How Japan Won the Lithography Industry
(& Why America Lost)
This is an older video. If you want to watch it, it is below
The Japanese VLSI Project was studied for many years thereafter. In some ways, I think researchers drew the wrong lessons from it. To me, the project was a unique thing that worked at the right time. I don’t see it as a map towards persistent success.
The Koreans tried to replicate its success with their own VLSI-style project. I might have mentioned it before in a prior video. There is some dispute about the results, but the consensus does seem to be that it did not yield the same results.
In 1978, 70% of all the world's lithography equipment came from an American supplier. As late as 1982, Americans still held 62% of the market.
Seven years later in 1989, Japanese firms held 70% share of the market - led by their two lithography giants: Canon and Nikon. The American once-market leaders, rapidly declining. One loses $100 million by 1986. The other withdraws from the market entirely by 1989.
The dominance of the Japanese lithography industry stunned the semiconductor world. Americans back home spilled gallons of ink, trying to figure out where it all went wrong. The answer, as always, is not what you might have expected.
In this video, a prequel to my ASML video, we are going to look at Japan’s famous cross-industry effort to develop an indigenous semiconductor industry and conquer the global lithography market on the side.
The US Creates the Market
Our story begins in America in the late 1950s. Scientists at Fairchild and Texas Instruments create and commercialize the first silicon-based integrated circuits. They are able to create, in volume, chips that are smaller, faster, and more power efficient than anything that has come before.
The United States government and military are immediately interested in this market. They order these chips for the Apollo space mission and the guidance computer for the Minuteman-II ICBM.
American demand for better chips was insatiable. Heavy sales volumes and government funding pushed the pioneers of Silicon Valley to descend the cost curve. To improve yields and performance. And so, the Americans first came to seize the global lead in silicon manufacturing.
The First Lithography Leaders
At the start, everybody made their own equipment in-house. Over time, the demand for newer and better chips forced the integrated semiconductor manufacturers to turn outwards for better equipment.
I spoke about this in my piece on Applied Materials, but what those manufacturers realized was that outside equipment vendors were able to bring more resources and effort to bear onto a very specific, very specialized problem.
The lithography exposure market grew in this way, with a variety of companies, most of them American, entering the market.
In 1973, a company called Perkin-Elmer introduces the Micralign, the first projection aligner. They developed this based off of projection work done for an Air Force research contract.
At its release, the Micralign produced far better defect rates than its competitors. With it, manufacturers improved yield rates from 10% to 70%, making it worth the device's higher prices. And thus, Perkin-Elmer emerged as the industry leader.
But uneasy lies the head that wears the crown. Projection aligners are great, but they had a weakness. Because they projected a chip design onto the entire wafer, they were limited in the amount of detail they could project.
A new generation of lithography tool began to emerge onto the market that addressed this weakness. This tool, rather than projecting across an entire wafer, moved across it step by step. Thus the name: Steppers.
In 1975, Geophysics Corporation of America or GCA, a former map-making company, acquired a small precision-motor manufacturer called David Mann. Using this technology, GCA was able to ship the first commercial stepper to IBM in 1977.
Steppers offered defect rates and yield rates superior to those from the Perkin-Elmer projection aligners. However, they still had two very significant weaknesses.
One, they had really slow throughput, since it takes so long to step across a wafer. This means manufacturers needed to buy and use more of them to make the same number of chips.
And two, a stepper cost $500,000 each. Half a million dollars! Over 30 times more than what the best Perkin-Elmer projection aligners cost at the time!
GCA's big customers - IBM, ATT, Fairchild and National Semiconductor - were willing to pay for the better performance. GCA’s revenues reached $62 million in 1976, an amazing achievement. But all the other smaller semiconductor makers, including those outside the United States, got shafted. Remember that part.
So what has happened here is that we have a market incumbent, quite successful but with clear flaws. There exists a brand new technology with clearly superior performance, but right now has significant flaws of its own and is far too expensive to win the market.
What's left is a market opportunity. A perfect spot for someone with the right mix of price and performance. But who would be able to meet that need?
Japan Falls Behind
Now it is time to head to the Land of the Rising Sun.
Throughout the 1960s, the Japanese government promoted a policy of semiconductor technology transfer. If a foreign company wanted to enter the then-lucrative Japanese market, they need to first establish a joint venture with a local player. The local player would absorb as much knowledge as they can from the foreigner, with the hope of eventually replacing them.
This allowed the Japanese semiconductor market to stay up to date with American semiconductor design and manufacturing practices. But this massive transfer of knowledge also obscured the fact that the Japanese themselves were not inventing semiconductor technologies of their own.
There were, of course, a few pockets of Japanese success. For instance, semiconductor assembly equipment. By now, the American semiconductor manufacturers had off-shored their low-level assembly work to Southeast Asia.
For political reasons, Japanese firms could not do the same, forcing companies like Shinkawa and Disco to build better performing assembly machinery that more heavily automated the process. This equipment was better than anything else out there and soon won the market from the Americans.
But these successes were rare. The reality was that the Americans still made the best computers, the best chips, and the best lithography equipment. And as demand for those items ramped up, the distance between the Americans and the Japanese only grew. Texas Instruments alone spent more on semiconductor R&D than Fujitsu, Hitachi and NEC combined.
Japan vs. the Monster
Stuck between the lithography dominance of GCA and Perkin-Elmer, as well as the computer making dominance of IBM, Japan started to feel anxiety throughout the 1970s over the lagging competitiveness of their computer industry.
Within the ruling Liberal Democratic Party, Japanese politicians heard news of a "one chip computer" from IBM and saw it as another Commodore Perry black ships event. A point in time when Japan needed to see that they were hopelessly behind and had to catch up with the West.
Powerful LDP politician Hashimoto Tomisaburo famously said, "We have too many computer makers in Japan to cope with the monster, IBM ... the reorganization of the computer industry and the establishment of a more unified and more integrated development organization for VLSI technology are urgently needed."
The LDP pressured the powerful Ministry of International Trade and Industry or MITI to reorganize Japan's computer industry and close ground with the West.
Thus, in 1976 the VLSI project was born. VLSI standing for Very Large Scale Integrated.
Not to be confused with the identically named 1978 DARPA project out in the US.
The VLSI Project
The Japanese VLSI project was budgeted to spend a total of $281 million USD - half of those funds coming from the government - and would last for four years.
The end date of 1980 was carefully chosen. It was the year that IBM was expected to release their "one chip computer" (which never saw the light of day, FYI).
The project's goal was to bring together Japan's various computer players to develop the next generation of semiconductor device technology - VLSI. They would all get to share in what they learned.
The companies were Fujitsu, Hitachi, Mitsubishi Electric, NEC, and Toshiba. Computer companies only, so consumer electronics companies like Sharp and Sony were excluded. MITI wanted only the very best R&D-intensive computing companies involved.
So kind of like the Avengers of Japan's computer industry. And also kind of like the Avengers, the superheroes fought a whole lot.
They were different companies in different cliques, staffed with people who normally competed against each other.
The project's managing director Masato Nebashi would later write, "They made no attempt to disguise their hostility; they discussed without disguising their selfish desires."
For instance, the project would establish two types of labs. A core "cooperative lab", where all the members met to work together on basic, fundamental research. It would also be the project's main office. And a bunch of "group" labs for the various company members to research applied technologies.
The group labs were chosen without issue. But the five members of the Japan Chip Avengers hotly disputed the location of the cooperative lab for six months. It was eventually placed in NEC's central lab in Kanagawa.
Here’s how the consortium worked.
The VLSI Project set up a board of directors, staffed by each of the five companies' presidents. Below was a managing director - the aforementioned Nebashi.
Under him was a general committee, staffed with the companies' directors.
Below that, an operating and technical committee. These committees were filled with the department heads of each of the five companies.
Based on the technical committee's guidance, goals would be set for the 100 or so employees loaned from the companies to research and achieve. Critically, the loaned employees were not chosen by the companies, but rather by the head of the cooperative laboratory.
The project had a variety of research goals. They were lithography technology, crystal technology, design technology, processing technology, testing technology, and device creation. They were doled out to the various companies to work on.
Team of Rivals
Nebashi was a retired MITI bureaucrat with decades of experience in managing national cooperative projects. He knew to leave the science to the scientists. His job was to manage the humans:
> I did not interfere in the research itself. My great interest in the organization was the human problem: how to coordinate the researchers from different companies and make them interact. I wanted them to become good friends, communicate to each other, and open their hearts ...
> So, what I did was the typical Japanese way: All I did for this four years was to drink with them as frequently as I could. I wanted to understand their complaints on those occasions and tried to eliminate problems.
But it would be a mistake to think that Nebashi was a drunken therapist for angry employees. He was considerate of people's feelings, but pushed them to stay the course and realize the project's mission. When problems came up, he made it his job to fix them.
Nebashi succeeded in his goals. The various researchers eventually ended up being friends. Upon the Project’s dissolution in March 1980, they held a big farewell party. The expensive equipment was divided up fairly, and people even started an alumni newsletter.
The whole project would yield over a thousand new patents in all of the technology research goals. It greatly advanced Japan's indigenous knowledge of semiconductors and their manufacturing, turning them into more than just fast-followers.
Competitors worked together towards greater things. 16% of these patents were filed as joint inventions by members of competing companies, a huge improvement.
Critically, a lot of these patents came in spaces where a single individual company knew they had a problem with, but did not have the resources to tackle all by themselves. For instance, the work to keep a silicon wafer flat during processing.
Beyond that, there was one other big benefit from the VLSI Project. Of the six major research goals, the project got the farthest with lithography technology.
At the end of the four years, the project built three kinds of lithographies using electron beams. This allowed for resolutions of 1 micrometer or less.
As the research consortium progressed along their lithography efforts, they began commissioning equipment from suppliers.
Lithography is a multi-disciplinary effort. In order for a company to get really good at it, they need to be good in three separate fields: High-resolution optics, advanced electron beam or X-ray lithography techniques and high precision mechanics.
The VLSI Project members eventually settled on working with a couple of Japanese companies with those special experiences. Their names were Canon and Nikon.
Canon and Nikon
Nikon was more ideally positioned than Canon, with expertise in both optics and mechanics. They had been selling medical microscopes since 1925 - optics, check - and made massive telescopes for observatories - high precision mechanics, check.
Canon did not have Nikon's experience in high precision mechanics. But they did have entrepreneurial gusto. Back in 1970, the company began developing aligners for the IC industry - the Micron Project.
Their first aligner, the PLA-series, had already been in the market for three years when the project members came calling. But that first product - the PLA 300, introduced in 1973 - was not market-competitive, due to Canon engineers' deficiencies with high-precision mechanics. It left the aligner with accuracy issues.
The VLSI Project contacted Canon and commissioned a type of aligner called a proximity aligner. This uses an older lithographic technique that has a gap between the photomask and the substrate. The fruits of that labor, the PLA-500 and 600 series, were a massive hit and certain machines were used for decades thereafter.
Canon hoped to also get the chance to work on steppers too. If you recall from earlier, the stepper is the next-generation device from GCA that cost an arm and a leg.
But Canon did not have the resources to do both projects. Thus, Nikon was commissioned in 1977 to work on a lithography stepper. They had never done anything like this before, but the VLSI project members worked very closely with them on the requirements.
How VLSI Helped
It is critical to point out that Canon and Nikon were not part of the Japan VLSI Avengers. The two companies were not privy to internal discussions between the various superheroes. And the VLSI Project never told them to go out into the world market with whatever was built for them.
What the Project did do for them was twofold.
First, they served as a foundation customer for whatever Canon or Nikon would make. This is a bigger deal than at first glance. Companies venturing out into new industries and product lines often face a dilemma: "How do we get the money to develop this product? Who's going to pay for our crappy first iteration?"
The Japanese government basically said that they would "invest" in the final product. They would be willing to pay for crappy first, second, and third iterations en route to the final product.
This is much like how Intel, TSMC, and Samsung contributed billions of dollars over many years to fund ASML's laborious journey towards commercializing EUV.
Second, the VLSI members clearly laid out what they were looking for and was willing to closely work together with the vendors so to achieve it. This matters. Who has ever worked with a client with no idea of what they are looking for?
Nikon first started into stepper development by purchasing a machine from GCA. GCA shipped it to them because technology exports to Japan were allowed. Nikon tore it apart, studied every piece, and then cobbled together a Nikon prototype that worked in the same way.
Naturally, that first prototype worked terribly. But rather than throwing the whole thing into the trash, the VLSI Project members committed to working closely with the company to make it better. After the project disbanded, both Canon and Nikon now had products that they knew could win the market.
The Japan Market Explodes
The VLSI Project succeeded in kicking off Japan's indigenous semiconductor industry - building massive fabs to make state-of-the-art memory chips. From 1980 to 1982, Japan's lithography industry grew 66%. The United States by contrast grew only 10% during this same period.
Japanese firms leveraged their skills at optimizing industrial and manufacturing workflows to surge into the market lead - beating their American counterparts to 64kb DRAM. There is a lot more to say about this, perhaps in another video, but the industry in Japan won by doing what the West was doing, but much better.
It was the same with lithography. The Japanese were able to make older technology go farther and at better prices. Canon shocked the industry by developing a proximity aligner capable of reaching 2-3 micron design rules, which many experts thought was not possible.
Then in 1980, Nikon released into the market the fruits of their work with the VLSI Project: the NSR-1010G stepper. The first customers were NEC and Toshiba, who found it perfectly suited for their current production lines.
The Japanese market's explosive growth caught the American lithography makers off-guard. 1981 was an American recession year, and the Americans were not attuned to the market - relying on the Japanese trading houses for market intel.
GCA tried to ramp up their production, but their sole optics supplier Carl Zeiss (remember them?) could not get lenses out fast enough. As delivery delays stretched on and on, the Japanese semiconductor makers could not wait any longer.
They defected, and Nikon snatched 20% of the market in a single year. American market share for every category of semiconductor production equipment collapsed. The war was over. The American lithography industry shriveled and died.
I suspect people watching this are scrutinizing this history very closely for insights about the present and the future. Perhaps, regarding another lithography situation underway somewhere in East Asia.
But I really caution against that. The players are not the same. The market scenario is not the same. And most importantly, the technology is far from the same.
The Japanese victory was not complete. Because as we all know, it would not last for long. That's why prequels suck. Canon and Nikon could not hold onto their lead in the lithography market. The Japanese semiconductor market would collapse in the late 1980s and 90s.
The Americans were moved to respond to their loss of market leadership, plunging forward into the research to develop revolutionary semiconductor production technologies and applications. They loved the VLSI Project idea so much they made their own - SEMATECH. There, they hit upon the next generation of lithography techniques: EUV.
By now, the American lithography industry no longer existed. So SEMATECH and the government turned to the next best alternative: a small up-and-comer in Europe called ASML.