What surprises me with Infineon is that they seem to not work on a LLM on a chip. In particular that here in Linz Infineon, Apple's 5G modem designer and the Johannes Kepler University with Sepp Hochreiter (Inventor of LSTM) are just next to each other.
I watched a interview of Infineon head of AI just recently, and I was like excited to hear about some cool hardware. And all he did is talk about how he uses LLMs.
Yeah sadly. I think they have all the know how and people and connections to pull that off. Yet as so often, Austrian (Linz) companies are all about being a supplier not a disrupter.
> Yet as so often, Austrian (Linz) companies are all about being a supplier not a disrupter.
That makes a lot of sense when you consider the origins. After all, Infineon used to be a subsidiary of Siemens, a famously relatively conservative company
Infineon marks innovation milestone with world’s first 300-millimeter power silicon carbide (SiC) wafer
PBusiness & Financial Press
Nov 09, 2026
Infineon presents industry-first functional power devices processed on 300-millimeter silicon carbide wafers
Technological breakthrough underlines Infineon’s innovation strength and leadership in power semiconductors and wide-bandgap technologies
300-millimeter SiC process feasibility lays the foundation for secure, future-proof SiC supply to customers in automotive, renewable energy, industrial and AI data center applications
Infineon will present the first processed 300-millimeter SiC wafer to the public at the electronica trade show starting 10 November 2026 in Munich
Anyone know why it took so long for them to transition to 300mm? As mentioned in the article, they're mainly using existing 300mm equipment designed for silicon, which transitioned years ago. I imagine some process steps are different (mainly growing to larger boule, but probably others), but it doesn't seem like they would be huge blockers.
Making a SiC crystal that's 300mm large is much more difficult than silicon. SiC can't use the Cz process that silicon uses for one. SiC doesn't even melt at normal temperatures. This is great for thermals but bad if you want to use molten phase techniques.
did anyone say anything about larger die? I think this is a more-die play, so a given defect density won't affect yield. (maybe a small improvement due to less circumference for given area.)
Not really. The article is about a larger wafer, not larger dies.
They are mostly cutting transistors, voltage regulators, and such out of those. There's already no practical limitation on the size of those components.
European investors and governments are risk averse. DRAM is one of the riskiest silicon products. They talk a lot about independence but European governments lack the raw investment power and willpower to make decade-long investments that are guaranteed to lose money in the next decade as DRAM. If Europe wants DRAM made in European soil, the governments and people should be ready for losing tens to hundreds of billions of taxpayer money.
When Korean and American factories ramp up, the price will plummet. Basically whatever business will not make a single Euro profit for years.
European silicon manufacturers target industrial and automotive for a reason. They want large, guaranteed contracts for older silicon nodes which are cheap and profitable. They let the US, China and Taiwan to take risk of developing new ones. When they do research, they prefer incremental ones.
When it comes to real talk Europe doesn't really care about independence more than it cares about capital and profits. Only an actual war will make the governments care and it will be too late. Without a strong drive to create one's own nuclear arsenal, don't expect any groundbreaking risky investment out Europe.
I doubt it will. You hv to consider memory tech may shift a bit towards LLM-first, so a kind of more computations in memory, and other innovations will be common and demand will not be lower especially if you consider average lifetime of a GPU card is 5 years, there will be demand like just to utilize all those racks to pay out all the money they borrow now. In memory compute will make AI more decentralized (still a hopium).
I think you missed the part where they(SK, Samsung and Micron) are a cartel forming a monopoly, and can control market prices regardless of demand via controlling the supply.
Sure, cartel behavior is illegal, but if you're in bed with the right political administrations, then eyes get closed.
Wolfspeed/Cree does not have a fab and does not create wafers, let alone LEDs. They are purely a research/licensing company with a side business of selling luminaire fixtures.
Different kind of applications. GaN has no reverse recovery at all and a way smaller gate source capacity and threshold voltage so you can crank up the frequency higher.
But GaN conducts heat worse and also anything that can withstand more than 650V becomes a huge die due to no vertical mosfet tech in GaN yet.
SiC is more or less like super Silicon, almost negligible reverse recovery, very fast switching, way smaller gate charge and a 1200V 1700V is economical and very performant.
For high power like 10s-100s of kW SiC is the best. Higher than that passive components become so big you don't need as high a frequency so you would go with Si IGBTs, lower than that GaN allows you to increase frequency by a lot and reduce passive components size AND in a lot of cases total losses.
So yeah, don't expect to see GaN in a car power train and SiC in a server power supply.
GaN is (much) better than SiC for certain FOMs, but worse in terms of thermal conduction and technology maturity. In addition, you can't yet make "vertical" GaN devices (there's a few flamed out startups that tried) which forces certain undesirable tradeoffs at high blocking voltages. So GaN's sweet spot is 100-650V devices, for 48V and mains voltage applications.
There is an overlap between SiC and GaN at rated voltages between 600 V and 750 V, but at lower voltages than that only GaN devices are used and at higher voltages than that only SiC devices are used.
Even at 650 V where there are available both GaN and SiC devices, there is little direct competition between them. For such voltages, SiC devices are used for controlling greater powers, where also the power dissipated by the device is great enough to require a big package that enables efficient cooling, while the pure GaN devices (i.e. not the hybrid silicon-GaN cascode transistors) are available only in small surface mounted packages, which enable much higher switching frequencies, but at lower output powers, because it is more difficult to cool them well enough.
For now, the SiC devices are only unipolar devices, i.e. MOSFETs and Schottky diodes. But in the future, when SiC crystals with fewer defects will become available, there will also be SiC bipolar devices, i.e. thyristors and IGBTs. The bipolar devices will be available for much greater powers than the current devices, while also being much faster than the silicon bipolar devices. On GaN, bipolar devices are not possible, so GaN will always be used only for higher switching frequencies, but lower output powers than SiC.
Silicon carbide MOSFETs are the best devices for the SOTA electric vehicles and also for their chargers, which work at 800 V (where 1200 V SiC MOSFETs are used) or at 1000 V (where 1400 V SiC MOSFETs are used).
>You can likely live without a new gaming rig. You likely cannot live without replaceable parts for trucks and cars.
Bad logic for an argument. With this logic you can say you can live fine with your 30 year old diesel car that doesn't need these new cutting edge SiC wafer chips, and that making diesel engines is more important for national security than making chips, and so on and so forth till you reach knowing how to make fire, swords an arrows as the most important for national security.
Truth is fabbing CPUs and GPUs is miles more profitable than what Infineon and EU are fabbing, and they're just as important for national security/sovereignty if not even more:
Did Infineon's engineers use EU or US CPUs and GPUs for the complex CAD/CAE simulations needed to develop those SiC wafers? Why else do you think the US got TSMC and Samsung to make cutting edge fabs on US soil if CPUs and GPUs are not important for national security? When you need to process TB of live battlefield data in case of war, do you not need domestic CPUs and GPUs?
Ten years old CPUs are "okay" and there are fabs in Europe that could make them. Faster is better, but slower isn't entirely useless and hugely better than no CPU. Like a 100 hp heavy truck is greatly underpowered, but almost infinitely better than no truck.
Yeah but in a (trade) war against a power who has modern chips and can process data and intelligence faster, who has better leverage? The one with cutting edge tech, or the one with "okay" tech?
>and there are fabs in Europe that could make them.
Which EU fabs make 10 year old 14nm CPUs and GPUs? Which EU domestic CPUs and GPUs designs are there at the moment that can be fabbed in case of emergency?
The truth is EU is painfully absent from this field. Sure, in theory they can duct tape together some RISC-V CPU and some ARM MALI GPU production on Dresden 16nm fabs, in case of a war, but since the EU never shipped a commercial domestic CPU and GPU from scratch in the last 20+ years , it's not as easy as turning a screw, as it is for nations that ship new CPUs every year to consumers: US, Korea, Taiwan and to an extent Japan and China.
Now SiC MOSFETs are the most efficient devices for the power converters used in electric vehicles, where DC voltages of 800 V or 1000 V are used in the latest models, which require, respectively, 1200 V or 1400 V SiC MOSFETs.
GaN HEMTs (high-electron-mobility transistors) are the most efficient devices for the power supplies that are fed from mains electricity, where 650 V GaN devices are used for the universal power supplies, which must accept up to 240 V AC on input.
I'm curious about RF uses (esp phased-array radar / AESA). usually GaN is treated as the the best for power/efficiency reasons, but does SiC enter into the picture? not fast enough?
Wooooo let's go back to 1827 for a moment and have a chat with Anyos Jedlik. Or maybe Andres Flocken in 1881. Let's see how they did electric vehicles.
Yes?! Infineon's fabs and new SiC wafers aren't built for the small feature sizes of CPUs and GPUs.
Making power FETs uses different process nodes than CPUs and GPUs. All semi fabs are highly specialized: some for power ICs, some for RF, some for DRAM, some for logic, some for NAND.
What surprises me with Infineon is that they seem to not work on a LLM on a chip. In particular that here in Linz Infineon, Apple's 5G modem designer and the Johannes Kepler University with Sepp Hochreiter (Inventor of LSTM) are just next to each other. I watched a interview of Infineon head of AI just recently, and I was like excited to hear about some cool hardware. And all he did is talk about how he uses LLMs.
AFAIK, they do power and analog stuff and digital logic for automotive and industry, not cutting edge digital logic at the smallest feature sizes.
Yeah sadly. I think they have all the know how and people and connections to pull that off. Yet as so often, Austrian (Linz) companies are all about being a supplier not a disrupter.
> Yet as so often, Austrian (Linz) companies are all about being a supplier not a disrupter.
That makes a lot of sense when you consider the origins. After all, Infineon used to be a subsidiary of Siemens, a famously relatively conservative company
The summary at the top of that page is:
Infineon marks innovation milestone with world’s first 300-millimeter power silicon carbide (SiC) wafer
PBusiness & Financial Press Nov 09, 2026
Infineon presents industry-first functional power devices processed on 300-millimeter silicon carbide wafers
Technological breakthrough underlines Infineon’s innovation strength and leadership in power semiconductors and wide-bandgap technologies
300-millimeter SiC process feasibility lays the foundation for secure, future-proof SiC supply to customers in automotive, renewable energy, industrial and AI data center applications
Infineon will present the first processed 300-millimeter SiC wafer to the public at the electronica trade show starting 10 November 2026 in Munich
Anyone know why it took so long for them to transition to 300mm? As mentioned in the article, they're mainly using existing 300mm equipment designed for silicon, which transitioned years ago. I imagine some process steps are different (mainly growing to larger boule, but probably others), but it doesn't seem like they would be huge blockers.
Making a SiC crystal that's 300mm large is much more difficult than silicon. SiC can't use the Cz process that silicon uses for one. SiC doesn't even melt at normal temperatures. This is great for thermals but bad if you want to use molten phase techniques.
SiC is just in general an extremely unforgiving material to work with. The stars just didn't align for 300mm wafers until now.
From the wiki, SiC monocrystals are made in an "Acheson graphite electric resistance furnace."
Scaling this up does not appear to be trivial.
CVD and the "Lely" process are also mentioned.
https://en.wikipedia.org/wiki/Silicon_carbide#Production
Larger dies == more material == more chance for impurities to wreck the crystal.
Chasing 9’s is very hard already (both uptime and purity), and by the time you start getting into 6+ digits of 9’s? Really really hard.
did anyone say anything about larger die? I think this is a more-die play, so a given defect density won't affect yield. (maybe a small improvement due to less circumference for given area.)
It says right in the title?
Not really. The article is about a larger wafer, not larger dies.
They are mostly cutting transistors, voltage regulators, and such out of those. There's already no practical limitation on the size of those components.
Can you guys restart DRAM production in Europe again? If 20 or so years ago it was possible, today is the best economical opportunity.
European investors and governments are risk averse. DRAM is one of the riskiest silicon products. They talk a lot about independence but European governments lack the raw investment power and willpower to make decade-long investments that are guaranteed to lose money in the next decade as DRAM. If Europe wants DRAM made in European soil, the governments and people should be ready for losing tens to hundreds of billions of taxpayer money.
When Korean and American factories ramp up, the price will plummet. Basically whatever business will not make a single Euro profit for years.
European silicon manufacturers target industrial and automotive for a reason. They want large, guaranteed contracts for older silicon nodes which are cheap and profitable. They let the US, China and Taiwan to take risk of developing new ones. When they do research, they prefer incremental ones.
When it comes to real talk Europe doesn't really care about independence more than it cares about capital and profits. Only an actual war will make the governments care and it will be too late. Without a strong drive to create one's own nuclear arsenal, don't expect any groundbreaking risky investment out Europe.
In a few years dram will be a losing money segment again, why bother?
I doubt it will. You hv to consider memory tech may shift a bit towards LLM-first, so a kind of more computations in memory, and other innovations will be common and demand will not be lower especially if you consider average lifetime of a GPU card is 5 years, there will be demand like just to utilize all those racks to pay out all the money they borrow now. In memory compute will make AI more decentralized (still a hopium).
I think you missed the part where they(SK, Samsung and Micron) are a cartel forming a monopoly, and can control market prices regardless of demand via controlling the supply.
Sure, cartel behavior is illegal, but if you're in bed with the right political administrations, then eyes get closed.
Memory packaging is non-trivial and Europe currently doesn't have players in that space.
The primary players are South Korea, Taiwan, the US, and China and most of their capex in the space is in Asia and the US.
Goodram exists. Last sticks I bought were from them because they are not that expensive now.
as far as I recall it, packaging wasn't done in Europe even then
Global Foundries (fab) is in Germany and Amkor (packaging) is in Portugal
I almost forgot abuot GlobalFoundries and I wonder if they could profitably make DRAM at current prices...
I thought Wolfspeed was already at 300mm at the start of this year?
I don't think they're claiming to be first in producing the substrate, just first to process it into functional power devices.
Wolfspeed/Cree does not have a fab and does not create wafers, let alone LEDs. They are purely a research/licensing company with a side business of selling luminaire fixtures.
Huh? Wolfspeed grows SiC crystals, makes substrates, fabs SiC wafers, and packages these.
How does this compare to GaN? Better or worse?
Different kind of applications. GaN has no reverse recovery at all and a way smaller gate source capacity and threshold voltage so you can crank up the frequency higher.
But GaN conducts heat worse and also anything that can withstand more than 650V becomes a huge die due to no vertical mosfet tech in GaN yet.
SiC is more or less like super Silicon, almost negligible reverse recovery, very fast switching, way smaller gate charge and a 1200V 1700V is economical and very performant.
For high power like 10s-100s of kW SiC is the best. Higher than that passive components become so big you don't need as high a frequency so you would go with Si IGBTs, lower than that GaN allows you to increase frequency by a lot and reduce passive components size AND in a lot of cases total losses.
So yeah, don't expect to see GaN in a car power train and SiC in a server power supply.
GaN is (much) better than SiC for certain FOMs, but worse in terms of thermal conduction and technology maturity. In addition, you can't yet make "vertical" GaN devices (there's a few flamed out startups that tried) which forces certain undesirable tradeoffs at high blocking voltages. So GaN's sweet spot is 100-650V devices, for 48V and mains voltage applications.
SiC is for high voltage, GaN is for high switching frequency.
There is an overlap between SiC and GaN at rated voltages between 600 V and 750 V, but at lower voltages than that only GaN devices are used and at higher voltages than that only SiC devices are used.
Even at 650 V where there are available both GaN and SiC devices, there is little direct competition between them. For such voltages, SiC devices are used for controlling greater powers, where also the power dissipated by the device is great enough to require a big package that enables efficient cooling, while the pure GaN devices (i.e. not the hybrid silicon-GaN cascode transistors) are available only in small surface mounted packages, which enable much higher switching frequencies, but at lower output powers, because it is more difficult to cool them well enough.
For now, the SiC devices are only unipolar devices, i.e. MOSFETs and Schottky diodes. But in the future, when SiC crystals with fewer defects will become available, there will also be SiC bipolar devices, i.e. thyristors and IGBTs. The bipolar devices will be available for much greater powers than the current devices, while also being much faster than the silicon bipolar devices. On GaN, bipolar devices are not possible, so GaN will always be used only for higher switching frequencies, but lower output powers than SiC.
Silicon carbide MOSFETs are the best devices for the SOTA electric vehicles and also for their chargers, which work at 800 V (where 1200 V SiC MOSFETs are used) or at 1000 V (where 1400 V SiC MOSFETs are used).
This may not appear sexy but this is a huge deal. Power semiconductors are critical for most NatSec and industrial usecases.
You can likely live without a new gaming rig. You likely cannot live without replaceable parts for trucks and cars.
>You can likely live without a new gaming rig. You likely cannot live without replaceable parts for trucks and cars.
Bad logic for an argument. With this logic you can say you can live fine with your 30 year old diesel car that doesn't need these new cutting edge SiC wafer chips, and that making diesel engines is more important for national security than making chips, and so on and so forth till you reach knowing how to make fire, swords an arrows as the most important for national security.
Truth is fabbing CPUs and GPUs is miles more profitable than what Infineon and EU are fabbing, and they're just as important for national security/sovereignty if not even more:
Did Infineon's engineers use EU or US CPUs and GPUs for the complex CAD/CAE simulations needed to develop those SiC wafers? Why else do you think the US got TSMC and Samsung to make cutting edge fabs on US soil if CPUs and GPUs are not important for national security? When you need to process TB of live battlefield data in case of war, do you not need domestic CPUs and GPUs?
Ten years old CPUs are "okay" and there are fabs in Europe that could make them. Faster is better, but slower isn't entirely useless and hugely better than no CPU. Like a 100 hp heavy truck is greatly underpowered, but almost infinitely better than no truck.
>Ten years old CPUs are "okay"
Yeah but in a (trade) war against a power who has modern chips and can process data and intelligence faster, who has better leverage? The one with cutting edge tech, or the one with "okay" tech?
>and there are fabs in Europe that could make them.
Which EU fabs make 10 year old 14nm CPUs and GPUs? Which EU domestic CPUs and GPUs designs are there at the moment that can be fabbed in case of emergency?
The truth is EU is painfully absent from this field. Sure, in theory they can duct tape together some RISC-V CPU and some ARM MALI GPU production on Dresden 16nm fabs, in case of a war, but since the EU never shipped a commercial domestic CPU and GPU from scratch in the last 20+ years , it's not as easy as turning a screw, as it is for nations that ship new CPUs every year to consumers: US, Korea, Taiwan and to an extent Japan and China.
Trucks and cars worked fine without them for decades
Because they did not use electric motors.
Now SiC MOSFETs are the most efficient devices for the power converters used in electric vehicles, where DC voltages of 800 V or 1000 V are used in the latest models, which require, respectively, 1200 V or 1400 V SiC MOSFETs.
GaN HEMTs (high-electron-mobility transistors) are the most efficient devices for the power supplies that are fed from mains electricity, where 650 V GaN devices are used for the universal power supplies, which must accept up to 240 V AC on input.
I'm curious about RF uses (esp phased-array radar / AESA). usually GaN is treated as the the best for power/efficiency reasons, but does SiC enter into the picture? not fast enough?
...Can't we just go back to thermionic vacuum tubes at this point? Those worked just fine at kilowatt powers and megahertz frequencies.
"Because they did not use electric motors."
Wooooo let's go back to 1827 for a moment and have a chat with Anyos Jedlik. Or maybe Andres Flocken in 1881. Let's see how they did electric vehicles.
Oh, lookie there, no SiC.
A yes, and because the electric vehicles in 1881 were so convincing, we are all driving EVs today....wait
That's a funny way to frame things. Do you think Infineon is forced to choose between those two markets?
Yes?! Infineon's fabs and new SiC wafers aren't built for the small feature sizes of CPUs and GPUs.
Making power FETs uses different process nodes than CPUs and GPUs. All semi fabs are highly specialized: some for power ICs, some for RF, some for DRAM, some for logic, some for NAND.