Will a fabless approach work in the power semi market?
Several third-party foundry vendors are entering or expanding their efforts in the silicon carbide (SiC) business amid booming demand for the technology.
However, making a significant dent in the market will not be so easy for SiC foundry vendors and their customers. They are facing stiff competition from traditional SiC device vendors such as Cree, Infineon, Rohm and STMicroelectronics.
SiC, a compound semiconductor material based on silicon and carbon, is used to make specialized power semiconductor devices for high-voltage applications like electric vehicles, power supplies, solar and trains. SiC stands out because it’s more efficient with higher breakdown voltages than traditional power semis.
In SiC, the integrated device manufacturers (IDMs) dominate the landscape. Cree, Rohm and others make devices in their own fabs and sell them under their own brand names. Those companies use proprietary processes, which enables them to differentiate their products.
The IDMs compete against each other, as well as an emerging crop of SiC fabless design houses. Fabless companies and others have their products made by a foundry vendor. SiC foundries give customers access to manufacturing capacity, but there are some cost and supply-chain challenges here. Generally, SiC IDMs don’t provide foundry services or make chips for others, but that could change one day.
The SiC foundry business is just getting off the ground. At some point, though, SiC foundries hope to replicate the successful silicon foundry model. In this model, many chip companies outsource their IC production to the silicon foundries, such as TSMC, Samsung, GlobalFoundries and UMC. These foundry vendors are not participating in SiC. Today, the SiC foundry business is still small.
As it turns out, the SiC foundry business is more difficult than the silicon foundry segment. Near term, the SiC foundry business will resemble other power semi markets. The best example is the insulated-gate bipolar transistor (IGBT) segment, a power semi type that competes against SiC devices.
“When you’re talking about the silicon foundry model, it is very successful,” said Hong Lin, an analyst at Yole Développement. “I am not here to say there is no chance for the foundry model in silicon carbide. In the power semiconductor business, if we’re looking at today’s IGBTs designs, there are foundries here. But it’s more of an IDM business.”
Still, the industry needs to keep an eye on the events in the arena. Among them:
What are power semis?
In total, the SiC device business grew from $420 million in 2018 to $564 million in 2019, according to Yole. The big growth driver is battery-electric cars. Power supplies and solar are also strong markets.
“In our forecast, there is still growth in 2020,” Yole’s Lin said. “If we look at the five-year CAGR, it’s close to 30%. If we look at this year, there are some products that will continue to ramp up. But the real ramp up, in particular with the automotive market, is later. It’s not in 2020.
“There is no doubting the technical benefits of using wide band-gap instead of silicon-based technology for different inverters and converters for EV/HEV,” Lin said. “R&D programs and technical developments have shown positive results, including reduction in size and weight and efficiency improvements for both SiC and GaN.”
Fig. 1: Power SiC device forecast Source: Yole Développement
SiC is also competitive. Nearly two-dozen SiC device suppliers compete in the business. “The market is still very nascent. We are still in the phase that we are seeing more and more companies here. We may see it go towards a consolidation phase,” Lin said. “It’s premature to predict if the fabless companies will be successful or not. Today, it’s too early to tell.”
SiC power semis are one of many types of power devices in the market. Power semis are specialized transistors that operate as a switch. They allow the power to flow in the “on” state and stop it in the “off” state. These devices boost the efficiencies and minimize the energy losses in systems.
For years, the power semi market has been dominated by silicon-based devices, namely power MOSFETs and IGBTs. Power MOSFETs are used in applications up to 900 volts. These include adapters and power supplies. The term “volts” denotes the maximum allowed operating voltage in the device.
IGBTs are used in midrange applications from 400 volts to 10 kilovolts. IGBTs are used in automotive, industrial and other applications.
Both power MOSFETs and IGBTs are mature and inexpensive, but they are also reaching their theoretical limits. That’s where SiC and gallium nitride (GaN) fit in. Both GaN and SiC are wide-bandgap technologies, meaning they are more efficient than silicon-based devices. For example, compared to silicon, SiC has 10 times the breakdown field strength and 3 times the thermal conductivity.
“SiC and GaN are promising components for the power semiconductor market due to the higher efficiency and smaller form factor characteristics compared to their silicon-based peers like power MOSFETs and IGBTs” said Steven Liu, vice president of corporate marketing at UMC. “The devices can be made much smaller in size for the same relative voltage and current handling capability. GaN has made inroads in applications requiring 600 volts and below, while SiC has made inroads in applications requiring 1,200 volts and above.”
Device makers sell SiC power MOSFETs and diodes, which are used in 600-volt to 10-kilovolt applications. A SiC power MOSFET is a power switching transistor. A diode is a device that passes electricity in one direction and blocks it in the opposite direction.
The downside with SiC is cost. The devices are more expensive than power MOSFETs and IGBTs.
The IDM world
Power semis are different than digital CMOS chips. They can withstand high voltages and large currents in systems.
In CMOS, chip suppliers put an emphasis on IC design to differentiate their products. Many chip suppliers also outsource some or all of their production to silicon foundries. Foundries develop a baseline process and customers design a chip around it.
The power semi segment is different. “Many foundries provide MOSFET/IGBT production services to customers, including UMC,” UMC’s Liu said. “However, IDMs still continue to dominate discrete ICs in terms of shipment volume.”
Generally, in power discretes, the foundry doesn’t develop a baseline process. Instead, a fabless design house develops a device based on the company’s proprietary process. Then, the process from the fabless company is ported to a foundry.
Each fabless customer may have a different process. So the foundry must port and maintain different processes for numerous customers.
“That flies in the face of the normal foundry model. That’s how you get your high volume and cost savings by running a single process that everybody designs into. It’s just different masks and ways to wire up a circuit,” said John Palmour, CTO of power and RF at Cree. “That doesn’t work for power devices. I’m not saying that a foundry model won’t work. They are doing business. But the big guys in silicon carbide do not use foundries. They all use their own fabs.”
There are several reasons for this. In simple terms, a given device is developed and optimized around a process under one roof in the fab. The device and process are tightly coupled in the fab. So the emphasis is developing a proprietary process, as opposed to IC design. In effect, design and process are one and the same.
“In other words, you can’t just draw up a power MOSFET, throw it over and say print this,” Palmour said. “The reason is that there is no circuit design. You can’t differentiate yourself from your competitor with circuit design because there’s no circuit. It’s one big device. The product that you sell is actually the process. Your design is entirely dependent on the process and vice versa. And that’s how you differentiate yourself from your competitor.”
So a SiC power device isn’t exactly a circuit from a traditional standpoint. “It’s a bunch of little discrete MOSFETs all tied in parallel,” Palmour said. “Think of it as one of the bits in CMOS. You’re going to design your own bit and then you’re going to parallel 500,000 of them. It’s not a standard process. But you are changing the fundamental structure between every customer.”
All told, the IDM model works for SiC. The SiC market is dominated by several IDMs, such as Cree, Infineon, On Semiconductor, STMicroelectronics, Rohm, Toshiba and others.
Not all IDMs are alike. A few are vertically integrated. For example, Cree makes its own SiC substrates. It not only uses the substrates for its own products, but it also sells them to competitors. In addition, Cree fabricates and sells SiC devices to customers.
Rohm and STMicroelectronics are also vertically integrated. Integrated suppliers can control the supply chain, enabling them to quickly react to the demand cycles.
Most IDMs aren’t vertically integrated. Most are buying the substrates from Cree, Rohm or third-party suppliers.
For all vendors, though, there are several manufacturing challenges in SiC. In the SiC flow, a vendor obtains a SiC wafer, which is then processed in a 100mm (4-inch) or 150mm (6-inch) fab. This, in turn, creates a SiC power device.
The biggest challenge is the SiC substrate. It’s too expensive, which drives up the cost for SiC power devices. “One of the key challenges we see in the SiC market today is securing high quality substrates for any company’s production plans at a price that makes sense,” said Mukund Raghunathan, product marketing manager at KLA.
The SiC substrate production process is complex. It starts with silicon and carbon materials, which are inserted in a crucible. In the crucible, a boule is formed and then sliced into SiC substrates.
“There are challenges associated with SiC as it has proved difficult to handle, grind and saw, compared to silicon,” said Rich Rice, senior vice president of business development at ASE.
During this or other processes, the SiC substrates are prone to various defect types. “Most of the challenges are related to the quality of the SiC material,” said Llewellyn Vaughan-Edmunds, director of strategic marketing at Applied Materials. “Killer defects such as basal plane defects as well as general defects such as threading screw dislocations need to be reduced.”
Following the SiC wafer process, an epi layer is grown on the substrate. Then the wafer is processed in the fab, resulting in a SiC device.
SiC MOSFETs are based on two structure types—planar and trench. Planar incorporates a traditional source-gate-drain structure. Trench forms a “U shape” vertical gate channel.
“Turning to wafer processing and plasma etch, there are two choices in trench profile when it comes to non-planar SiC devices. Users ask for trenches with no micro-trenching (flat base) or bottom rounding (test tube shape). Whichever profile is used, end-pointing is desirable to stop the etch front at the right depth,” said Kevin Crofton, president of SPTS and senior vice president at KLA. “For SiC devices, conducting film deposition by PVD is very similar to that used in the silicon MOSFET lines. In both applications, the device makers want to deposit thick (4-microns plus) aluminum alloys with no defects. One of the challenges is to deposit these films at a high rate without whiskers. Another challenge is to track and align the transparent wafers around the wafer processing system, particularly with legacy hardware.”
The SiC industry requires some breakthroughs. “Today, the substrate costs are approximately 50% of the processed wafer. The goal and focus in the industry are to reduce this cost and increase the supply, while also improving quality,” Vaughan-Edmunds said. “Therefore, new ways to speed up the growth of SiC boules, improve uniformity, enable faster slicing and high-precision CMP are a big focus.”
Others agree. “Despite the considerable focus on expanding the commercialization of SiC, it is still immature compared to silicon-based IGBT technologies,” said David Haynes, managing director of strategic marketing at Lam Research. “This is particularly true in automotive applications where reliability is critical and product certification can be complex and time-consuming. To close this gap and accelerate adoption, a holistic understanding of defectivity in SiC device fabrication is required.”
Fabless-foundry model
Meanwhile, customers can buy SiC devices from IDMs and fabless companies. As stated, fabless companies and others use foundries.
In CMOS, this relationship is often referred to as the fabless-foundry model. This model was first introduced in the 1980s, but was quickly dismissed. Later, though, it proved to be a big success.
The SiC fabless-foundry model is relatively new, and it comes with an assortment of challenges. While IDMs will continue to dominate, there is also room for the fabless and foundry vendors. In fact, several fabless companies already are ramping up products using foundries.
“The fabless model allows startups and smaller companies to test their products without significant process equipment investment,” KLA’s Raghunathan said. “Conversely, traditional fabs retain the advantage of being the strategic vendor of choice for major customers. Both models are playing to their respective strengths, serving the varied needs of the current industry landscape and finding ways to coexist.”
Over the years, a few vendors provided SiC foundry services. But the big effort started in 2015, when the U.S. Department of Energy and North Carolina State University formed PowerAmerica, a partnership between industry, government, national labs and academia. PowerAmerica’s goal is to accelerate the commercialization of GaN and SiC.
As part of its efforts, PowerAmerica in 2016 provided support for X-Fab, which was (and still is) developing a SiC foundry service in its 150mm fab in Texas. Working with PowerAmerica, X-Fab devised process kits and other technologies for making SiC devices.
Others are developing similar services. Episil is in the process of transforming its SiC foundry fab from 100mm to 150mm. And Sanan launched a 150mm SiC foundry service.
Still others are exploring the business. “More and more foundry players are now interested to be involved in this market,” UMC’s Liu said.
Today, X-Fab is in production with several SiC customers. But bringing up a SiC foundry business presents some challenges. “Investments need to be made to support the business considerably in advance of volume production,” said Christopher Toelle, business unit manager at X-Fab. “Creating the right foundry infrastructure to align to the needs of SiC power customers is also important.”
According to Toelle, here are the other challenges in the arena:
Despite the challenges, several SiC device suppliers are producing products using foundries. These suppliers include ABB, GeneSiC, Global Power, Microchip, Monolith, and UnitedSiC.
It makes sense to have a fab. But it doesn’t make economic sense to have a fab unless you can process from 10,000 to 30,000 SiC wafers a month, according to UnitedSiC.
Only a few vendors are producing devices at those volumes. If not, it makes sense to use foundries. “The reason for going down this path is that it’s really the most capital efficient way to build a semiconductor business,” said Chris Dries, president and CEO at UnitedSiC, a fabless SiC device supplier. UnitedSiC has its 150mm products made by X-Fab. It also uses an undisclosed vendor for 100mm capacity.
Having a foundry is only part of the equation. “If we were just to make a ‘me too’ MOSFET, we wouldn’t be competitive. As a fabless company, you have to buy a substrate and get epi put on it. Then, you need a foundry and a packaging house,” Dries said. “How do we as a fabless company compete against the larger vertically integrated competitors? You have to out innovate your vertically integrated competitors. That’s exactly what we’ve done. We have the lowest specific on-resistance devices in the world today.”
UnitedSiC sells several products, including SiC junction gate FETs (JFETs). “We make normally-on JFETs, and then cascode them with low-voltage silicon MOSFETs. It’s more of a composite device,” he said.
Still, some customers may be skittish in dealing with a fabless company. To help matters, UnitedSiC has strong backing. ADI has invested in UnitedSiC and has a supply agreement with the company.
But it won’t be easy to compete against IDMs. Process costs are high. “Trying to compete on price with the IDMs is nearly impossible,” according to one expert.
There are other hurdles. In CMOS, many IDMs stopped building advanced fabs because they became too expensive. Instead, many IDMs decided to outsource some or part of their production to foundries.
It’s unclear if the outsourcing model will work for SiC IDMs. “They already have huge, fully depreciated fabs. They want to put stuff in their own factories and leverage the volumes. They have no motivation to use a foundry,” Cree’s Palmour said.
Conclusion
Clearly, SiC is a hot market, thanks to potential demand from the electric vehicle market. This in turn has attracted a growing number of new players, including device makers and foundries.
But it’s unclear if there is room for everyone. It’s highly doubtful, even if the market takes off in a big way.
Related Stories
 |
 |
 |
 |
 |
 |
 |  |
Very good SiC market analysis report. Thanks.