Book: 20230515 to 20230810, "Chip War" by Chris Miller

20230515 - Introduction

Chip manufacturing industry is extremely high efficent, and extremely fragile.

How long will it take to turn it into distributed manufacturing mode? 5 years (2028)?

20230517 - Part I: cold war chips

From steel to silicon

Good to know where the word "debug" come from: vacumn tube attract moth, and moth cause problem, so it needs "debug". At the end of WW2, people realized that precise attack is much more powerful than large scale attack, but that need fast and low cost computation.

The switch

The breakthrough is inevitable. Why silicon has such property?

Noyce, Kilby and the integrated circuit

This is the low hanging fruits after the breakthrough.

Liftoff
Mortars and mass production
"I...WANT...TO...GET...RICH"

Like "Why Nations Fail" said, innovation needs many necessary factors. After WW2, USA has all of them: money, talents, fair institutions, knowledge base.

20230530 - Part II: the circuitry of the American world

Soviet Silicon Valley

In 1950s, there were low hanging fruits everywhere. USSR only needs knowledge or even just inspiration to replicate the "breakthrough".  Not anymore.

To replicate ChatGPT, it needs the most advanced supercomputers, the most advanced software system, huge amount of data and dozens of highly qualified AI engineers.

"Copy it"

USSR doesn't have inclusive institutions. Most of people there don't try their best working. It's a bit like Spain vs British in 1550. The explorers from Spain work for Spain King, and the people from British work for themselves. We can see the result of Spanish Armada in 1588.

China is similar to USSR. Copying is not a good strategy regarding high tech innovations.

The transistor salesman
"Transistor girls"

Why US chip maker want to build factory in Japan? Because Japan has high quality workers with low wage.

Why Japan could invent new electronic products but US couldn't? Apple finally caught up but that's mainly because of their software development advantage.

Why the software from Japan is so poor?

Computer is powerful enough to enable manufacturing automation now. Especially with the help of AI. Does that mean those developing countries are doomed?

Precision strike

Microelectronics were tested at the end of Vietnam War. High precision striking is at least 1000 times more powerful than traditional bombs.

It's not hard to imagine the power of smart missiles when they are integrated with AI.

Can China catch up? It's not just about chips, but more about the ecosystem. For example, satellites and autonomous drive. AI needs to understand the real world in very short of time.

Supply chain statecraft

No wonder that US didn't want to retreat from Vietnam. All asian countries were watching.

What can we learn from it? Economy(unemployment rate) is more important than millitary power!

Those asian countries are lucky. The semiconductor industry just emerged in time.

Intel's revolutionaires

CPU is designed to replace many different types of specialized chips with low cost general chip, and DRAM make large scale software possible. Together ,they can be used to collect, process and deliver information to billions of people.

The Pentagon's offset strategy

USA has many visionaries. Inclusive institutions create visionaries.

Microelectronics changed war forever.

20230612 - Part III: Leadership lost?

"That competition is tough"

Why is the chip quality from Japan, Korea, and Taiwan superior to that of other countries?

Surely it's caused by their history and culture.

From the other side, they don't make enough effort to build software.

"At war with Japan"

Government direct support and low capital cost (low interest rate and loan get easily get extended) is also critical in helping Japanese chips companies to win the competition against USA.

"Shipping junk"

Companies go up and down in competition. GCA fell, and Nikon rose.

I don't see anything strange in that.

Japanese chip makers also went down in the past 30 years, when competing with Korea and Taiwan.

Will Tesla go down? It's just matter of time. But not that soon. Even in 1987, 10 years after the peak time of GCA, GCA still has around 50% of market in USA.

The crude oil of the 1980s

USA realized the importance of chips at early stage, but there was something the government could not do, because it's against free market.

USA is already the victim of globalization by that time. I think customs tax is the only solution.

Software development is always the weak point of Japan. Weird.

Death spiral

SEMATECH was set up too late to save GCA. Huge advantage may not last for more than 3 years, and no one can save it after 10 years!

The Japan that can say no

It's interesting to see how people get confused by the information from surface.

It's hard to compare two complicated system fairly.

Maybe we can use game theory? Which system create more innovation? Which one give people more fair reward for their hard thinking and work?

20230625 - Part IV: America resurgent

The potato chip king

The common features of potato chips and computer chips? Large quantity, small products.

Cutting cost and improving quality are the way to win competition. I think this is the same strategy to many manufacturing business.

Fear is the ultimate motivation. If all employees try their best to cut cost and improve product quality, they can win the game. Just like Tesla?

Disrupting Intel

Only the paranoid survive? Maybe. But Intel was very lucky too.

Intel had chance to copy Micron's strategy, but it didn't.

"My enemy's enemy": the rise of Korea

Korea is lucky.

"This is the future"

Using computer chip to design and build computer chip, that's the key of Moore's Law.

Same thing will happen with AI. When will that happen? 2029 or 2045?

The KGB's directorate T

USSR is 5 years behind USA regarding silicon chips. Is that enough?

Why USA allowed Samsung and Micron to build new chip manufacturers in China? I guess USA wants to stop China building any basic devices for even low end silicon chip industry, instead, relying on the devices that Samsung and Micron bought oversea.

It's all about the machines that build machines, no the products.

"Weapons of mass destruction": the impact of the offset

Without the silicon chip industry, long range weapons don't have precision, and even rader is primitive.

How about China?

I think that satellites are the computer chips 40 years ago. The in-orbit costs is the key.

What's the next one? Maybe AI. If Tesla can build a supercomputer with 100 ExaFlops, surely USA is far ahead of other countries.

War hero

"The Cold War is over and you have won"

Why did Russia invade Ukraine? It should know that it has no chance to win against US.

Why doesn't US stop Russia from expanding?

20230705 - Part V: integrated circuits, integrated world?

"We want a semiconductor industry in Taiwan"
"All people must make semiconductors:
Sharing God's love with the Chinese"
Lithography wars
The innovator's dilemma
Running faster?

US was overconfident in the first decade of 21st Century, and China wasted this opportunity.

But I think US's strategy is correct. Allowing allies to exclusively control some critical technology shows that US trust them, so those allies can also trust US.

I guess EU, Japan, Korea, US, etc. will soon build new alliance against China. What would Chine do in that case?

20230720 - Part VI: offshoring innovation?

"Real men have fabs"
The fabless revolution
Morris Chang's grand alliance
Applie silicon
EUV
"There is no plan B"
How Intel forgot innovation

The complexity of EUV lithography is insane!

How is it possible to develop next generation of EUV machine?

"Use infinite money for solving an impossible problem" p225

Well said.

One thing for sure: whoever want to make or use the cutting-edge EUV machine, needs to be friend of all major developed countries.

Intel's strategy is not that bad. At least it sounds not bad. But since they cannot expand their product scope, the cost of EUV machine is too high to them. For TSMC, there are many big customers: Apple, Nvidia, AMD, Qualcomm, all Android phone manufactures except Samsung, and many small chip design companies. EUV machine is $1B each, which means around $1M per day in 3 year lifespan.

So, the real problem of Intel's strategy is that they didn't realize how many types of high demand chips will be there.

20230731 - Part VII: China's challenge

Made in China
"Call forth the assault"
Technology transfer
"Mergers are bound to happen"
The rise of Huawei
The 5G future
The next offset

Globalization means integration and dependency.

China says it support globalization, but want to remove dependency on other countries. That means it actually support anti-globalization.

AI needs both hardware and software. It needs creativity, and creativity can only come from fair institutions.

Not likely that China get a chance to win over US.

20230810 - Part VIII: the chip choke

"Everything we're competing on"
Fujian Jinhua
The assault on Huawei
China's Sputnik moment?
Shortages and supply chains
The Taiwan dilemma

Surely China will be able to produce low end chips independently, maybe in 10 years.

What's the difference between missles loaded with high end chips vs the one with low end chips?

In what case will China decide to block Taiwan? When the damage is much greater to the world than to China itself? However, start from the first day of that conflict, China will lose access to any high end chips from Korea, Taiwan, Japan and US. What will push China to do that inspite of all the potential loss?

What push Putin to invade Ukraine? Overconfidence?

China won't repeat the mistake from Russia. Hopefully.

What's the ultimate speed to silicon chip? At the moment, it's around 10B transistors per square centimeter, which reach around 100TFlops/s. Jim Keller said the transistors could be 50 times more, which means 500B per square centimeter. Let's assume the speed is 1PFlops/s. Superconductors could make it 100 times faster, I guess. So, the ultimate goal is 50PFlops/s.

Let's say we can integrate 1 million such chip in a pod, then the speed may reach 10^24

Can we get there in 10 years(2044)?