SILICON CARBIDE – RISK OF OVERSUPPLY IN 2025?
By Stephen M. Rothrock, President & CEO, ATREG, Inc.
The global silicon carbide (SiC) market has started to transition to eight-inch, an inevitable trend expected to peak in 2026-2027. Today’s market share of eight-inch SiC products is currently less than 2%, but forecasts say it will grow to around 15% by 2026. The number of planned eight-inch SiC fabs worldwide speaks for itself – 14 total, of which 12 are currently under construction. Financed in part by government subsidies over the past couple of years, many chip makers have massively invested in greenfield SiC production lines, with some of these gradually coming online and becoming operational. As the battle for SiC market share heats up in such an uncertain global semiconductor market, companies must balance the potential rewards with the risks that such a competitive landscape brings. Do we run the risk of an SiC overcapacity?
Source: Trendforce, October 2024
As a crucial development in future power electronics, SiC has rapidly penetrated markets such as automotive and renewable energy where power density and efficiency are paramount. About 70% of SiC volumes go to automotive customers, with around 12% to 13% going to industrial applications and a similar amount going to renewable energy. The amount of electricity used by data centers, roughly 2 to 3% of all the electricity generated worldwide, continues to skyrocket due to the demands of artificial intelligence (AI).
But with weaker electric vehicle (EV) demand, industrial end-markets still correcting, and capacity still coming online in China, some SiC semiconductor companies have been hit hard over the last few months. STMicro and Infineon are among those that have both revised their SiC forecasts lower. Wolfspeed, who was the first to market with the world’s largest SiC plant in Mohawk Valley, New York, is one of the only companies who can produce eight-inch SiC wafers in the short term. Due to slower EV adoption, however, the U.S. chip maker had to delay its plans to build a new $3 billion semiconductor facility in Germany. BOSCH has also been manufacturing eight-inch SiC wafers at its Reutlingen, Germany fab. Following a retooling phase beginning in 2026, it will start producing its first SiC chips on U.S. soil at the Roseville, CA fab acquired from TSI Semiconductors in 2023.
So is a lasting SiC glut really on the way? There are valid reasons to be worried. Research reports show that a large amount of capacity was added in 2023 and the same in 2024. With EV sales falling short of expectations in Western markets, there are real concerns that there won’t be enough demand in 2025. Making matters potentially worse, Chinese SiC production is ramping up, raising the risk of the Chinese EV market, the strongest EV market so far, turning away from Western suppliers. One thing is certain – 2025 will be an interesting year to watch for silicon carbide.
AVAILABLE FOR SALE: OPERATIONAL 200MM COMPOUND SEMI FAB, EUROPE
ATREG has been exclusively retained to facilitate the sale of a state-of-the-art operational compound semiconductor fab located in Europe. The facility offers extensive GaN capabilities as well as a track record of doing high-volume CMOS production. It features 40,000 ft² of operational cleanroom space (20,000 ft² immediately available) and approximately 200 process tools (80% 200mm). This facility is currently engaged in R&D development and low-volume micro LED manufacturing with existing profitable revenue.
Key attributes include a comprehensive IP portfolio, 260+ skilled employees (of which 110 are engineers), and specialized capabilities in MOCVD, CMP, and wafer bonding. In addition to the current micro LED capabilities, the fab is suited for GaN power devices, advanced photonics, GaAs processing, and MEMS/sensors manufacturing. Recent infrastructure investments and a 13-acre campus with expansion potential further enhance this opportunity. The existing revenue stream helps defer operational expenses during product qualification and production ramp-up.
The facility is well supplied with water, natural gas, as well as specialty chemicals and features a favorable regulatory ecosystem for compound semiconductor production and R&D. The fab has received significant investment over the past five years, including sub-fab upgrades, cleanroom expansion, utility improvements, and other facility upgrades.
For more details about this 200mm European manufacturing facility, please email ATREG Senior Associate Stuart Smith.
AVAILABLE FOR SALE: OPERATIONAL 200MM DISCRETE FAB, USA
ATREG has been exclusively retained to facilitate the sale of an operational 200mm, automotive-certified discrete fab located in the United States. This cost-effective facility produces 3µm to 0.35µm low-, medium-, and high-voltage MOSFETs as well as IGBT technologies for a variety of end-applications. The opportunity includes a high-end, fully integrated line of 200mm tools capable of manufacturing more than 20,000 wafer starts per month (based on a ~9 mask layer process average). This facility offers significant flexibility and opportunity for expansion both within existing buildings/cleanroom and on adjacent land, and comes with a supply agreement.
The fab team of approx. 200 experienced employees includes an efficient and cohesive process integration workforce consisting of a team of 60+ engineering, functional, and management staff, as well as 130+ technicians, operators, and other operations support staff. The facility offers an efficient 37,500-ft2 Class 1000 cleanroom with a sub-fab (Class 1 SMIF). The facility features roughly 200 well-maintained operational front-end tools consisting of equipment from OEMs such as AMAT, LAM, TEL, KOKUSAI, ASML, NIKON, DISCO, KLA-TENCOR, AXCELIS, and NOVELLUS.
For more details about this 200mm U.S. manufacturing facility, please email ATREG Senior Vice President Nick Papa.
EXECUTIVE Q&A WITH JON HERLOCKER, CO-FOUNDER & CEO OF TIGNIS
ATREG recently sat down with Jon Herlocker, Co-founder and CEO of Tignis, a Seattle-based provider of next-generation AI-enabled process control and process analysis using machine learning (ML) and digital twins for semiconductor manufacturing facilities. We discussed why every semiconductor manufacturing facility should be using artificial intelligence (AI) as well as the implementation of AI in legacy wafer fabs.
What is your take on the AI hype cycle and where is it heading for the semiconductor manufacturing industry?
Before answering that, I think it’s important to define what I mean by AI. For me, AI is inclusive of all forms of machine learning which means that there are so many different aspects of it – some of which are more hyped than others. Today when somebody uses the word AI, they are often referring to Generative AI (GenAI), Large Language Models (LLMs), and other large neural network models. LLM-based chatbots are contributing very real value every day. If you are a knowledge worker and you are not using Gemini, ChatGPT, or some other equivalent LLM every day, then you are working harder than your competition. While the semiconductor industry isn’t fully ready to trust LLMs for technical decisions, I expect that within three years, we will have LLM-based agents that can automate some subset of engineering activities and decisions within fabs. For today, keep in mind that there is a longer tail of AI capabilities – machine learning applied to process control, virtual metrology, predictive maintenance, inspection, etc. You don’t hear much about it because in the semiconductor vertical, only the most advanced node fabs are employing it, and much of it has been “home grown” by internal development teams. Legacy fabs do not have the resources of advanced node fabs, so they are just now coming up the adoption curve, and they are embracing commercial solutions instead of rolling out their own.
Why are legacy process nodes so critical to many industries?
Great question. The word “legacy” probably has a bit of a negative connotation to it and nothing could be further from the truth. It just means that the fab is manufacturing semiconductors at an older (e.g. larger) process node that was actually a “leading-edge” node at one point in time. But due to cost and market demands, there was no need to manufacture certain devices at successively smaller (and more expensive) process nodes. One semiconductor company we know shipped eight billion devices last year at “legacy” process nodes from 130 nanometers to one micron. These are process nodes that were considered “advanced” in the mid-1980s. But the parts they make are still in high demand for many consumer and industrial uses. During COVID, the reason many automakers were unable to ship more cars was because of a lack of these “legacy” semiconductor devices. They aren’t as sexy and newsworthy as AI chips, but they are essential for the devices that we use every day.
In your opinion, what industry is the most dependent on legacy process nodes?
As we just talked about, the automotive industry is very dependent on semiconductor devices manufactured at legacy fabs. Mil-Aero and defense companies are also big consumers of legacy process node chips. And while most people don’t think of it, legacy chips power many everyday appliances from refrigerators to dishwashers to microwave ovens. In fact, legacy chips are essential in the production of most automobiles, aircrafts, broadband, consumer electronics, factory automation systems, home appliances, military systems, and medical devices.
Why should every wafer fab be using artificial intelligence? What are the key benefits?
As you know from ATREG’s decades of success in semiconductors, many of these legacy fabs were built 20 or more years ago before storage and compute power were cheap and plentiful. Fabs accomplished amazing results with what they had back in the day, but they couldn’t leverage all the data they had back then, and they definitely don’t do it now. And with fabs adding more sensors to increase visibility, there just aren’t enough engineers to look at all the data and make decisions rapidly enough. This is why AI is becoming an essential tool for legacy fabs. It helps all aspects of the fab (yield, quality, OEE) and performs better while simultaneously reducing costs and ultimately improving profitability. But perhaps more importantly, AI/ML solutions will increase a fab’s agility and resiliency, ensuring they remain competitive. Given we’ve been talking about legacy fabs, I think it’s important to mention that there is a lot of low-hanging fruit in applying AI/ML to reliability centered maintenance. These legacy fabs have tools that are decades old and finding parts for maintenance is very challenging. Having predictive insights into what parts will be needed for upcoming maintenance events is very beneficial for fab operations.
What about the implementation of AI in legacy semiconductor manufacturing facilities? In what areas can AI be particularly beneficial?
For process or equipment optimization, AI can help at every process manufacturing step in the fab, e.g., etch, litho, CMP, PVD, and ALD. However, AI is most likely to be applied to the bottleneck steps in the process. The more complex the process, the more AI can outperform humans. Another area where we are starting to see AI solutions is the forecasting, planning, and scheduling of everything – wafer moves, maintenance, people, and more. One area where AI can help that most people don’t realize is in the sub-fab where all the supporting OEM tool vendors such as pump and gas vendors are located. When their tools go down for unplanned maintenance, a lot of the big tools in the cleanroom go down as well and that can be very expensive for the fab.
What are some of the obstacles that chip makers should be aware of when implementing AI in brownfield facilities?
Probably the biggest issue with legacy chip makers adopting AI is what we call “AI data readiness”. Historically, fab data is siloed and not well structured for analysis or discovery. AI can ingest tremendous amounts of manufacturing data, but that data needs to be well organized and curated. Because brownfield fabs have run the same way for years, typically there hasn’t been a lot of effort spent on data engineering to create an intuitive semantic model for all the fab data that is well organized and aligned and is ready to be used by AI tools.
What is the way forward for legacy manufacturers as opposed to greenfield fab owners? What advice would you give them?
Legacy manufacturers need to get started now! It doesn’t have to be a complete implementation across the entire fab, but I recommend they start somewhere. Other than their staff, AI is the ONLY appreciating asset in the fab. AI and the machine-learning models they put into manufacturing start learning on day 1 and with the right vendor, those models are better and smarter on day 365 and they will continue to improve by learning all the nuances of their fab operations, helping the team make faster, smarter, and better decisions for all aspects of fab operations. On the other hand, the greenfield fabs we are talking to are planning for the entire fab to be “smart” at inception. Their teams are making sure that AI is fully integrated across the fab and encompasses data engineering, data storage, data science, as well as predictive and prescriptive machine learning. Legacy fabs are in a different stage in their lifecycle versus a greenfield fab when it comes to AI, but they can still greatly benefit from using AI in their fabs. They just need to start somewhere and they need to do it yesterday.
About Jon Herlocker
Jon is Co-founder and CEO of Tignis, Inc., a software company providing artificial intelligence (AI)/machine learning (ML) solutions for smart manufacturing and process control in the semiconductor manufacturing space. Jon is a deep technologist and experienced executive who has a long history of leading innovation in big data analytics and applied AI and ML. His career began in the 1990s where, as the Lead Engineer for MovieLens, he helped create the field of recommender systems as popularized by such companies as Netflix and Amazon. In his prior leadership roles, he served as Vice President and Chief Technology Officer (CTO) of VMware’s Cloud Management Business Unit, CTO of Mozy, and CTO of EMC’s Cloud Services division. Jon is an experienced entrepreneur, having founded three start-up companies. He is a former tenured professor of Computer Science at Oregon State University and his highly cited academic research work was awarded the prestigious 2010 ACM Software System Award for contributions to the field of computer science. Jon holds a Ph.D. in Computer Science from the University of Minnesota as well as a B.S. in Mathematics and Computer Science from Lewis and Clark College.
As 2024 comes to an end, the ATREG team is very grateful and thankful for the continued loyalty and support of its valued clients and partners. This year again for the holidays, we will be making donations on their behalf to the following local and international non-profit organizations:
- L’Arche International – An international federation dedicated to the creation and growth of homes, programs, and support networks for people with and without intellectual disabilities to live mutual relationships, share daily life, and build a community together.
- The Solutions Project – An organization that funds and amplifies climate justice solutions created by Black, Indigenous, immigrant, women, and communities of color building an equitable world.
- Global Partnerships – An international impact investor with the mission to expand opportunity for people living in poverty via micro-financing.
- The Homeless Project / Chief Seattle Club – An organization dedicated to physically and spiritually supporting American Indian and Alaska Native people in the Seattle Area.
Wishing you and yours health, happiness, and success in 2025!
The ATREG Team
The ATREG team looks forward to reconnecting with you in the new year at SEMI’s annual Industry Strategy Symposium (ISS) to be held from January 12 to 15 in Half Moon Bay, CA. Need to divest or acquire a brownfield fab or cleanroom? Looking for a brand-new greenfield building location or extra loading capacity? ATREG can help! Please email us to set up an appointment with one of our fab transaction advisors at the event and discuss your global infrastructure-rich manufacturing asset needs. See you there!
Click here to register for the event.