News Global Foundries Expansion (Anandtech)

[Ajay]

Well, business is looking good for GloFo: https://www.anandtech.com/show/1683...illions-doubling-fab-8-creating-new-fab-in-ny

They are in the midst of a huge expansion. Kind of surprised myself. I wonder how long it will be till they move to a 7N node. Can't milk 12nm forever.

 

[Ajay]

Ugh, made this a 'question' instead of 'news' 🙄

I changed it for you if thats OK, back to news. If you want it back to question, let me know.
Markfw
Anandtech Moderator

 

[Hitman928]

Well, business is looking good for GloFo: https://www.anandtech.com/show/1683...illions-doubling-fab-8-creating-new-fab-in-ny

They are in the midst of a huge expansion. Kind of surprised myself. I wonder how long it will be till they move to a 7N node. Can't milk 12nm forever.

Glofo abandoned 7 nm Finfet. Their next node is 12FD-SOI which is an evolution of their 22FD-SOI. Maybe they decide to pick 7 nm Finfet back up but I doubt it.

 

[Ajay]

I imagine that there is currently allot of IoT, automotive, etc. that can be done on current and near term node. Not sure what the breakdown is for GloFo sales by node. Is Singapore 12FD? Seems like the new NY Fab would be a generation after that.

 

[LightningZ71]

Their current, most advanced, finger node is 12lp+, which incorporates some of their IP developed for their 7nm node. It's a non-trivial improvement, but isn't in the same class as N7 or i10,sf,esf. GF are seeing substantial revenue from all of their fans currently, and are essentially investing what they are making in expanding their capabilities with their current nodes.

In the future, they will likely partner with another foundary to purchase trailing node IP and equipment and update their offerings with newer model capacity, but are so far removed from leading edge development that they would likely never get back into leading edge development.

 

[semiman]

They'll not move to sub 7nm node where EUVs dominate. You can't really buy only 1 or 2 EUV scanners(~10% of their annual revenue) to set up a new process. Setting up a new process requires a new methodology, PDK, ecosystem support, and factory. There will be massive losses if one EUV scanner stops (-50% throughput).
Still, DUV-based processes are good and have a strong demand so GloFo is doing what they can.

 

[Mopetar]

I don't think Global Foundries needs to be on the bleeding edge to compete. Here's a graph from an AT article earlier this year that shows TSMC's revenue breakdowns per node:

Even with 5nm ramping up, half of their revenue is still using nodes that are effectively older than the best node that Global Foundries has. For all the fuss about the latest and greatest process nodes, a lot of people are still paying for wafers on a process that's over a decade old. Looking at the historical trend it doesn't appear that the bleeding edge node ever accounts for much more than a third of the revenue, or at least not in the time scale for this data.

There's no sense in Global Foundries trying to run a race that they can't hope to win. Imagine if every creature in the sea were desperately trying to occupy the ecological of the shark. It just wouldn't work.

 

[DrMrLordX]

I don't think Global Foundries needs to be on the bleeding edge to compete.

They don't. I expect that under the present circumstances, GF is selling every wafer they can crank out.

 

[Ajay]

They don't. I expect that under the present circumstances, GF is selling every wafer they can crank out.

They are. Based on the above graph, I guess than can get along with a 12nm leading edge node for some time.

 

[cytg111]

I don't think Global Foundries needs to be on the bleeding edge to compete. Here's a graph from an AT article earlier this year that shows TSMC's revenue breakdowns per node:

Even with 5nm ramping up, half of their revenue is still using nodes that are effectively older than the best node that Global Foundries has. For all the fuss about the latest and greatest process nodes, a lot of people are still paying for wafers on a process that's over a decade old. Looking at the historical trend it doesn't appear that the bleeding edge node ever accounts for much more than a third of the revenue, or at least not in the time scale for this data.

There's no sense in Global Foundries trying to run a race that they can't hope to win. Imagine if every creature in the sea were desperately trying to occupy the ecological of the shark. It just wouldn't work.

Damn. Someone explain the 90nm trend please? Especially cost-effective node?

 

[Doug S]

Damn. Someone explain the 90nm trend please? Especially cost-effective node?

What is it about the 90nm trend that has you confused? As more nodes are added below it, 90nm becomes a smaller percentage of TSMC's overall revenue but the 90nm+ revenue in absolute terms is likely unchanged as there will always be customers for those nodes.

 

[cytg111]

What is it about the 90nm trend that has you confused? As more nodes are added below it, 90nm becomes a smaller percentage of TSMC's overall revenue but the 90nm+ revenue in absolute terms is likely unchanged as there will always be customers for those nodes.

I dont know what to say, for its age it looks to have a disproportionate big slice of the pie.

 

[beginner99]

I dont know what to say, for its age it looks to have a disproportionate big slice of the pie.

I think it says 90nm+ which I think means all nodes 90nm or bigger. So it's actually 2 or even 10 different nodes.

 

[Doug S]

I think it says 90nm+ which I think means all nodes 90nm or bigger. So it's actually 2 or even 10 different nodes.

Easily more than 10 I'm sure, if you include the variants at nominally the same linewidth.

 

[ThatBuzzkiller]

A late entry to 7nm is definitely a real possibility if EUVL technology gets more efficient and mature. Could happen as soon as 2023 ...

 

[Ajay]

A late entry to 7nm is definitely a real possibility if EUVL technology gets more efficient and mature. Could happen as soon as 2023 ...

7nm only needs DUV.

 

[NostaSeronx]

GlobalFoundries will probably go NZ2C/NZ3C from Canon Nanotechnologies or any other nanoimprint technique+device. Since, they want to go into other markets than leading edge logic nodes.

For example: https://www.nanotech-now.com/columns/?article=902

The patent space for NIL is more liberated than EUVL.

 

[ThatBuzzkiller]

7nm only needs DUV.

Yes but multi-patterning isn't ideal since you need to use a lot of masks so EUVL looks to be a promising proposal to lower the entry cost to manufacturing 7nm process in the future if we can improve the throughput of these systems ...

In the not far future, we can expect to see the introduction of anamorphic high-NA EUVL systems in a couple of years which will help even further in the reduction of mask count ...

 

[Arkaign]

I don't think Global Foundries needs to be on the bleeding edge to compete. Here's a graph from an AT article earlier this year that shows TSMC's revenue breakdowns per node:

Even with 5nm ramping up, half of their revenue is still using nodes that are effectively older than the best node that Global Foundries has. For all the fuss about the latest and greatest process nodes, a lot of people are still paying for wafers on a process that's over a decade old. Looking at the historical trend it doesn't appear that the bleeding edge node ever accounts for much more than a third of the revenue, or at least not in the time scale for this data.

There's no sense in Global Foundries trying to run a race that they can't hope to win. Imagine if every creature in the sea were desperately trying to occupy the ecological of the shark. It just wouldn't work.

Thank you for this.

It's super easy for enthusiasts to focus only on the latest processes and x86/arm products, but the semiconductor world is SO much bigger than that, and the vast majority of devices needing silicon don't need the latest and greatest by any stretch. Capacity is a real issue globally at present, so expansion eases some of the supply pressure.