Such a nice fellow, and thanks for the interview. Very valuable to keep in the know about companies like GF even though or even because they are not chasing the leading edge anymore.
I was always putting the high hopes for GF and it finally looks like they're on a path to solidify their position, it's not all in just leading edge Looks like good CEO with just right amount of tech knowledge as in economics
I'm sure there's microcontrollers, simple CPUs/SoCs, or RF components in all kinds of consumer electronics that would benefit from a shrink to a smaller GF node. Manufactured on a newer process, these components could increase many aspects such as efficiency, performance, or potentially even manufacturing capacity as they'll get more chips per wafer on, say 28nm as opposed to 130nm. As the automotive chip crunch during the last few years showed, a lot of those companies are still reliant on manufacturing at MUCH older nodes that are very much already at capacity. Also, with all the latest geopolitical strife, it's good to have distributed manufacturing so that conflict in a region doesn't impact as much of the global economy. Happy to see GF finding their market.
A problem with going to smaller nodes is that at some point simple chips are pad limited as the pads dictate how long each edge needs to be. So you can't shrink the die further without loosing any functionality.
RF doesn't really get better with smaller nodes. RF performance pretty much plataued at 45 nm (though the SOI nature of GF22 makes it uniquely suited to millimeter-wave stuff). The only reason it shrinks after is because you might have to integrate it with a lot of digital in a SoC, and that digital wants to go to the smaller node because it does benefit a lot from the shrink. Even GF's 45 SOI is somewhat better than 22 SOI in raw performance. Anything that is simple will not go to new nodes because it is so prohibitively expensive. The mask set and tools just cost more and more and if you have small dies, you don't have the wafers/month to amortize those costs over.
The question about investment in rd in non-leading edge nodes was a good one. I wonder if ocie will help shine a spotlight on designs that aren't leading node. Building a whole CPU on 5nm starts to look wasteful.
This man is a good salesman for GF. I still wish they were in the HPC areas. I guess given enough incentive and the $$ to go along with it they would. Thanks for this interview.
So what are examples of the innovation that happens with the "essential" nodes? I can imagine the yield increases steadily over time but is there anything else?
Theres still lots to do. Especially in analog, design lacks technology behind quite a bit. For example more flexibility and higher performance metal stack for RF passives.
lots of RF stuff (70 GHz automotive radar uses 'legacy' nodes like 45nm and 22nm), precision analog, sensors, analog compute (some AI startups doing analog compute in memory in 22nm)
Well, their business is the trailing edge, but it's easy to forget that the trailing edge is a moving target. If some of their customers wanted that functionality I can't imagine they wouldn't jump on the opportunity if the economics made sense.
I wish them all the luck in developing edge inference machines. Do we have some sort of timeline and information on who they are partnering with to make that happen? What about software support?
Well for edge inference, producing the cheapest possible chip could be significant. You don't need that much compute. Training ai is the expensive part. Imagine a raspberry pi with a 5dollar inference chip, where you can run 30fps full HD object detection. Does not seem far fetched.
Ian, have you made any videos about the Fotonix technologies? I found some info on GF website but it's pretty vague. Main thing I can see is it's the most advanced and largest scale microphotonic production available.
Nice interview, unfortunate you didn't ask about whether glofo sees itself moving down into the denser nodes at a later point. Do they see the economics of those nodes becoming more manageable over time or will they ride it out evolving their existing nodes for the time being?
This channel has a video about that, cost per transistor goes up as you go to the latest node but I think the sweet spot currently is no longer at 16nm but rather at 7nm
As said in the interview, 70% of the market (# of chips) isn't on leading edge. We're talking power management, modems, photonics, controllers, embedded memory (RFID) etc - all the mundane stuff that sits around the big powerful chips. There's 500mm2+ silicon in your smartphone, and the big SoC and DRAM is only 150 of that. You need someone to build the rest, and that's where GF plays.
@@BeyondImaginationzz I think he is trying to say how long can their business model last as the leading nodes become legacy nodes. Remember GB's nodes were once leading edge.
For cutting edge technology, yes tsmc is at top , but tsmc only cater to developing CPUs and GPUs mainly , like apple CPUs , amd CPUs and GPUs , Nvidia GPUs , AI CPUs .....etc But things like that don't scale very well at lower nodes still needs to be developed on 10 nm nodes , and so many companies can't afford cutting edge 3 nm or 5 nm nodes,( tapping out cost is $500 millions ) those use 10 nm nodes ( tap out cost only $40 million ) , huge difference . GF still has place for long time . Amd still buys $1.2 billion worth wafers from them ,
It is good to see a CEO with a clear vision and future path.
Such a nice fellow, and thanks for the interview. Very valuable to keep in the know about companies like GF even though or even because they are not chasing the leading edge anymore.
I was always putting the high hopes for GF and it finally looks like they're on a path to solidify their position, it's not all in just leading edge
Looks like good CEO with just right amount of tech knowledge as in economics
Last question from audience amused me 😂, but I think CEO of GF had a great response. 👍 Ha ha~
I'm sure there's microcontrollers, simple CPUs/SoCs, or RF components in all kinds of consumer electronics that would benefit from a shrink to a smaller GF node. Manufactured on a newer process, these components could increase many aspects such as efficiency, performance, or potentially even manufacturing capacity as they'll get more chips per wafer on, say 28nm as opposed to 130nm.
As the automotive chip crunch during the last few years showed, a lot of those companies are still reliant on manufacturing at MUCH older nodes that are very much already at capacity.
Also, with all the latest geopolitical strife, it's good to have distributed manufacturing so that conflict in a region doesn't impact as much of the global economy.
Happy to see GF finding their market.
A problem with going to smaller nodes is that at some point simple chips are pad limited as the pads dictate how long each edge needs to be. So you can't shrink the die further without loosing any functionality.
@@12Lego12ts that's a good point. I guess in those cases a redesign might be necessary, which adds cost and complexity and won't always be possible.
I'm pretty sure RF doesn't shrink very well, or easily, as it is analog and high power.
RF doesn't really get better with smaller nodes. RF performance pretty much plataued at 45 nm (though the SOI nature of GF22 makes it uniquely suited to millimeter-wave stuff). The only reason it shrinks after is because you might have to integrate it with a lot of digital in a SoC, and that digital wants to go to the smaller node because it does benefit a lot from the shrink. Even GF's 45 SOI is somewhat better than 22 SOI in raw performance.
Anything that is simple will not go to new nodes because it is so prohibitively expensive. The mask set and tools just cost more and more and if you have small dies, you don't have the wafers/month to amortize those costs over.
The question about investment in rd in non-leading edge nodes was a good one. I wonder if ocie will help shine a spotlight on designs that aren't leading node. Building a whole CPU on 5nm starts to look wasteful.
Outstanding coverage, thank you!! Great to hear from Mr Caulfield
This man is a good salesman for GF. I still wish they were in the HPC areas. I guess given enough incentive and the $$ to go along with it they would. Thanks for this interview.
So what are examples of the innovation that happens with the "essential" nodes? I can imagine the yield increases steadily over time but is there anything else?
Different features on different nodes, such as embedded memory, body bias, RF, IP support etc.
Theres still lots to do. Especially in analog, design lacks technology behind quite a bit. For example more flexibility and higher performance metal stack for RF passives.
lots of RF stuff (70 GHz automotive radar uses 'legacy' nodes like 45nm and 22nm), precision analog, sensors, analog compute (some AI startups doing analog compute in memory in 22nm)
What about Infineon or Intel arent they foundries too?
And isn't GF getting absolutely smacked by Infineon?
My thought as well
That is the one major thing I noticed. the rest was solid lol
If the cost of sub-10nm foundary went down like 50% and much mature, will GF upgrade then?
Well, their business is the trailing edge, but it's easy to forget that the trailing edge is a moving target. If some of their customers wanted that functionality I can't imagine they wouldn't jump on the opportunity if the economics made sense.
I wish them all the luck in developing edge inference machines.
Do we have some sort of timeline and information on who they are partnering with to make that happen? What about software support?
I think he meant they'll be doing the analog signal devices and the communication infrastructure devices.
Well for edge inference, producing the cheapest possible chip could be significant. You don't need that much compute. Training ai is the expensive part. Imagine a raspberry pi with a 5dollar inference chip, where you can run 30fps full HD object detection. Does not seem far fetched.
Could those essential nodes profit from stacked cmos to increase density? Maybe even befor singel digit nodes start to do this?
Ian, have you made any videos about the Fotonix technologies? I found some info on GF website but it's pretty vague.
Main thing I can see is it's the most advanced and largest scale microphotonic production available.
Nice interview, unfortunate you didn't ask about whether glofo sees itself moving down into the denser nodes at a later point. Do they see the economics of those nodes becoming more manageable over time or will they ride it out evolving their existing nodes for the time being?
That's literally addressed in the video.
Microchip Technology have some real essential foundry capacity even further down the curve iirc.
Any link to the original presentation and the stats etc?
Hmm. Is it cost per transistor really increasing at 16nm and below, or cost per mm2?
This channel has a video about that, cost per transistor goes up as you go to the latest node but I think the sweet spot currently is no longer at 16nm but rather at 7nm
what about Intel IFS what is Intel's approach win foundry customers?
Not the topic of this interview
@@TechTechPotato did you ask him about Foveros like chip packaging technologies?
Was the room very dark? The image looks a bit grainy.
ISO was set wrong on the camera. Had to turn up the brightness to get something useful
@@TechTechPotato Thanks for replying!
where is the 7nm discussion?
It's always astonishing to me how only one smart man can turn around and save such giant companies.
Great man of history theory shows up in many places.
Not sure he inspires confidence, he seems like Intel, missing buses and giving excuses
the world would be a less worse place without AI
Don’t understand the place of GF going forward. Can just be crushed by Samsung or TSMC at any moment if they wish
As said in the interview, 70% of the market (# of chips) isn't on leading edge. We're talking power management, modems, photonics, controllers, embedded memory (RFID) etc - all the mundane stuff that sits around the big powerful chips. There's 500mm2+ silicon in your smartphone, and the big SoC and DRAM is only 150 of that. You need someone to build the rest, and that's where GF plays.
you are underestimating demand of non-leading chips specially in developing countries, high end chips are too expensive in those countries
@@BeyondImaginationzz I think he is trying to say how long can their business model last as the leading nodes become legacy nodes. Remember GB's nodes were once leading edge.
For cutting edge technology, yes tsmc is at top , but tsmc only cater to developing CPUs and GPUs mainly , like apple CPUs , amd CPUs and GPUs , Nvidia GPUs , AI CPUs .....etc
But things like that don't scale very well at lower nodes still needs to be developed on 10 nm nodes , and so many companies can't afford cutting edge 3 nm or 5 nm nodes,( tapping out cost is $500 millions ) those use 10 nm nodes ( tap out cost only $40 million ) , huge difference . GF still has place for long time .
Amd still buys $1.2 billion worth wafers from them ,
So if those other fabs could get their transistors/chips per dollar to GF level and scale that business, then we would get scary for GF.