Power Semis Usher In The Silicon Carbide Era

Silicon carbide production is ramping quickly, driven by end market demand in automotive and price parity with silicon.

Many thousands of power semiconductor modules already are in use in electric vehicles for on-board charging, traction inversion, and DC-to-DC conversion. Today, most of those are fabricated using silicon-based IGBTs. A shift to silicon carbide-based MOSFETs doubles the power density and accelerates switching speed in a smaller and lighter package.

The high voltage and ability to operate in hot, harsh environments are growing requirements in EVs and charging stations, but silicon carbide (SiC) has taken time to gain solid footing due to the cost of fabrication and packaging of this wide-bandgap material. That is changing, however. SiC power modules now are at price parity with silicon-based modules, according to Victor Veliadis, executive director and CTO of PowerAmerica, which in turn is prompting supply partnerships and the construction of new SiC fabs.

There is still much to be done. SiC wafer technology needs upgrading. To fabricate these devices requires 20% new process tools, and 80% modified tools. And the goal is faster turnaround for integrated and discrete power devices, which is why carmakers are moving to direct fab-to-module collaboration.

Fig. 1: Acquisition and partnership agreements to ensure supply and rapid technology advancement. Source: Yole Intelligence, 2023 Power SiC report, in press

The new wafer process tools include high-temperature epi growth (>1,500°C), hot ion implantation, rapid thermal processing (RTP), and faster, pulsed atomic-layer deposition. Significant modifications are taking place in wafer grinding, CMP, polishing pads, and slurries for the hard yet brittle SiC material. New materials including strippers and cleaning chemistries address device and sustainability needs.

From the packaging end, high-power printed circuit boards with discrete components are being replaced with integrated circuits and integrated packages like chip-scale packages (CSPs) for smaller, more reliable high-voltage operation. That enables EVs with smaller, lighter battery packs, which helps to increase driving range. While the focus today is on SiC power, and extending Si power modules for hybrid and EVs, tomorrow’s SiC modules will dominate in EVs. In addition, GaN will find niches in EVs, grid power, and smart energy.

Market, technology for SiC and GaN power
The world could produce 39 million battery electric vehicles by 2030, corresponding to a growth rate of 22% CAGR from 2022 to 2030. This, in turn, fuels the power semi market, which is forecast to utilize about 50% Si devices, 35% SiC devices, and 12% GaN devices by 2030. In EVs, the traction inverter converts DC from the battery pack to AC, which powers the motor that drives the front and rear axles. SiC also accelerates on-board and off-board charging, bringing power from the grid into EVs.

Most importantly, SiC modules form the cornerstone of the switch from 400V batteries to 800V batteries. Consumers will adopt EVs faster when there is access to faster vehicle charging, adequate range, and battery costs of less than $10,000 per vehicle.

SiC modules are reaching the tipping point where they are at price parity with silicon-based power solutions, while enabling a much more efficient, compact system. This, combined with the extended range of 800V batteries over the 400V used today (containing 600V or 650V devices), is spurring high volume production of 1,200V SiC devices. However, the yield hits from crystalline defects in wafers, losses from device packaging and module integration, and supply chain changes such as closer links between carmakers and power system makers are still works in progress. And from a practical standpoint, new SiC wafering and fab capacity will take time to ramp to high volumes.

That hasn’t affected enthusiasm for the technology, however. Analysts continue to adjust their SiC market forecasts upward. Yole Group is expecting the power semi market to hit $6.3B by 2027, of which 70% is for automotive applications. Looking just at SiC wafer production (starting SiC wafers), TECHCET forecast a 14% compounded annual grow rate from 2022 to 2027.

Fig. 2: Carmakers are moving toward more direct collaboration with module suppliers, and eventually chipmakers. Source: Yole Group/SEMICON West

IDMs, foundries, fabless activity
Leaders like Wolfspeed, STMicroelectronics, onsemi, Rohm, Infineon, and Bosch are key players on the chipmaking side. The largest cost contributor to these devices, the SiC wafer, is beginning to migrate from 150mm to 200mm fabrication, but the growing, slicing, and preparation processes still rely on costly, time-intensive manual labor operations.

All parties, especially IDMs and foundries, are pushing hard on lowering defectivity in the SiC lattice, developing SiC-specific tool platforms like high temperature ion implantation, epi deposition furnaces that operate above 1,500°C, and improved CMP slurries, pads and cleaning chemistries to process a material that is nearly as hard as diamond.

The technology switch is complete because silicon-, SiC-, and GaN-based power circuits all compete in the 400V battery range. However, SiC power systems are capable of delivering much higher power levels than GaN (see figure 3).

“I call this the 650V battleground, because really all three technologies are competitive in this range,” said PowerAmerica’s Veliadis, who took part in the “Connecting the Automotive Ecosystem with SiC Manufacturing” forum at SEMICON West. GaN has higher electron mobility than SiC, but it is less mature and cannot match SiC’s high power levels. Even so, GaN has substantial appeal in manufacturing high-frequency devices. In addition, some current GaN-on-silicon approaches by Intel, imec, others look quite promising.

Fig. 3: The power density operating windows of SiC, Si and GaN (left) overlap at the low end for 650V devices (400V batteries), but 1,200V devices for 800V batteries are coming. Source: PowerAmerica

Silicon carbide modules are considered essential for enabling higher electric drivetrain efficiency of EVs. The dramatic shift from silicon-based to SiC-based device will go a long way toward increasing the power density of electric systems while reducing the size, weight and most importantly, cost of EVs. This is happening because silicon-based power semiconductors, though still being optimized, are reaching their operational limit in terms of conduction and switching losses. Silicon carbide’s wider bandgap (3.26eV vs. 1.12eV for silicon) reduces such losses and delivers superior high-temperature and high-frequency performance.

To date, many SiC chipmakers have converted 150mm silicon fab lines to SiC fabrication. “The model that’s been very successful so far has been to process silicon carbide in mature, fully depreciated silicon fabs with a modest capital investment of about $30 million, and the reward, of course, is massive,” said Veliadis, noting that it’s system cost that matters most in power modules. “With silicon carbide, you’re going to pay about three times more for the semiconductor chip, but you end up with a system cost that is lower than that of silicon power modules, which is counter-intuitive. But the answer is simple. The ability to operate efficiently at high frequencies reduces the volume of the magnetics and passive components so significantly that it outweighs the higher cost of chip manufacturing.”

However, the industry is running out of older fabs that can be refurbished for $30M. New SiC fabs are being facilitated quickly. In the meantime, fabless companies are scrounging for capacity.

“We have two markets that are competing with each other — the automotive market and the renewable energy market, which are hunting for capacity,” said Ralf Bornefeld, senior vice president for power semiconductors and modules at Robert Bosch. “We learned from the Covid pandemic that a competing market can shut off another market, so we need to take that into account.” Bosch currently is producing its third generation of SiC MOSFETs modules, with breakdown voltages of 1,200V.

SiC devices are particularly suited to automotive because they can deliver high power density with higher temperature operation in harsh environments. SiC power devices can achieve extremely low switching losses and ultra-low RDSon (resistance between source and drain during operation). A smaller RDSon correlates with lower power loss in MOSFETs.

Device capability starts with the SiC material. ”The crystal quality is the number one element that the key players have been addressing over the past 20 years, but there are still basal plane dislocations, stacking faults, and so forth in the crystal that need to be engineered to make 20, 30 and 40 square millimeter devices,” said Christophe Maleville, chief technical officer and senior executive vice president of SOITEC’s Innovation. “When we got into the silicon carbide pool four years ago, the first thing we noticed was the viability in every boule and every wafer is different, and often engineers need to adjust and verify the epitaxy. In order to implement a lean manufacturing process, SOITEC has developed its SmartSiC substrate.”

Electrically, power devices can be sensitive to parasitic inductance, sparking, and other challenges. Not unlike analog mixed signal factories, where parametrics are a primary concern, power engineers contend with variation.

“In the past, (analog) lacked the shrinking. But they already had mature processes in terms of defects,” said Dieter Rathei, CEO of DR Yield. “As compound semiconductors like SiC, GaN, and GaAs become more mainstream and have faster growth rates, parametric yield issues will improve.”

Vertical integration vs. collaborative wafer development?
Today’s wafers in the 100mm and 150mm sizes mostly use monocrystalline silicon carbide with hexagonal lattice structure (4H and 6H indicates 4-in. and 6-in. hex wafers). But the move from 150mm to 200mm is well underway by the largest SiC device producers, and others are tapping into that supply.

Infineon, for example, obtains wafers from multiple suppliers, according to Yole Group analysts. These include ST’s majority acquisition of Nortel in Sweden. And Renesas, a provider of silicon power devices, is beefing up capacity and its partnerships. In July, Renesas signed a 10-year agreement and placed a $2B deposit with Wolfspeed to supply 150mm bare and epitaxial SiC wafers. Renesas also has an agreement with Mitsubishi, which is spending ¥260 billion on technology and expansion including a new SiC fab in Japan.

“[Renesas] was a latecomer in conventional power semiconductors, but now [our products] are valued for their high efficiency,” said Hidetoshi Shibata, the company’s president, in a recent release. “The same can be done with SiC.”

Meanwhile, SOITEC and STMicroelectronics are exploring monocrystalline-on-polycrystalline SiC approaches, which splits the monoSi wafer into multiple slices and re-uses the donor wafer substrate to reduce waste. The advantage to a polySiC base is the ability to channel heat through the substrate to metal connectors, for faster switching and superior heat dissipation.

In some ways, SiC is following silicon’s trajectory. But because of the defectivity levels in SiC, some data sharing is needed.

“We exchange device data with the silicon wafer supplier’s raw material data,” said Bosch’s Bornefeld. “We also use advanced AI-based systems to identify good correlations and share this so that both companies make steps forward.”

Nonetheless, data sharing is not widespread. Also unlike silicon, boule scaling from 150 to 200mm does not have a high payoff in the form of many more wafers/boule. In addition, a larger seed is needed for 200mm, which requires much longer to grow at 2,500°C. Increased productivity (wafers/boule) could be in the 20% range today. TECHCET analysts estimate the cost contributor for boule growth will drop relative to slicing, grinding, polish and CMP, etc. (see figure 4).

Fig. 4: Maximizing the number of SiC wafers per boule is critical due to the high cost of material per millimeter of boule height. Source: TECHCET

Automotive chips outpace EVs
“We see that the growth in demand for semiconductors for automotive applications is changing at a much faster pace that the growth in the production of electric vehicles,” said Lee Bell, director of product marketing for automotive smart power and discrete product marketing at STMicroelectronics. “This is due to a number of factors. Advanced driver safety features, autonomous vehicle controls, advanced connectivity, and convenience features all drive semiconductor demand, but not in the same way that the electrification of the power train does,” he said. “In 2022, about two-thirds of all electric vehicles were hybrid vehicles, with about one-third being battery powered. By 2030, that trend will reverse itself. This is due to increased market acceptance, greater availability of charging infrastructure, but probably most importantly, this is where the car makers are placing their R&D and manufacturing budgets.” This change is the key driver toward using SiC MOSFETs.

Bell noted that traction inverters tend to be larger die. He added that the charging system in the vehicle, and the DC-DC converters that lower the voltage from battery to IoT systems, are huge consumers of power semiconductors. Neither exists in hybrid car architectures.

He also emphasized the primary focus on efficiency – packaged device and module – because less power loss in the system translates directly to longer range cars and trucks. “We did a study comparing a 210kW inverter system, equivalent roughly to 280 horsepower, versus a SiC MOSFET and silicon IGBT (insulated gate bipolar transistor),” he said. “The silicon carbide approach consistently gives 98% operating efficiency, while the IGBT approach delivers lower efficiency particularly in the low operating load range, where the vehicle spends about 95% of its life.”

Total power is the on-state losses plus switching losses. “Switching loss in silicon carbide is reduced by a factor of four,” he said. ST is producing its fourth generation of SiC products, which offer a 30% improvement in RDSon.

Bosch’s Bornefeld showed the demand and capacity estimates through 2030, suggesting the global wafer and fab capacity coming online in Japan, Korea, China, Malaysia, Germany, Austria, and the United States is quite substantial. In fact, the industry needs to be careful not to overbuild (see figure 5). “The question is, ‘what’s going on in China?’ China is already leading in terms of raw material for silicon carbide, and they are delivering very high-quality, reasonably priced wafers,” Bornefeld said. “They are catching up quickly on devices, as well. So we really need to observe and track the overall capacity.”

Fig. 5: Mapping of worldwide facilities for processing SiC raw material into wafers. Source: Semiconductor Engineering/Laura Peters

Finally, Veliadis of PowerAmerica talked about the need for workplace training to skillfully implement wide bandgap semiconductors such as SiC and GaN in fabs. “Engineers with significant experience in building MOSFETs in SiC and GaN are in short supply and there are significant differences between a SiC fab and a silicon fab.”

Conclusion
The clean energy and EV transformations will require alternative semiconductor materials like SiC and GaN, and power devices are sure to be significantly optimized over the next couple decades. The frenzy of technology improvements and capacity expansions may not last, but power devices will remain key to many company’s roadmaps.

“We know that the semiconductor industry is on a path to reach a $1 trillion market, but everyone wants to know what’s going to happen after 2030,” said David Britz, head of strategic marketing of ICAPS at Applied Materials. “I’m here to make the case that the fifth era of semiconductors is really getting driven by transformation in energy generation and transportation.”

The management of growth in SiC wafers, devices, and modules may be the toughest aspect of the SiC market so far, along with supply chain issues, filling in technology gaps, and geopolitical changes. Still, it appears that the semiconductor tech community is in agreement on many things — particularly the need for next-generation power efficiency and performance.

Related
Power Semiconductors 2023 eBook: Deep Dive Into Materials, Manufacturing & Business

How these devices are made and work, challenges in manufacturing, related startups, as well as the reasons why so much effort and resources are being spent to develop new materials, and new processes.