Steve Blank on the Chip Wars of the 21st Century
warontherocks.com
warontherocks.com
The fundamental problem is that it is extraordinarily expensive to build a state of the art chip fab and it is extraordinarily difficult to make start of the art chips (process technology in the 7nm node range). The technology is heavily patented and, what isn't patented, is tightly controlled as trade secrets by fabs.
TSMC[1] started as a contract fab house with competent, but not state of the art, fab capabilities. Over time and with immense investments, they have become a leading (arguably the leading) fab. Intel is on par (arguably) or has fallen slightly behind TSMC in terms of state of the art fab capabilities. Global Foundries[2] (made up of former IBM foundries[3], Motorola foundries, and others, acquired by AMD and subsequently spun out) has fallen behind in terms of state of the art (e.g. 7nm nodes). Europe also has quite a few competent, but not state of the art, foundaries.
There are a ton of fabs that were state of the art but have not been upgraded - they are quietly fabbing billions of chips that don't need smaller geometry fab nodes and cannot justify the expense of moving to more state of the art fabs. "Moving" to a new fab node technology requires a significant redesign of the chip and would require requalifying, both of which are very expensive.
Ref:
https://en.wikipedia.org/wiki/List_of_semiconductor_fabricat...
[1] https://en.wikipedia.org/wiki/TSMC
[2] https://en.wikipedia.org/wiki/GlobalFoundries
[3] Trivia: IBM paid Global Foundries to acquire their (IBM's) chip foundries.
This is why chip "yields" change over time - the process is slowly tweaked (maybe the temperature get changed 0.001 degree or something) to improve yield.
Chip manufactures use a mix of patents and trade secrets to protect their advantages.
That is what makes me dislike the idea of patents - if two or more teams working in silos on developing and refining the process would have gotten similar results, then the process should not be patentable.
> It would seem that European countries have great education and economy and yet we cannot see any innovation in this space?
Some of the important hardware for chip manufacturing is developed and produced in Europe.
I don't understand why the US Government can tolerate losing the edge in this area to other nations. Maybe its because TSMC is based in Taiwan, which is an American Ally?
It seems like one would want to retain these skills even if they may not be profitable. Are there military contractors dedicated to manufacturing microchips?
Because the US is a Capitalist state. It serves first and foremost the interests of large corporations and capital holders operating in the US. As (some of) those have become less interested in nationalism and are multi-national - including a few that are not US-based to begin with - the incentive for maintaining "national edge"s wanes.
Each machine costs $100 - $255 million. ASML ships 35 this year and 45-50 in 2021.
TSMC define their processes in great detail: the exact timings of photolithographic processes, the exact composition of the chemical processes, all the little material handling details, wafer sawing, packaging, wire-bonding, testing and so on.
They also define standard cell libraries, the little blocks of rectangles that make up the gates. Given the way patterning geometry works, if you want your chip to actually work you realistically need to use their libraries.
For major customers there will be a certain amount of tech support, especially on a new process; inevitably the first few wafers through will have terrible yields and it's to the benefit of both parties to work out why.
A factory - any factory - is not just the set of capital equipment that makes it up, but the collection of staff that run it, including their often unwritten expertise on how the process works in ways that may differ from the theory.
Key steps: floorplan, place and route, layout.
It's automatic with manual tuning and fix-up; just as PCB autoroute isn't perfect. Getting timing closure can require a lot of work; picking the right size of cells to drive the length of wire and fan-out is a serious problem.
Also even you got your hands on them, you need a sufficient expertise and huge investments to build a fab. If you are part of the US block, then it's too expensive, if not, you can't buy, so you have to develop them yourself, which takes a lot of money and more importantly time. Even with all those money poured into this area, China plans to become self-sufficient on the chip front not earlier than 2030.
I think the unique thing about TSMC is that they're an "unaligned" fab, not part of one of the giant chip companies. They'll take contracts from anyone. This leads to their name getting mentioned a lot by all the fabless semiconductor companies, while Intel, AMD, Samsung etc. turn out billions of chips with zero news coverage.
They are the free market in chip manufacture.
Edit: On the note of Europe, Philips was one of TSMC's original investors and customers. Philips also co-founded ASML which other people have mentioned.
I believe the author's position is that previously, one of the only considerations was, "how much will it cost." Or, "Does it make sense for us to invest in X fab?" Now there are other considerations beyond the economic and I'm certain you will begin to see not only fabs, but the full chain, in the EU and China. Probably sooner rather than later.
What does that mean for the different economic blocks?
What does that mean for tech in general?
What does that mean for global security?
Do we, (the US), have sufficient creativity to wargame all that out and get an idea of the future?
Etc
I think that's what the article is about.
Of course these countries want to keep the US, China, Russia, etc "interested" to give them more goodies - tech/markets/access/etc. The more suitors, the greater the competition, the more plentiful the gifts.
Vietnam is doing this right now with china and the US and getting investment/political favors/market access/etc from both.
Currently at 65 nm, good enough to manufacture chips like these: https://en.wikipedia.org/wiki/Elbrus-2S%2B
Russia still depends on TSMC for chips at 28nm like these though: https://en.wikipedia.org/wiki/Elbrus-8S
In comparison, here's BAE's RAD750 at 150 nm: https://en.wikipedia.org/wiki/RAD750
65 nm is plenty good enough for rad-hardened military equipment, though good server chips still have to be outsourced to TSMC.
In comparison, China's SMIC is at 28 nm CMOS and ramping up 14 nm FinFET processes but it depends on foreign companies for lithography machines. China's SMEE has a 90 nm lithography machine and announced that they're delivering a 28 nm machine sometime next year. I can't seem to find any Russian lithography companies though.
Also there is a big segment of small chips (e.g. RFID) for which a bigger process is cheaper. IIRC most of public transport tickets and cards in Russia are produced locally.
One aspect that is less explored is the actual depth and breadth of the semiconductor supply chain. Moreover, that supply chain is bound with technical expertise at every level. You may have the fastest car at Le Mans, but you're not going to win without the best driver.
Chip design starts with design software. The most sophisticated software is made in the US. To use it, you need a variety of highly experienced designers of the right ilk (think neurosurgeon vs. cardiologist), supported by program management, organizational management, test and validation engineering and other disciplines. This synthesis of expertise and tools and equipment has to be at the highest level of capability all the way up and down the supply chain before you can get a 7nm chip to boot up.
Additionally, there are about 40 different types of process equipment used in chip fabrication. Every one of them requires dozens of product development engineers to design and build. Those teams rely on decades of both personal and organization experience to build on before they can produce tools that can, for example, apply a material like, say, tungsten, to a 12" slice of pure silicon in a layer that can be precisely measured in atoms, uniformly, across the 'wafer'.
Of all the 40 process tool types, the most complex work of science is the lithography tool made by ASML in the Netherlands. This thing produces the light needed to print the chip circuits at 7 nanometer geometries by emitting droplets of molten tin into a chamber and hitting the droplets with a 25,000 KW laser that vaporizes the tin and produces the light, firing at 50KHz[0]. ASML has spent more than 20 years, employing 100s of engineers and scientists, in order to produce this wonder. There is no company (and certainly no government) on Earth that can catch up to ASML in any reasonable amount of time.
Next in the chip production chain is called 'Assembly' where the wafers are cut into individual chips and packaged into what you'd recognize. That node in the chain has it's own set of non-trivial equipment and expertise.
All of this is to say that there are vulnerabilities all up and down the supply chain, bouncing all over the planet. No one country has the technology to build state of the art chips from both the supply chain and expertise standpoints all within its own border. I'd guess that the least number of nations to do this would be the USA + Netherlands. But as Steve Blank alludes, it would take the US a decade to complete the development and ramp the capacity just to meet its own demand, and 100s of billions of dollars.