shops face unparalleled complexity
From Electronic News--June 7, 1999
By Gale Morrison
San Jose--Spot sightings of half-million dollar photomask sets have IC manufacturers
worldwide on high alert.
The full-blown shift this year to lithography below 0.25-micron is confronting chipmakers
with soaring mask costs. The situation could become troubling for IC vendors that lack
the volumes to amortize the nonrecurring engineering (NRE) charges.
"Today, a set will have between 15 and 30 layers. That could cost you anywhere from
$50,000 to $100,000 and up. I know of mask sets that have sold for over a million,"
said Paul Warkentin, senior vice president for R&D at Etec Systems, a manufacturer of
maskmaking equipment.
Such an NRE expense presents an acute problem for foundries and their fabless
customers who might be selling, or evaluating, the die from just 300 or fewer wafers.
"Microprocessor and DRAM makers can get tens of thousands of wafers from one set.
The cost is recovered," said S.Y.Chiang, head of R&D and president of leading foundry
Taiwan Semiconductor Manufacturing Co. (TSMC). "Our customers might want 300
wafers and that's it. The mask gets very, very expensive then."
Chiang said that mask costs increased 50 percent in the 0.35-micron to 0.25-micron
shift and now they are staring down the barrel of another 200 percent increase in mask
cost in the 0.25-micron to 0.18-micron shift.
There are fundamental reasons for the rapid escalation in price, say maskmakers and
Etec. In the 0.25-micron to 0.18-micron shift, the data to produce the precise pattern for
every layer is four to six times as large. And the throughput of the Etec tool – meant for
0.25-micron but being pushed to 0.18-micron – to pattern all of the 15 to 30 masks
involved is four times as slow.
The mask industry has scaled a tall technological mountain, particularly in light of the
commercial beginning in the 1980s when masks were drawn out at five times (5X) the
one micron-plus wafer feature size sought. Ken Rygler, executive vice president
worldwide marketing and strategic planning at DuPont Photomasks Inc. (DPI), terms this
"the 5X holiday."
"We have come more than full circle now. The shrinks have come so far and so fast,"
Rygler said. "Our life has gone to multiple writing tools, multiple photoresists (with
plasma instead of wet etching), multiple mask types..." Even the mask coating is
changing, from chrome to a complex "MoSi" material.
"The leading edge is always expensive," Rygler argues. "Yields are not as high. You're
building them with tools that are a generation behind because the (patterning) tools aren't
ready yet."
Considering all that, DPI's concern is where to find the profit. "I have to sell them for half
a million, so I sure have to make them for less than that. I've got to turn a profit, and the
captive shops don't," he said.
Still, IC makers can enjoy an unforeseen lithography equipment-purchasing holiday.
These most sophisticated mask techniques, like all manner of phase-shifting, have
allowed fab operators to continue using their 248nm (or 0.248-micron) wavelength
lithography equipment to create 0.18-micron feature sizes. The masks are allowing the
light source to make smaller features than its own wavelength, something that is possible
now for the first time in industry history.
Photomask makers are quick to point this out, and herald the "sub-wavelength era" when
they do.
"If this (more expensive) mask set stops you from buying another $10 million lithography
tool, you're coming out ahead," said Mike McCarthy, spokesman for Photronics. "You
can use a tool you've already paid for, one that's pretty well depreciated" to make
leading edge feature sizes.
Concurrently, the maskmakers with more process tweaking get their laser beam,
0.25-micron writing tools (predominantly Etec's ALTA 3500) to write the 0.18-micron
masks, albeit not as quickly as they would with a new, roughly $13 million Etec electron
beam MEBES 5000. Such a break for the mask shops could be partly to blame for
Etec's third quarter loss and recent layoffs.
And the maskmakers also have new and improved capital expenses to contend with,
before they decide to buy a MEBES tool. "The second largest capital expense in a mask
shop is usually inspection," Warkentin said. "They have to go to KLA-Tencor and get
much more complex equipment than has been the case. The metrology has gone from
simple optical to much more demanding CD-SEM (critical dimension-scanning electron
micrograph)," he said.
Photronics and DPI are quick to point out that mask sets in the majority cost in the low
six figures, as they are being made for production on the 0.35-micron scale, where there
is plenty of mask-making, as well as fab line, capacity. The great majority of IC
production today is still on the 0.35-micron scale.
Photronics' McCarthy said that getting a handle on NRE was the chief motivation for the
company's new Design-to-Wafer (D2W) business unit. Photronics struck a deal in
March with Cirrus Logic to acquire the Milpitas, Calif. fabless pioneer's mask
engineering group and let it run autonomously. Richard Morse is D2W's general
manager, reporting directly to James R. Northup, Photronics' president.
And Photronics' rival DPI is thinking along the same lines. "We're working with our
customers to reduce the costs," Rygler said. "In every reticle set, you have critical layers
and non-critical layers. The pad mask is easiest, that might be $2,000. A (phase shift)
mask for your poly via gate device level could cost $20,000. What you want to do is
optimize each layer."
At TSMC, which runs its own captive mask shop, Chiang said that the company is
contemplating – as many have over the last 20 years and IBM and NEC still are – using
a "direct write" path to cheaper production of small wafer batches.
Warkentin, Rygler, McCarthy and others don't see it happening. Etec shipped several
ABELE electron beam direct write systems from 1988 to 1990, Warkentin said, but
most of them have been decommissioned, except for a few being used in advanced
military applications where data security is so highly prized.
Direct write could come about though. Etec continues to pursue a microcolumn program,
which the Defense Advanced Research Projects Agency and the Naval Air Systems
Command have supported to the tune of $10.8 million. In the program, Etec has turned
what is typically a three-beam, roughly two- to five-feet long patterning source into
integrated parallel columns that are one centimeter in size. The shorter length and parallel
writing ability might bring direct write back into the light, as it were. |
