Hacker News Reader

Hacker News

Comment by brcmthrowaway | Hacker News Reader

Full thread

brcmthrowaway

What exact IP makes TSMC so valuable?

We know the EUV machines are produced by ASML

We know ASML optics are produced by Zeiss.

So what is it?

View on HN

It's probably not just a piece of IP. The question of competitive advantage (or comparative advantage) by companies (IBM in the 70s, Microsoft in the 90s, Intel for decades, TSMC, nVidia, and Apple recently) or by nation states (Japan in the 80s, Taiwan and S. Korea recently) is complex and often mysterious. Economists talk about core competencies, which is a combination of culture, training, work-ethic, supply chain, IP, and other things. If teasing these apart was easy, competitors would emerge more quickly.

Equally fascinating is why it goes away (Japan in the 90s, Intel recently).

I have been trying to figure out TSMC in particular but haven't yet, and I'm skeptical of anyone who claims it is one simple thing, or even anyone who claims they know all the reasons.

They built the supplier global and local ecosystem and they have a local pipeline for engineering and technician talent. The US has good graduate and university systems but the industrial culture is different and you see that in operations. There are different workplace incentives.

They’ve been working for a long time to reach this superior position and made good decisions business partners and investments along the way. These investments in local national economies take time, decades. I agree with you, IP is reductive where I think there is a combination of tangible and intangible factors, trade secrets, talent, operational discipline, management and long term thinking.

It’s hard to see in present time, but it becomes clear in retrospect when the winning companies started making failing decisions while on top. Then doubling down on those poor decisions or switching strategies too frequently to make an effect.

>It’s hard to see in present time, but it becomes clear in retrospect when the winning companies started making failing decisions while on top. Then doubling down on those poor decisions or switching strategies too frequently to make an effect.

It is even more infuriating when you see a big player makes a bad decision, but it takes years to see the effects (on the market for example).

It's about the semiconductor wafer manufacturing and processing techniques. Just wafer manufacturing alone, to get the desired electrical characteristics, doping and the lowest number of defects possible is a mighty task. TSMC has perfected this black art. They are head and shoulders above anyone else.

They know how to integrate and operate those and all the other pieces required to go from a hardware design to hardware you can touch. If you just had to take delivery of ASML machines Intel wouldn't be struggling to get below a (so called) 7nm node.

Also TSMC pays far far more than Intel on a PPP basis for the same role so it's not like Intel can poach that talent.

Intel: USD$77k for a hardware engineer role (low i know - Intel have had 20years of wage freezes and they has an across the board wage cut last year). Link: https://www.indeed.com/cmp/Intel-Corporation/job-titles/Inte...

TSMC: Pays the Taiwanese equivalent of USD$58k for the same role. Link (in NT$ but conversions easy): https://www.salaryexpert.com/salary/job/signal-integrity-eng...

Now before you say aha! that's less on a direct dollar to dollar comparison remember that Taiwan has slightly better than 2x rate of PPP.

Contrary to what a lot of people seem to think, TSMC in Taiwan has better pay, working conditions and benefits than Intel in USA. The childcare mentioned in the article is one such example as is the pay i just mentioned. This has been true for a long time now. Intel is the primary Western competitor and they don't compete for talent at all. They haven't done for a long long time now.

I think of it like this:

ASML makes the film.

Zeiss makes the cameras.

TSMC is the movie studio. They'd be making hit movies no matter who was supplying the film and the cameras.

Zeiss - lenses both in analogy and irl

ASML - cameras

Silicon - film

TSMC - film house (manufacturing, color correction, managing storerooms, mastering, distribution brokering with theaters)

Taiwan - Hollywood with just the right climate for talent and business

Fabless chip firms - movie studios

Chip designers - writers

Specialized modular chip IP - actors

… the analogies really break down from there…

Are software engineers the ad agencies and the code execution just the generation of value by playing the movie for people???

I think that's a little too simplistic. They would not be the world leading fab without ASML. ASML's EUV process may be the most complex piece of machinery man has ever created.

It's maybe more like ASML's machines are like a Stradivarius, and TSMC is the best in the world at playing the instrument.

ASML make the tool, TSMC make the process.

The fab itself, for example, you can't just setup a Twinscan in your garage. Even the floors are marvels of engineering. Culture is another subtle but huge factor. When chasing nanometers the fab becomes sensitive to things like deodorant and cigarette smoke. You can of course mitigate these issues, but having an aligned and motivated work force makes it easier.

Forgive my ignorant question, but if it's that sensitive why not automate it with robotics and vacuum seal it?

I understand that this is a sensitive process but so is surgery which robots are able to do with remote control.

They do their best but they can't seal away everything. Raw materials have to enter, chips and waste come out, and the whole thing comes apart for maintenance.

Manufacturing is a game of yields and margins. When you have thousands of operations and inputs, small changes have big effects. TSMC and Intel both have access to the same machines, it's what they're doing with them that differs.

You have no idea how many steps, robotics, offsets, material feeds, alignments, tool setup, recipe fixing, materials troubleshooting go into a fab. These are all to run automated processes. You underestimate how difficult it is to sink heat in a vacuum and the cost of scaling volume and keeping pressure to a low exponent (high vacuum).

Robot surgery is tolerant of a much greater length scale of defect than a modern IC chip and is orders of magnitude fewer process steps.

Try to watch a youtube video or something but even then you won’t see the full picture. There are so many layers and the wafer passes through different flavors of the same processes over and over.

Read this interview, and you will see that it's the work ethic.

https://www.computerhistory.org/collections/catalog/10265812...

https://en.wikipedia.org/wiki/Tacit_knowledge

making a chip is not like hitting 'print'. there is an insane amount of engineering and consulting that goes on. yes, they buy the machines, but they develop the process. and all the tooling around that, including quite a bit of IP.

intel has tried to compete with them - they have their own ip, machines, and processes. but they are totally missing the consultancy and full package access that TSMC has, and they have failed.

In simplistic terms, they buy machine with 1000 knobs. They have engineers who know how much and when to turn each knob.

For the same reason as buying the same guitar used by John Lennon make most of us playing music as good as John Lennon.

Reply on news.ycombinator.com