Sony and TSMC Said to be Planning US$7 Billion Chip Fab in Japan

[TheLostSwede]

There doesn't seem to be a single month where rumours about new TSMC plants around the world are popping up and this time around it looks like there might be a joint venture with Sony in Japan. According to the Nikkei, the Japanese government is likely to be involved and might foot as much as half of the US$7 billion bill.

Another much more unknown player, Japanese auto parts maker Denso is also said to be a potential participant in the new fab. Denso is said to supply Toyota among others and with a shift towards more EVs, this might not be such a strange move. The new fab is expected to be located in Kumamoto Prefecture on land owned by Sony. It should be noted that Sony already manufactures image sensors here and the factory was hit badly by a large earthquake back in 2016, which led to a global shortage of certain image sensors.



If the joint venture goes ahead, this would be TSMC's first chip plant in Japan and also its first joint venture with Sony at this scale. The fab would, at least at the early stages, be focusing on image sensors, image processors and components for EVs, although as it'll take a couple of years at least to build the fab, this isn't set in stone. The Japanese government is said to be looking at getting a commitment from TSMC that chips made at the new fab will be prioritised for the Japanese market.

View at TechPowerUp Main Site

 

[mtcn77]

Omg, this is big. Japan used to be big in the foundry business, but TSMC developed pureplay foundry first and threatened to sue if they copied their foundry model(since they looked down upon their business in the beginning). Good thing they are cooperating because Japan and TSMC are two sides of the same coin, private vs. licensed enterprise.

 

[mechtech]

A question for those in the silicon industry.

so doesn’t ASML make the machines that etch/process the silicon wafers and fab buy the machines from them?

so why can’t any company buy a machine and start fabbing? Are the machines sold incomplete or is the final set up done by the end fab such as TSMC and that’s what is the secret sauce to getting a small nodes and good yields? Or is it more than that??

 

[mtcn77]

Or is it more than that??

You know, the whole Intel method is copy exactly. When one factory builds something the challenge is replicating the process in others. Like Intel, other foundry businesses like GF build entire factories around specific lithography devices(it is faster that way). As such, competition demands they follow each other's methods since no one can drop behind the leading edge without losing market share. Like socrates said, lack of education is more expensive in the process.

PS: time costs money.

 

[TheLostSwede]

A question for those in the silicon industry.

so doesn’t ASML make the machines that etch/process the silicon wafers and fab buy the machines from them?

so why can’t any company buy a machine and start fabbing? Are the machines sold incomplete or is the final set up done by the end fab such as TSMC and that’s what is the secret sauce to getting a small nodes and good yields? Or is it more than that??

Sure, how many billions do you have in the bank?

Money aside, a lot of it comes down to expertise as well. Just look at how both Intel and GloFo struggled at one point. Intel obviously had more money to throw at the problem than GloFo, who gave up and stopped working on node shrinks, at least for the time being. On top of that, the machines from ASML are only part of the process.

Just as in any other industry, starting from scratch, even if you can use the same equipment and infrastructure, doesn't mean you'll make equal quality products. It's also a huge difference between screwing up in a sawmill and screwing up in a chip fab, as a dumb example, but I hope you get my point.

 

[95Viper]

Keep it on the Topic.
This is not a political who/what versus who/what thread.
This is a tech forum not your political soapbox.

 

[Wirko]

A question for those in the silicon industry.

so doesn’t ASML make the machines that etch/process the silicon wafers and fab buy the machines from them?

so why can’t any company buy a machine and start fabbing? Are the machines sold incomplete or is the final set up done by the end fab such as TSMC and that’s what is the secret sauce to getting a small nodes and good yields? Or is it more than that??

At the bleeding edge, each ASML machine in a fab is basically a prototype, requiring a bunch of engineers with electron microscopes and whatnot, on constant alert as they struggle to keep yields good, and improve them over time. Each 0.1% gained or lost is worth many millions.

The machines are not some kind of mass-produced stuff, either - you order one a couple years in advance, it's made to your specification, you help developing it, and it contains parts and know-how provided by you.

And all they do is projecting patterns on your wafers in UV light. Etching comes later. ASML scanners are just the most notorious because they are so complex and costly. There are other machines that also work close to the limits of physics, like those for vapor deposition, and I'm sure they're also complex and costly and hard to manage too.

Sony and Denso aren't going for all that as no one but TSMC/Samsung/Intel itself can afford that. Their goal is to make "image sensors, image processors and components for EVs", and something like 28 nm is going to be good enough for that.

 

[The red spirit]

Sure, how many billions do you have in the bank?

Is that even a problem? We have stocks, bonds, P2P loans and bunch of other financing instruments. It's not even particularly new to start something like that without having cash on hand.

 

[Wirko]

TSMC developed pureplay foundry first and threatened to sue if they copied their foundry model(since they looked down upon their business in the beginning).

I don't understand that. What parts of the foundry model that you mentioned can be legally protected (by means of patents and possibly copyright)?

 

[Marshal_90]

Oh this made me so happy! Good move!

 

[Rxzlion]

A question for those in the silicon industry.

so doesn’t ASML make the machines that etch/process the silicon wafers and fab buy the machines from them?

so why can’t any company buy a machine and start fabbing? Are the machines sold incomplete or is the final set up done by the end fab such as TSMC and that’s what is the secret sauce to getting a small nodes and good yields? Or is it more than that??

1. ASML makes the lithography machines. they "burn" the blueprint that you need to etch (that is a really simplified example) on the wafer abd it'sa multi step process.
2. etching machines are another beast and they are made by multiple diffrent companies you also have dry eych and wet etch.
3. on top of those machines you have testing machine of many types.
4. a normal fab can have up to 40 types of machines with the range of 30 to 80 of each one.
5. ASML can manufacture up to 50 a year at current.
6. Each one of those can run up to 170 new model (old model was 134 if i remember ) wafers per hour and like i said it a multi step process and iit can be anywhere form 100 steps up to 600.
7. Fabs like intel, TSMC, GF and others run from 5000 to 20000 wafers in the line.
8. From all of the above you can understand why you need multiple of those machines.
9. There is a wait list and even if you order now you'll probaly get the machine in 2025.
10. I'm not going to go into what it even takes to build a fab, develop your process, train personal to keep the line runing and do all of the maintenance on the different machine and not to forget the cost of chemical, gass, electricity, water, wages and etc.
11. Even a small fab can need an investment of up to 2 billion if you want the use EUV.
12. All of what i wrote is a super simplification.

 

[zlobby]

12. All of what i wrote is a super simplification.

Yes, and that doesn't even touch the topic of raw materials, some of which are very rare and very expensive.

One can't simply get a few billions out of their pocket (ironically, many people can) and build themselves a fab.

All this is why there are not many fabs that run EUV.

Just imagine if 5nm chips were so common that even your stove would use them. We'd probably be heavily augmented cyborgs waging space wars. Or most likely we'd be all dead. If anything, human kind has shown that they can't be trusted with high tech en masse. Inb4 mobile phones are more of a point in case than on the contrary.

 

[mtcn77]

I don't understand that. What parts of the foundry model that you mentioned can be legally protected (by means of patents and possibly copyright)?

Maybe it was a joke I failed to register. The CEO was pretty self righteous while making it though. You have to hand it to him, the first years were unprofitable, although ripe with opportunity - no one dared make an investment where there were no profit. That gives him all the leeway to file all patents he can muster, don't you think? And we are dealing with a man who started his own company when they said it straight they wouldn't make him the CEO of Texas Instruments...

so why can’t any company buy a machine and start fabbing?

I will point you in the right direction with a quote, a writer of our neighbours the greeks wrote, "You cannot be king by trading in coal"...
This was before globalisation which had people forget this. Your money is as good as its origin. You have to be entrenched in the economy you are dealing with in order for it to have currency value.

Once you begin thinking this way, you figure there are no middle men in high tech.

 

[midnight100]

this is great news. i'm glad tsmc is expanding into more nations.

 

[mechtech]

At the bleeding edge, each ASML machine in a fab is basically a prototype, requiring a bunch of engineers with electron microscopes and whatnot, on constant alert as they struggle to keep yields good, and improve them over time. Each 0.1% gained or lost is worth many millions.

The machines are not some kind of mass-produced stuff, either - you order one a couple years in advance, it's made to your specification, you help developing it, and it contains parts and know-how provided by you.

And all they do is projecting patterns on your wafers in UV light. Etching comes later. ASML scanners are just the most notorious because they are so complex and costly. There are other machines that also work close to the limits of physics, like those for vapor deposition, and I'm sure they're also complex and costly and hard to manage too.

Sony and Denso aren't going for all that as no one but TSMC/Samsung/Intel itself can afford that. Their goal is to make "image sensors, image processors and components for EVs", and something like 28 nm is going to be good enough for that.

Ahh ok.

So the machines themselves, in compare again to another analogy, is like buying a shortblock engine, you get the basic machine, then your staff/plant does the tweaking (heads, camshaft, carb/fuel injection) to get the results/node/output it desires (over simplification) but the gist of it?

So say an old 45nm node/machine/equipment, say if someone like TSMC itself is selling that off, and a smaller company like UMC buys it, then the tooling should be negligible?

 

[TheoneandonlyMrK]

Am I alone in thinking 7Bln seams like a small percentage of a plant?!.

 

[mechtech]

1. ASML makes the lithography machines. they "burn" the blueprint that you need to etch (that is a really simplified example) on the wafer abd it'sa multi step process.
2. etching machines are another beast and they are made by multiple diffrent companies you also have dry eych and wet etch.
3. on top of those machines you have testing machine of many types.
4. a normal fab can have up to 40 types of machines with the range of 30 to 80 of each one.
5. ASML can manufacture up to 50 a year at current.
6. Each one of those can run up to 170 new model (old model was 134 if i remember ) wafers per hour and like i said it a multi step process and iit can be anywhere form 100 steps up to 600.
7. Fabs like intel, TSMC, GF and others run from 5000 to 20000 wafers in the line.
8. From all of the above you can understand why you need multiple of those machines.
9. There is a wait list and even if you order now you'll probaly get the machine in 2025.
10. I'm not going to go into what it even takes to build a fab, develop your process, train personal to keep the line runing and do all of the maintenance on the different machine and not to forget the cost of chemical, gass, electricity, water, wages and etc.
11. Even a small fab can need an investment of up to 2 billion if you want the use EUV.
12. All of what i wrote is a super simplification.

Got ya. My background is heavy industry, mining, metallurgy, hydro. I can't even guess on the price it would cost to build a new metallurgical plant here in North America today, but if I were I would guess anywhere from $1-$5 billion depending on size/output and amount and type of metals to refine.

So like everything in heavy industry its a big complex beast of a system, but also like heavy industry, pretty much all the items to build the plants are off the shelf so to speak, it's getting the expertise and capital to make it happen.

Sounds like cool place to get a tour.

 

[TheLostSwede]

So say an old 45nm node/machine/equipment, say if someone like TSMC itself is selling that off, and a smaller company like UMC buys it, then the tooling should be negligible?

Actually, some of the machines might not fit in at all, since as you know, even "identical" nodes don't use the same underlying technology.
This is why you can't easily transfer a design made for TSMC's nodes to say Samsung on a similar node. It's in fact often easier to start a new design for a specific fab node than to try and make an existing design work with a different foundry node.
The only time this sort of worked, was during the various fab node alliances, but even those required a lot of extra work if you want to move between foundries, as not everything was the same.

Am I alone in thinking 7Bln seams like a small percentage of a plant?!.

How so? This isn't going to be a cutting edge fab, at least not based on what they're planning to manufacture there, if the Nikkei news post is anything to go by.

Sounds like cool place to get a tour.

Actually, you can't see much at all, since almost everything takes place inside the machines, due to the fact it has to be dust free.
Been to an Intel fab in Ireland and an NEC memory fab in Scotland and it was really quite boring. Never actually been invited to TSMC...

 

[TheoneandonlyMrK]

Actually, some of the machines might not fit in at all, since as you know, even "identical" nodes don't use the same underlying technology.
This is why you can't easily transfer a design made for TSMC's nodes to say Samsung on a similar node. It's in fact often easier to start a new design for a specific fab node than to try and make an existing design work with a different foundry node.
The only time this sort of worked, was during the various fab node alliances, but even those required a lot of extra work if you want to move between foundries, as not everything was the same.


How so? This isn't going to be a cutting edge fab, at least not based on what they're planning to manufacture there, if the Nikkei news post is anything to go by.


Actually, you can't see much at all, since almost everything takes place inside the machines, due to the fact it has to be dust free.
Been to an Intel fab in Ireland and an NEC memory fab in Scotland and it was really quite boring. Never actually been invited to TSMC...

Just machine costs and fab coats , ok smaller scale, at least initially.

 

[TheLostSwede]

Just machine costs and fab coats , ok smaller scale, at least initially.

Yeah, it's not going to cover any of the operating costs of course.
According to Wikipedia, a "normal" fab costs US$2-3 billion, but TSMC paid US$9.3 billion for at least one of its fabs and it's likely to cost even more in the future.

Semiconductor fabrication plant - Wikipedia

en.wikipedia.org

 

[R-T-B]

so why can’t any company buy a machine and start fabbing? Are the machines sold incomplete or is the final set up done by the end fab such as TSMC and that’s what is the secret sauce to getting a small nodes and good yields? Or is it more than that??

You need a lot more than just the machines to make a modern process. You need all sorts of films, masks, etc, many of which are considered proprietary patented tech.

 

[mechtech]

You need a lot more than just the machines to make a modern process. You need all sorts of films, masks, etc, many of which are considered proprietary patented tech.

gah

patents............

 

[mtcn77]

gah

patents............

You know patents are exclusivity rights? Anybody producing without them would nullify them. They don't actually do anything, they just let you claim it.

 

[mechtech]

You know patents are exclusivity rights? Anybody producing without them would nullify them. They don't actually do anything, they just let you claim it.

yes I am aware.