My understanding is that 1 company in Taiwan makes the greatest chips in the world and no one else can replicate them. How is that possible?
In: Engineering
In addition to all the other things everyone else has said, features on top-of-the-range modern chips are just a few atoms across.
Everything is super sensitive to the environment at these scales and the kind of physics we are used to in our macroscopic world is overwhelmed by considerations like quantum effects.
People can do this stuff in the lab, but that is very different to doing stuff profitably at scale. When your yield rates are affected by e.g. a roadworks two blocks over, you have the manufacturing process from hell. It takes a long time, a lot of trial and error and a huge investment to get everything working smoothly; it’s not just a matter of doing the right thing, but also of working out what the right thing to do even is given what it is like where you are and what is around you. Since the outcome is the result of many many iterations of trial and error, people may not even be explicitly aware of which tiny details matter and are critical to get right.
It is certainly possible to set up a workable fab node elsewhere, and others do exist, but it would take a vast amount of time, money and effort to go from building site to commercially profitable facility even if you had complete access to TSMC facilities to use as reference.
Intel and Samsung can. The ‘nanometer’ thing is mostly marketing fluff. It actually doesn’t refer to any single measurement on a transistor. Thus, there is no part on a 3nm chip that is 3nm. TSMC just has the largest capacity to produce a chip that is in the ‘3nm class’.
Is it telling that my first thoughts were of
A) potato chips, and I was confused because, well, who would associate Taiwan with Potato Chips, and what have I been missing out on? OR,
B) poker chips, and I was wondering if this were somehow influenced by the cultural predilection most Asians seem to have for gambling.
Just to add some flavour to the mix here and demonstrate how frickin’ tricky the chip fab thing is, here’s how NXP’s fab got utterly wrecked by a little cold weather:
[https://www.nxp.com/company/blog/through-the-storm-the-complex-process-of-restarting-a-semiconductor-facility:BL-RESTARTING-SEMICONDUCTOR-FACILITY](https://www.nxp.com/company/blog/through-the-storm-the-complex-process-of-restarting-a-semiconductor-facility:bl-restarting-semiconductor-facility)
These factories are some of the most insanely precise and sensitive places on the planet, they cost a billion dollars to set up, as TFA says it takes a WEEK to safely shut one down for cleaning and 2 weeks to safely start it all up again and to get the air clean enough that they can even *think* about making chips.
Semi conductors manufacturing is the most difficult mass production on the planet.
It requires tremendous amounts of investment to build and operate a fab. Each successive semi-conductor generation (often called a “node”) is getting more difficult and expensive to design and produce, with the possibility of it being delayed / faulty increasing (this happened with Intel’s 10nm process). As a result, a lot of companies dropped out, and at the bleeding edge we’re left with 3 main players: Intel, Samsung, and TSMC.
Currently, TSMC has the best process – right now it’s being used for the new iPhone Pro and Apple M3 series and should expand to more products from AMD and and Nvidia by the end of the year.
There’s also one main company left producing the machinery to make these highly advanced nodes economical: ASML. These machines are very complex, expensive, and limited in quantity per year.
TSMCs leadership role isn’t a guarantee. Intel and Samsung aren’t THAT far behind, and Intel in the last few years has dumped tremendous amounts of money into R&D (more than AMD, Nvidia, and TSMC combined) in an effort to close that gap. Intel claims they will take leadership from TSMC within the next 18 months of so (but this is certainly not a guarantee and remains to be seen).
Another aspect is that Intel is only now just beginning to open their fabs to external design companies. For a long time, companies like AMD, Qualcomm, Nvidia, etc. Went with TSMC because they were the best available for them. A few years ago, Global Foundries was close enough to be considered, but they’ve since dropped out of the leading node market, and last gen Nvidia sourced from Samsung because their 8nm node was “good enough” and cheaper than TSMC’s 7nm node and Nvidia was able to compete due to their extremely good architecture design that offset some of this discrepancy.
TSMC, the company in question doesn’t design chips, and is dependent on a Dutch company for the machines.
That’s not to say they don’t have their own process, one that’s given them a competitive edge, but it’s not like other “foundries” (the name for this type of company) couldn’t achieve the same result next year. TSMC has been ahead of their competition, but they do have competition.
Said Dutch company (ASML) however does not. No one can make current gen high end chips without a machine from Eindhoven, and despite the Chinese stealing company secrets, that’s unlikely to change any time soon.
The best analogy I can give you is putting screen protectors onto phones. Bubbles are the result of a tiny piece of dust that you cant see. That’s why it’s so hard to put a screen protector on without bubbles. Now imagine a Walmart sized warehouse where the entire place must not have a single spec of dust, and trying to put a warehouse sized screen protector onto a giant phone. That’s how difficult it is to build chips. The equipment is super sensitive. An extra microscopic drip of moisture in the air will result in a laser going off course and burning something in the wrong place. Equipment like this is super expensive, and years to build by hand. Even if you can put together a facility, you need to do it quickly enough where your equipment is not obsolete by the time it’s done. Someone will be the leader. TSMC is the leader today, but it may not alway be the case.
As other responders have said, it’s not that other vendors don’t have leading edge transistor technology. So what TSMC does can, to a degree, be replicated. But what is very difficult to replicate is the combined *volume* and *reliability* at which TSMC does it. It’s a weird side-effect of semiconductor manufacturing: the more volume (number of wafer starts) you have, the more reliable (higher yield) the semiconductors produced will be. TSMC has been a leader at this for 30 years, putting a steady percentage of their profits into technology and reliability improvements that bring them more and larger customers, which bring them higher profits and … around it goes. The only other companies which achieve a similar degree of volume and quality are the 3 remaining DRAM and Flash makers: Samsung, SK Hynix and Micron. There are no non-DRAM, non-Flash foundries that are better than TSMC in terms of volume and reliability.
Its not
TSMC is a factory / machinery provider, other people buy time on its machinery and make THEIR designs.
TSMC doesn’t design chips, it simply mass produces them cheaply and reliably.
It also doesn’t make the machines it makes chips with, it buys them.
So REALLY, TSMC just buys a huge amount of chip making machines, employs a very skilled workforce to use said machines, and leases those machines operational hours on favourable rates to others.
Chip foundries are expensive and to be profitable you need to run them 24/7/365
Intel makes 70% of the worlds PC’s it can run its fabs none stop, AMD makes 20%, better and cheaper to forward order hours for new products then pick up slack hours for the back orders.
Here are the steps:
1. Design a process to do something that other people haven’t developed yet.
2. Patent it.
Latest Answers