The Cook’s Tour: SkyWater Technology Foundry

The Chip Insider®

Copyright © 2019 VLSI Research Inc. All rights reserved

June 28, 2019 – Strategy and Tactics: The Cook’s Tour: SkyWater Technology
Foundry. WildPhotons: The most damaging phrase in the language is `It's always
been done that way’ …
Focal Points:
The Cook’s Tour: SkyWater Technology Foundry is an emergent semiconductor manufacturer
for whi ... See more

h its uniqueness lies in being a solely U.S.-owned-and-operated, DMEA1 Trusted
Semiconductor Foundry with ITAR compliance. The operative word in front of ‘foundry’ is
‘technology’ as opposed to ‘specialty.’ We mine the difference and take a tour to review the fab,
tactics, and strategy. And to assess SkyWater’s position in the semiconductor industry.
“Intelligence is quickness in seeing things as they are”
— George Santayana

The Cook’s Tour: SkyWater Technology Foundry is an emergent semiconductor

manufacturer for which its uniqueness lies in being a solely U.S.-owned-and-operated, DMEA
Category 11 Trusted Semiconductor Foundry with ITAR compliance. Like most such enterprises,
SkyWater has a long history. It was spun off from Cypress Semiconductor in 2017, who
originally bought it from Control Data (CDC) in 1991. Built in 1986 by CDC as a secure facility to
make chips for the military, it was a ‘Trusted Foundry’ long before the term was invented. And
‘Trusted’ is not just a word they tacked on in a tagline-branding exercise. It’s a U.S. Department
of Defense certification program run by the DMEA.
SkyWater was formed in 2017 when it was divested from Cypress Semiconductor. By then
Cypress had transitioned to the No-More-Moore strategy of becoming fab-lite and, in doing so,
they began the smart move2 of converting the operation to being a foundry with excess
customers.
Today it has 80K square feet of clean room space with a capacity to process upwards of 100K
Mask-Layers-per-Week on 200mm wafers, with 500 employees to make it fully operational. This
makes it Right-Sized for the small-to-medium sized customers it is attracting — i.e. less-thangiant customers can have easy access to SkyWater’s A-team without having to be designated a
‘strategic customer’ by the giant foundry. 200mm makes it flexible, with low capital cost so a
customer’s development team can play in the fab without emptying their wallet. This doesn’t
mean the wafers come free, as they say they have already “delivered strong EBITDA.” So they
are here to stay, which is always another concern for fabless customers. Companies that can’t
earn will burn as they enter the earth’s atmosphere of semiconductor industry competition.
That brings up another potential concern: Leadership? Here they also pass the test with
leadership educated in the Ivy League of the semiconductor industry. Senior management

comes from the universities of AMD, Cypress, ST, Texas Instruments, and TSMC. So they know
what they’re doing and, as you will see, it shows in the fab. After that brief introduction, let’s
get on with the tour.
When going to SkyWater, the first thing one notices is how easy it is to get there. It’s about a
three-hour flight from Silicon Valley… with direct flights — no connections — no dinky little
plane… with no need to leave at midnight for a 10+ hour flight to catch a meeting the next day.
Landing in Bloomington, it literally takes longer to walk from the gate and get your rental car
out of the parking lot, than it does to drive to the fab. SkyWater’s so close to the airport, that
Waze couldn’t load directions before I’d missed the highway exit. One can see how close it is in
the aerial shot, where the airport is in the background. Now this may seem extraneous, but
fabless-foundry interface executives say travel-time is an important opportunity-cost in today’s
Time-To-Christmas driven design cycles — it’s an opportunity-cost that is both business and
personal, making it even more important.

Once in the fab, there’s a sense of déjà vu, as it’s in the mold of classic 200mm SMIF’d
production lines — the kind that launched TSMC to its prominence because of the technology’s
inherent yield and flexibility advantage.3 Like any well-run fab, there were plenty of green lights
to indicate the tools were functioning and processing wafers.

Lots were well organized to maximize tool utilization. SkyWater’s use of mini-environments
allow them to line-up lots of wafers near tools like airliners line up to land on a Friday night.
This is one of the most efficient ways to maximize utilization. What was most intriguing is that
SkyWater is running a very high-mix assortment of lots with a significant amount of process
variation between them.

As one moves from the fab’s lithography bays, the yellow lights go back to white, which saves
on energy — making it somewhat greener.4 Another key sign that it’s a well-run fab is there are
few people in the aisles and there were few systems in maintenance mode. This is critical
because maximizing tool utilization is crucial to managing costs. Importantly, the vibe from fab
workers was very positive, like the one below who one of the tour guides called out to.
Otherwise it was like the old Japanese fabs of lore: most everyone had a laser focus and never
looked up to see who was being paraded through. Clearly everyone has internalized the threat
they face in today’s world of Brontosaurus-sized competitors with multiple 300mm fabs.

There’s still little automation outside of the tools. However, you can see they do quite well
moving lots with carts. And they don’t have to deal with the problem of traffic jams in the
overhead transport track of AMHS5 systems, which becomes increasingly problematic with
higher numbers of different process flows, different lots on the line, lot priority changes, etc.
Variability is the bane of automation systems, which can result in high levels of inefficiency in a
high-mix manufacturing operation like SkyWater’s. The efficiency of this approach could be
seen in the diffusion bays.

Moving on, you pass through a test area and on to a separate CMP area, which is classic
cleanroom protocol. Notably, the test area could also provide space for extending parts of the
fab for disruptive material needs.

All told, what I have portrayed here is a fairly conventional 200mm fab running standard CMOS
production technologies, for which Cypress was so famous. That said, SkyWater has taken it
beyond this adding MEMS — which is CMOS compatible, plus more…
What I haven’t shown is some of the new tools I saw being installed. These were not for added
capacity, but for added capability SkyWater is developing to serve its new businesses in
Superconducting ICs, Carbon Nanotubes, and Photonics. These are all emergent markets with
disruptive potential in computing.
Serving these technologies is more than talk, as it has ongoing contracts with multiple
customers in these areas. One can see SkyWater is walking the talk in its list of tools and the
materials being piped in. Already, they can handle 40% more elements than typically found in a
CMOS fab of this capability.

SkyWater has been positioned as a technology foundry, in contrast to being a specialty foundry
or a throw-off from a company trying not to bleed on the razor of Moore’s Law — they don’t
call it “cutting edge” for nothing. The difference here is subtle, but real. Specialty foundries are
more focused on volume manufacturing outside the bleeding edge of the giants. They offer
specialized processes that they have developed on which the customer can design — using
existing libraries and PDKs. A technology foundry is more focused on developing new processes
that are more unique to the individual customer. This can be as simple as a typical ASIC
approach where the customer just brings product concept. But then it goes beyond this in its
ability to work hand-in-hand with customers to identify unconventional materials, integrate
process flows, and take it to manufacturable yield.
In terms of operational excellence, they are structured to run development and production in
the same fab. There are clear hand-offs between Development Engineers, Process Engineering
Managers, and Manufacturing Area Manufacturers — all aligned to specific process steps.
This is a structural advantage for achieving faster ramps and higher early yields on emergent
product lines — especially when new materials are involved. It also provides a seamless
interface from development, to LRIP5 products, to small volume production. This shoehorns
perfectly into their 200mm SMIF’d fab, as this class of fabs has the ability to turn on a dime
when it comes to running large volumes of small lots. Compared to a 300mm fab, it’s like the
difference between an attack sub and a boomer.
Now this doesn’t mean they lack automation. The tools are all mechanically automated in a
conventional way. What they do lack is automation of material flows between tools. They make
up for this with extensive automation of data flows. It’s far more important to know what lot
needs to be where than it is to move it to the next tool. If you know it data-wise the only

difference is machines versus humans moving wafers and 200mm cassettes, which are light
enough to be carried. When it comes to data automation for rapid decision making, SkyWater is
pretty state-of-art. This includes Advanced-Process-Control, that their President — Tom
Sonderman — became so famous for. Moreover, they can articulate well where the gaps are
that need to be filled. Since SkyWater’s founding, fab productivity has consistently risen as a
result of playing this game of data whack-a-mole. This has happened even though they’ve
added more product categories and are addressing more market segments.
SkyWater’s addressed end-markets are: Aerospace & Defense, Automotive, Computing,
Consumer, Industrial, and Medical. They are Medical Device 13485 Certified. Classical market
segments addressed are High-Performance Mixed Signal, MEMS, Power, ROICs6, Interposers,
and Rad-Hard. Processes to serve these can run from 350nm to 90nm to deal with
heritage/longevity needs of customers for short production runs of legacy products in markets
like auto and aerospace. Their processes can span power requirements as high as 200 Volts.
Packaging abilities include TSVs and active interposers. On top of this are the exotic
technologies SkyWater is addressing, like Superconducting ICs, Carbon Nanotubes, and
Photonics. Processes for these included Josephson Junctions, exotic memories to complement
them, CNFETs, photon waveguides, interferometers, and more.

Put it all together and you get the type of impressive market-to-technology stack that’s only
possible from a new/old company like SkyWater. New in the sense that they are right-scaled to
do truly innovative work. Old in the sense that they are mature enough to make a business out
of it, ensuring supply chain sustainability for future customers.
1) DMEA Trusted Foundry: Defense MicroElectronics Activity. Skywater has a Category 1 certification,
which means its products can be used for “mission critical” applications as defined in DOD 8500. ITAR:
International Traffic in Arms Regulations.
2) A smart move because the conventional fab-lite model was a halfway-house to being fabless, as with
it a company would try to keep a fab for research.
3) Today’s 300mm fabs all use a similar mini-environment approach called a FOUP.

4) While these just have yellow filters, the wattage of the bulbs has to be higher to overcome the loss of
light in order to get the same lux in the cleanroom. Also, the extra wattage and thermalization of
photons in the filter leads to extra cooling being needed as well. Whoever build the fab this way gave
some real thought had gone into having sections in white light, as it is also easier on worker’s eyes.
5) AMHS: Automated Material Handling Systems. These systems are great in fabs where the mix is much
lower relative to what SkyWater is running today. To be cost effective, 300mm fabs have to run very
high volumes of with few and similar process flows, with minor variants, due to the size of the wafer and
the cost of the fab. My point is not to demean automation, but to bring out the often overlooked
positives of carts in high-mix conditions where lots of LRIP products running. Especially since, most
people would see them as a negative, given what is highlighted in 300mm fabs.
6) LRIP: Low Rates of Initial Production.
7) ROIC: Read-Out Integrated Circuits are used to capture data from sensors — typically for reading
photocurrent off image sensor pixels.

Links to VLSI stuff in the Public Domain:
IBM Research AI Hardware Center ... with Mukesh Khare of IBM
SOI: Why it's no longer a niche technology - with Carlos Mazure of the SOI Industry Consortium
Mark Bohr's Rules of Innovation
Critical Subsystems: Probably one of the most important …
Mark Bohr: The Scaling + Heterogeneous Integration Era
Mark Bohr: The Power Constrained Era
Mark Bohr: 40 Years of Moore's Law - From Memory to Dennard Scaling

Archive of this Year’s Previous Issues:

Archive in the TCI section at: VLSIresearch.com

June 21, 2019 – Strategy and Tactics: The Silicon Cycle: Does Diversification Mute Volatility?
WildPhotons: We learn wisdom by 3 methods …
June 14, 2019 – Strategy and Tactics: Alibaba at 7nm. China’s Critical Tech Weaknesses. Out-of-Sample
Uncertainties. WildPhotons: We keep moving forward because curiosity keeps leading us down new
paths …
June 7, 2019 – Strategy and Tactics: What could push the downturn into 2020. Risk vs Uncertainty.
CHC: Ladybugs, Legos, and Bunnysuits. WildPhotons: Only a fool learns from his own mistakes …
May 31, 2019 – Strategy and Tactics: Life After CMOS. Apple vs. Nikon revisited. WildPhotons: I
choose to go to the moon …
May 24, 2019 – Strategy and Tactics: Trump's Trade War Impact on Semiconductors WildPhotons:
Intelligence is quickness in …
May 17, 2019 – Strategy and Tactics: China-fail factors, NAND Pricing stabilized, Applied Materials turns
market, Samsung’s reach for the technology crown. WildPhotons: Statistics are no substitute for
judgment …

May 10, 2019 – Strategy and Tactics: Inflation and how it is affecting globalization and changing the
structure of the semiconductor industry. WildPhotons: Partnerships help you weather long winters …
May 3, 2019 – Strategy and Tactics: The End of Moore’s Law: Why it won’t be the disaster so many fear.
GLOBALFOUNDRIES sells another fab. WildPhotons: An enemy can do more damage than the …
April 19, 2019 – Strategy and Tactics: China’s Semiconductor Equipment Industry Challenge.
WildPhotons: Changing Strategy …
April 12, 2019 – Strategy and Tactics: The Future of ATE. Testing Quantum Computers. Happenings.
WildPhotons: Every block of stone has a statue inside it
March 29, 2019 – Strategy and Tactics: Profitable Strategies in Buying and Partnering for process
development. WildPhotons: Teamwork is essential when opportunities are overwhelming …
March 15, 2019 – Strategy and Tactics: SPIE Week WildPhotons: Over the hill …
March 8, 2019 – Strategy and Tactics: Process Diagnostics Revolution. WildPhotons: I paint things …
February 22, 2019 – Strategy and Tactics: A Cyclical History of the Semiconductor Equipment Industry.
WildPhotons: It’s the eye which makes the horizon …
February 8, 2019 – Strategy and Tactics: IBM Research AI Hardware Center. AMD’s supply strategy with
GLOBALFOUNDRIES. Q&A. WildPhotons: Elevate the soul to great things …
February 1, 2019 – Strategy and Tactics: ASML’s AI, DRAM EUV, Mapper, and AI Strategy. CHC: KLATencor’s shrink-wrapping software in a system. WildPhotons: Sometimes leaders have to stand their
ground …
January 25, 2019 – Strategy and Tactics: 2019 Strategic Forecast. CHC: Sunlin Chou: A tribute.
WildPhotons: There’s plenty of room at the top …
January 18, 2019 – Strategy and Tactics: IBM Lays out the future of AI. in-Memory Computing defined.
Neuromorphic to change manufacturing. WildPhotons: I've always been more interested in the future
than in the past …

WildPhotons: light life lessons

VLSIresearch

Ph: 408.453.8844 | 2290 North First Street Suite 202 San Jose, CA 95131

… intelligence to make better decisions faster
VLSIresearch.com
ChipHistory.org
weSRCH.com
By accepting this report, opening it, or using it you are agreeing to these terms. This report contains valuable proprietary information
developed or acquired by VLSI Research at great expense. You have a limited license to hold these materials but do not become
the owner of any material. The materials provided are protected by copyright, trade secret, and trademark law. This presentation
has been approved by VLSI Research for public release with attribution. The information in the materials may be used by you on a
limited basis in your own documents provided that those documents are not-for-sale; VLSI Research’s name, brand, or trademarks
are not used to endorse a product or company for sales purposes; VLSI Research is clearly referred to as the source of such
information; and you obtain written approval prior to use.
This report is provided on an "AS IS," "WHERE IS", "WHERE AVAILABLE", “WITH ALL FAULTS” basis. VLSI Research does not
warrant these materials or the information provided therein, either expressly or impliedly, for any particular purpose and VLSI
Research specifically disclaims any express or implied warranties, including but not limited to, any express or implied warranties of
TITLE, ACCURACY, NON-INFRINGEMENT, MERCHANTABILITY or FITNESS FOR ANY PARTICULAR PURPOSE OR USE.
The sources of the information in this report include numerous individual reports, memos and bulletins from various segments of the
industry, annual reports, financial reports, interviews, questionnaires, surveys, technical symposia, trade journals, technical journals
and individual assessments by knowledgeable company or industry representatives as well as our own analysis and judgment.
Some companies are more cooperative about providing information than others and some companies decline to provide or validate
the accuracy of any information. Although the information provided is obtained or compiled from sources VLSI Research believes to
be reliable given the oftentimes difficult circumstances under which it is collected, VLSI Research cannot and does not warrant or
guarantee the accuracy, validity, truthfulness, timeliness, or completeness of any information or data made available to you for any
particular purpose. In no event will VLSI Research be liable to you or any third party, whether in contract, tort or under any other
legal theory, for any direct, indirect, special, consequential or incidental damages, or any other damages of any kind even if VLSI
Research has been advised of the possibility thereof.
We receive letters and e-mails on current topics covered in our services and/or reports that are of interest to our subscribers, as well
as comments on our reports. We value that subscriber input and like to use it. By submitting such material to us, unless you tell us
specifically not to publish it, or except to the extent that you give us an embargo date before which you instruct us not to publish it,
you authorize us to publish and republish it in any form or medium, to edit it for style and length, and to comment upon or criticize it
and to publish others’ comments or criticisms concerning it, as the case may be.
This report may contain information concerning stocks that is obtained from the opinions of industry analysts. Quoted past results
are not necessarily indicative of future performance. None of the information should be seen as a recommendation to buy or sell any
securities. We are not stock analysts or investment advisors. You should contact a registered investment advisor as to the nature,
potential, value or suitability of any particular investment action. No information provided is investment advice and any such
information is just an opinion and is not tailored to the investment needs of any specific person. Certain statements in this report,
other than statements of historical fact, and other written or oral statements made by VLSI Research may be forward-looking. In
some cases, you can identify forward-looking statements by terminology such as "may", "will", "should", "expects", "intends",
"plans", "anticipates", "believes", "thinks", "estimates", "seeks", "predicts", "potential", and similar expressions. Although VLSI
believes that these statements are based on reasonable assumptions, they are subject to numerous factors, risks and uncertainties
that could cause actual results and outcomes to be materially different from those stated or projected. Those factors, among others,
could cause actual results and outcomes to differ materially from the results and outcomes stated or projected in, or implied by, the
forward-looking statements. You should understand that forward-looking statements are not guarantees of results or outcomes. New
risks and uncertainties arise from time to time, and VLSI Research cannot predict those events or how they may affect you, the
reader. VLSI Research Inc does not have any intention or obligation to update forward-looking statements after the date of this
report.
No part of this report may be used in any legal proceedings nor may any of these materials or the information contained therein be
disclosed to any third party, including investors or affiliated firms belonging to investors, outside directors or to your affiliated
companies, or reproduced or transmitted to any third party, in any form or by any means – mechanical, electronic, photocopying,
duplication, microfilming, videotape, verbally or otherwise – without the prior written permission of VLSI Research.
Copyright © 2019 VLSI Research Inc. All rights reserved.
The Chip Insider® and the CSS 10 BEST logo are registered trademarks of VLSI Research Inc. All other trademarks, service
marks, and logos are the property of their respective owners.

... See less