Infineon to Start 300 mm GaN Wafer Production as TSMC Exits Market

[AleksandarK]

Infineon has announced that its gallium nitride (GaN) power semiconductor production is on schedule, confirming that its 300 mm wafer fab will deliver customer samples in Q4 2025. The German semiconductor manufacturer becomes the first company to successfully integrate 300 mm GaN wafer technology into its existing high-volume production infrastructure, achieving a 2.3x higher chip yield per wafer compared to conventional 200 mm processes. Infineon plans to capitalize on the projected annual market growth of 36%, with GaN power applications expected to reach $2.5 billion by 2030, according to an analysis by the Yole Group. The timing proves particularly strategic, as TSMC has announced plans to shut down its GaN production lines and dismantle facilities within two years, creating a significant market vacuum.

The new announcement addresses a fundamental industry challenge: scaling GaN production while maintaining cost competitiveness with silicon alternatives. Infineon's integrated device manufacturer (IDM) model provides complete process control from wafer fabrication through final product delivery, enabling what the company projects will be cost parity between comparable silicon and GaN devices. Johannes Schoiswohl, Head of GaN Business Line, emphasized that the scaled 300 mm manufacturing leverages Infineon's existing infrastructure investments while supporting rapid capacity expansion for emerging applications, including AI system power supplies, automotive charging systems, and industrial motor control. TSMC's strategic retreat from GaN manufacturing indicates that the company's focus remains on high-margin logic processors, leaving specialized power semiconductor companies like Infineon to dominate the expanding GaN market. In this market, GaN's superior power density, switching speeds, and thermal performance deliver measurable system-level advantages over traditional silicon-based solutions.



View at TechPowerUp Main Site | Source

 

[Panther_Seraphin]

Some legitimate questons from me

We know GaN seems to be smaller/more efficent than equivalent Silicon designs in Phone Chargers etc and seems to be favoured in EVs as well but are they applicable to say GPU/Motherboard VRM designs or is there caveats that prevents them?

If they are suitable what has prevented their usage so far?

Are they suitable for PSU designs as I know they are a lot higher voltages 12volts vs 1.2 for example and same questions as above.

 

[sudothelinuxwizard]

If they are suitable what has prevented their usage so far?

Cost.

 

[Rightness_1]

Some legitimate questons from me

We know GaN seems to be smaller/more efficent than equivalent Silicon designs in Phone Chargers etc and seems to be favoured in EVs as well but are they applicable to say GPU/Motherboard VRM designs or is there caveats that prevents them?

If they are suitable what has prevented their usage so far?

Are they suitable for PSU designs as I know they are a lot higher voltages 12volts vs 1.2 for example and same questions as above.

From what I understand GaN is more efficient, handles higher current and switching speed is faster. These traits are very useful in a motherboard VRM, as it would reduce thermal loss, package size on the board, and would provide higher current handling as well as faster switching rates, all plusses.

But as posted by sudothelinuxwizard, it's all about cost. But also it will take one motherboard company to come up with a board that makes use of them, and once that happens others will follow because of marketing.

 

[trsttte]

(...) the scaled 300 mm manufacturing leverages Infineon's existing infrastructure investments while supporting rapid capacity expansion for emerging applications, including AI system power supplies, automotive charging systems, and industrial motor control.

God dammit, they had to throw AI in there somehow didn't they!? They're just power supplies, if it's used in a datacenter, a phone charger or some dedicated AI accelerator makes no difference!

are they applicable to say GPU/Motherboard VRM designs

They're probably already being used, it's just not as much of a selling point. You have a lot of features you can publicize with a motherboard, it can have more IO, more memory expansion, support faster ram, whatever, so you don't need to know what's on the VRM - they do like to say how many phases of it there are, as if that meant anything, bigger number equals more better simply.

On a power supply there's not that much you can point towards against the competition, it's supplies power with 0.0001% regulation instead of 0.0002% on the competition, meh. So GaN became a marketing gimmick - hey look we use this cool new thing that makes the thing you don't care about smaller and cooler.

 

[sudothelinuxwizard]

They're probably already being used, it's just not as much of a selling point. You have a lot of features you can publicize with a motherboard, it can have more IO, more memory expansion, support faster ram, whatever, so you don't need to know what's on the VRM - they do like to say how many phases of it there are, as if that meant anything, bigger number equals more better simply.

I doubt that, because I've looked up quite a lot of high-end motherboard VRM part numbers and none have said anything about being GaN. There is no reason for them to use GaN unless they advertise it.

 

[Panther_Seraphin]

The little reading I have done suggenst they are better at higher voltages so I can imagine doing something like DC-DC conversion or rack level power supplies they would be really beneficial for.

 

[Prime2515102]

On a power supply there's not that much you can point towards against the competition, it's supplies power with 0.0001% regulation instead of 0.0002% on the competition, meh. So GaN became a marketing gimmick - hey look we use this cool new thing that makes the thing you don't care about smaller and cooler.

Well there's the selling point: they're more efficient, running cooler for a given power level. This would be very welcome on a motherboard (and especially a PSU) for longevity, if nothing else. The more heat, the shorter the lifespan, not only for the device itself but neighboring components like electrolytic capacitors for example: they dry up over time. Heat hastens that process.

It's also getting more popular in audio amplifiers. Increased power levels for a given size, or cutting the size in half for the same power level, which lowers the cost overall due to chassis and heatsink cost reduction (they ain't cheap!).

So bringing the cost of GaN to parity with Si is very good news indeed (if they can pull it off).

 

[Six_Times]

Why is TSMC planning to shut down its GaN production lines and dismantle facilities within two years?

TSMC is planning to phase out its Gallium Nitride (GaN) production lines by July 31, 2027, due to several strategic and market-driven factors: Price Pressure from Chinese Competitors: Intense competition from Chinese rivals, backed by significant government subsidies, has driven down global GaN chip prices, eroding profit margins. Rather than engage in a price war, TSMC has opted for a strategic exit from this low-profitability segment. Focus on High-Margin Advanced Nodes: TSMC is prioritizing advanced node development, particularly for AI applications, where demand and profitability are higher.

The company is redirecting resources from legacy businesses like GaN to capitalize on the growing demand for technologies such as Chip-on-Wafer-on-Substrate (CoWoS), Wafer-on-Wafer (WoW), and Wafer-Level System Integration (WLSI).Repurposing Facilities for Advanced Packaging: TSMC plans to repurpose its Hsinchu Fab 5, currently used for GaN production, for advanced packaging starting July 1, 2025. Reusing existing cleanroom facilities allows TSMC to expand capacity efficiently to meet surging demand for advanced packaging technologies. Limited Scale and Profitability: GaN production, with a monthly output of 3,000–4,000 6-inch wafers, represents a small and less profitable segment for TSMC. The company’s decision reflects a comprehensive evaluation of long-term market conditions and corporate strategy, prioritizing higher-value segments. Client Transition and Market Dynamics: TSMC’s largest GaN customer, Navitas Semiconductor, is shifting production to Powerchip Semiconductor Manufacturing Corp (PSMC) over the next 12–24 months, citing technical compatibility with PSMC’s cost-effective GaN-on-Silicon technology. This transition aligns with TSMC’s exit, as it works with clients to ensure a smooth two-year wind-down period. TSMC’s withdrawal is not expected to impact its 2025 financial targets, with projected revenue growth of 24–26% in US dollar terms, reflecting confidence in its core business and strategic reallocation of resources.

 

[kondamin]

Why is TSMC planning to shut down its GaN production lines and dismantle facilities within two years?

TSMC is planning to phase out its Gallium Nitride (GaN) production lines by July 31, 2027, due to several strategic and market-driven factors: Price Pressure from Chinese Competitors: Intense competition from Chinese rivals, backed by significant government subsidies, has driven down global GaN chip prices, eroding profit margins. Rather than engage in a price war, TSMC has opted for a strategic exit from this low-profitability segment. Focus on High-Margin Advanced Nodes: TSMC is prioritizing advanced node development, particularly for AI applications, where demand and profitability are higher.

Seems like a rather important technology to keep in house even at a reduced scale.
They aren’t short for cash

 

[Rightness_1]

Why is TSMC planning to shut down its GaN production lines and dismantle facilities within two years?

TSMC is planning to phase out its Gallium Nitride (GaN) production lines by July 31, 2027, due to several strategic and market-driven factors: Price Pressure from Chinese Competitors: Intense competition from Chinese rivals, backed by significant government subsidies, has driven down global GaN chip prices, eroding profit margins. Rather than engage in a price war, TSMC has opted for a strategic exit from this low-profitability segment. Focus on High-Margin Advanced Nodes: TSMC is prioritizing advanced node development, particularly for AI applications, where demand and profitability are higher.

The company is redirecting resources from legacy businesses like GaN to capitalize on the growing demand for technologies such as Chip-on-Wafer-on-Substrate (CoWoS), Wafer-on-Wafer (WoW), and Wafer-Level System Integration (WLSI).Repurposing Facilities for Advanced Packaging: TSMC plans to repurpose its Hsinchu Fab 5, currently used for GaN production, for advanced packaging starting July 1, 2025. Reusing existing cleanroom facilities allows TSMC to expand capacity efficiently to meet surging demand for advanced packaging technologies. Limited Scale and Profitability: GaN production, with a monthly output of 3,000–4,000 6-inch wafers, represents a small and less profitable segment for TSMC. The company’s decision reflects a comprehensive evaluation of long-term market conditions and corporate strategy, prioritizing higher-value segments. Client Transition and Market Dynamics: TSMC’s largest GaN customer, Navitas Semiconductor, is shifting production to Powerchip Semiconductor Manufacturing Corp (PSMC) over the next 12–24 months, citing technical compatibility with PSMC’s cost-effective GaN-on-Silicon technology. This transition aligns with TSMC’s exit, as it works with clients to ensure a smooth two-year wind-down period. TSMC’s withdrawal is not expected to impact its 2025 financial targets, with projected revenue growth of 24–26% in US dollar terms, reflecting confidence in its core business and strategic reallocation of resources.

Ask nGreedia why they screwed gamers when A.I. took off? They only care about what makes the most money.

 

[trsttte]

Seems like a rather important technology to keep in house even at a reduced scale.
They aren’t short for cash

TSMC has the most advanced fabs in the world right now, power transistors are a ballpark error compared to the money they get from doing SoC for big technology companies (Apple, Nvidia, Google, etc.)

Focusing on what they do well and leaving the scraps where they aren't cost competitive is just smart business

 

[kondamin]

TSMC has the most advanced fabs in the world right now, power transistors are a ballpark error compared to the money they get from doing SoC for big technology companies (Apple, Nvidia, Google, etc.)

Focusing on what they do well and leaving the scraps where they aren't cost competitive is just smart business

until you need the expertise in house.
Just think where intel would be now if they hadn't dumped their GPU division at the turn of the century or kept optane going

 

[R-T-B]

Ask nGreedia why they screwed gamers when A.I. took off? They only care about what makes the most money.

That's every company dude. Try to stay on topic please.

 

[trsttte]

until you need the expertise in house.
Just think where intel would be now if they hadn't dumped their GPU division at the turn of the century or kept optane going

I think it's very different situations. Optane was not cost effective too, they were right to dump it as unfortunate as it is seeing that tech going to waste, gpu's is something they still needed so not as smart.

TSMC doesn't depend on this product, it's just another thing they made, and it's likely they will continue to manufacturer other types of power semiconductors, gan was probably not the only thing in the portfolio so they're not really loosing any expertise.

 

[R-T-B]

gpu's is something they still needed so not as smart.

I dunno the gpu they had back then was basically "100 little x86 cores glued together" and it was kinda an awful idea from every angle. Probably best it was abandoned, tbf.

 

[kondamin]

I dunno the gpu they had back then was basically "100 little x86 cores glued together" and it was kinda an awful idea from every angle. Probably best it was abandoned, tbf.

I was thinking about the i740 from back in 1998.
They gave up on that in favor of integrating it on the motherboard later on the cpu die.

then again, it was a giant, and they probably would have gotten even more anti-trust litigation against them if they entered that market too.