TSMC’s Announcement Of A U.S. Fab Is Big News

The Wall Street Journal’s report that Taiwan Semiconductor Manufacturing Company (TSMC) intends to build a “fab” (semiconductor fabrication facility or factory) in Arizona is big news, although it is really just a toe in the water. TSMC is the world’s largest contract integrated circuit (IC) manufacturer, and it is far and away the technology leader, offering the most advanced manufacturing processes on the planet. As such, it is a critical supplier for companies like Apple
AAPL
, Qualcomm
QCOM
, Nvidia
NVDA
, AMD, and even Intel
INTC
.

TSMC has a remarkable history

TSMC’s roots go back to the Science and Technology Advisory Group (STAG) established by the Government of Taiwan in 1979. That group included Frederick Seitz, president of the National Academy of Sciences in the U.S. and a member of the President’s Science Advisory Committee, Bob Evans, a key figure in the development of IBM
IBM
’s System/360, and Kenneth G. Mackay, executive vice president of Bell Labs and a pioneer in communications engineering. The STAG played a key role in pushing Taiwan’s Industrial Technology Research Institute (ITRI) to investigate semiconductors. ITRI’s Electronics Research and Service Organization (ERSO) eventually licensed RCA’s CMOS semiconductor process, and sent a team of 35 engineers from Taiwan and five overseas Chinese living in the U.S. to New Jersey in 1976 to learn the technology and transfer it back to Taiwan, where a demonstration production facility was setup. That project was spun off as United Microelectronics
UMC
in 1982, though the demonstration facility continued to churn out 15,000 wafers a month.

In the early 1980s, Carver Mead at the California Institute of Technology began advocating a new approach for dealing with the complexity of very-large-scale integration (VLSI) design, in which the manufacturing of ICs would be separated from the design process. The U.S. Defense Advanced Research Projects Agency (DARPA) initiated projects to develop electronic design tools that enabled the separation of the design process from the manufacturing. This was the birth of the concept of having a separate “foundry” that would contract manufacture chips for multiple customers. 

By 1986, ITRI had completed building a VLSI model factory. Morris Chang, who was at ITRI at the time presented a plan to the Taiwan government to build the first pure-play foundry. Later that year, in partnership with Philips of the Netherlands, TSMC was launched with Chang as the founding CEO. The Taiwan government initially owned 49%, Philips 27%, and local private investors 24%. While other companies, notably LSI Logic and IBM in the U.S., Toshiba in Japan, and Samsung in Korea offered foundry services, they produced their own chips as well. TSMC was the premier pure-play foundry, and invested heavily in building a customer service model that was tailored to the needs of “fabless” semiconductor companies like Qualcomm, Nvidia, and Broadcom
AVGO
who never planned to have any manufacturing of their own.

20,000 Wafer starts a month is the minimum efficient scale

TSMC’s announced intention is for a fab with an initial capacity of 20,000 wafer starts per month. Fabs make ICs on silicon wafers, typically 300 mm (12 inches) in diameter, so that means 240,000 wafers per year. Depending on the size of the chip, a wafer might hold around very 500 large chips like datacenter microprocessors, to many thousands of smaller chips like those found in a cell phone. The production process takes 40 or more days, as wafers have to go through 300+ process steps (and considerably more for some of the most advance processes).

20,000 wafer starts a month is around the minimum efficient scale for a fab. TSMC’s three “Gigafabs” in Taiwan, Fabs 12, 14, and 15 handle around 150,000 per month each (based on the last time they would let me inside for a visit), and the company’s overall capacity is around 2.5 million per month. It’s new Fab 18 which is just ramping up production will produce over one million wafers per year. Thus 20,000 wafer starts a month is “dipping your toes in the water.”

An expensive investment

The $12 billion cost estimate for the Arizona fab should take the breath away of most CFOs. One of the reasons most U.S. semiconductors are fabless is that very few companies have the courage to make that kind of investment or have the confidence they can make the advanced process technology work. The fab would be targeted at 5 nm technology, currently the most advanced process that is actually in production anywhere. One reason TSMC probably has been reticent to establish a fab outside of Taiwan is the need for close production support from its existing teams, and the extent to which it shuttles work across its existing network of fabs which are largely in three science parks in Taiwan.

$12 Billion is a staggering level of investment, and most of that money will be going into specialized production tools that can make the extremely fine circuit patterns on these chips. The key tool in the process comes from ASML in the Netherlands, and that tool weighs 180 tons and costs $120 million each. Yet each one only can process 170 wafers per hour, so you multiple copies of these in each fab. The optical engine inside the tool is produced by Zeiss in Germany. The mirrors in that engine are polished to a surface roughness that is roughly the size of one atom. When I visited the factory a few years ago, I was told that if you used those mirrors to project a spot on the surface of the moon, it would be a pretty small spot.

A start, but this does not secure the semiconductor supply chain for the U.S.

Having a TSMC fab in the U.S., assuming it is built, should not make people think the entire semiconductor supply chain will be back in the U.S. Obviously we will still depend on Taiwan and TSMC for a lot of capacity. We will also rely on Asia for slicing those wafers up into individual chips and putting them into packages.

Decades ago when the IC industry was just getting started, wafers containing chips were sent to Asia for the labor intensive packaging step. Armies of workers looking through microscopes wire bonded leads to the ICs and placed the chips in packages. Today chip packaging is automated, but the expertise continues to lie offshore. It will take time and money to develop the capabilities domestically. The Chinese clearly feels the sector is strategically important, as reflected by Jiangsu Changjiang Electronics Technology’s 2015 purchase of STATS ChipPac, one of the leading providers at the time. As the industry moved towards 3D packaging, it will be a critical technology.

As I wrote earlier this week, Intel might still enter the foundry business, and that would provide some healthy competition. But Intel also realizes the technical challenges and magnitude of the investment required. It will be interesting to watch how this plays out.

Speak Your Mind

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Get in Touch

350FansLike
100FollowersFollow
281FollowersFollow
150FollowersFollow

Recommend for You

Oh hi there 👋
It’s nice to meet you.

Subscribe and receive our weekly newsletter packed with awesome articles that really matters to you!

We don’t spam! Read our privacy policy for more info.

You might also like

CNN’s Jake Tapper: ‘The President Thinks He Is Going...

CLEVELAND, OHIO - SEPTEMBER 29: U.S. President Donald Trump...

New Mahindra Scorpio S3+ launched; know everything about it...

New Delhi: Indian automobile manufacturer Mahindra has launched a new base trim to its...

Trump Hasn’t Attended Meeting Of Coronavirus Task Force In...

Topline Confirming a report from earlier in the morning, Assistant Secretary for Health Adm....

WWE Expected To Bring Back Huge Star For SummerSlam

WWE is expected to bring back a top star...