[ad_1]
During the last annual meeting held by PowerAmerica from March 12 to 15 at North Carolina (NC) State University, two panels were arranged on silicon carbide and gallium nitride and focused on U.S. chip production. SiC panel participants were Microchip Technology’s Kevin Speer, Wolfspeed’s GP Gopalakrishnan and Qorvo’s Andy Wilson. GaN panel participants were Infineon Technologies’ Tim McDonald, Navitas’s Llew Vaughan-Edmunds, GlobalFoundries’ Rajesh Nair, SkyWater Technology’s Ross Miller and Polar Semiconductor’s Tim Maloney. PowerAmerica’s executive director and CTO, Victor Veliadis, delivered the introductory speech. This article reports some highlights from the two panels.
Veliadis on U.S. SiC production
SiC fabrication in the U.S. is a success story. The processes were developed in the U.S. well in advance, thanks to investments in military applications. Veliadis remembers when he was charged $25,000 for 3-inch wafers needed to build 10-kV products for the feasibility study. The idea for SiC manufacturing was to bring extra volume to compensate for the drop in silicon wafer production, with the possibility of operating fabs efficiently. That enabled PowerAmerica and X-Fab in 2016 to create the world’s first open SiC foundry, converting a former Texas Instruments facility that made traditional silicon wafers. One major problem in implementing such a model is the high number of processes to replicate when dealing with many customers, each of them claiming their own patents. That can be overcome by establishing basic process blocks and add the innovative steps case by case for differentiation.
Veliadis offered Microchip Technology as an example of this funding model. It was “a company that started out at X-Fab, doing their production there. Eventually, they went to Colorado Springs [Colorado] and built a line of their own. So their silicon fab would do silicon carbide, and I funded some of that. They were the first company to come with a 3.3-kV MOSFET, which I partially funded. Of course, they put out a lot of money of their own.”
SiC substrate sourcing
Substrate sourcing has been a major issue for those companies jumping into SiC. Recently, the U.S. Department of Energy has agreed to include materials funding in what PowerAmerica is going to do in the future. Companies are trying to fill the gap in substrate supply, with some of China’s players having made significant progress in offering their products. SiC manufacturers have relied on Wolfspeed, which has about a 60% share in bulk wafers, but some companies are moving fast toward vertical integration. Luckily, Wolfspeed has been far-sighted in serving customers (and potential competitors) with whom long-term supply agreements have been signed.
Commoditization of substrates will benefit companies like Microchip that have everything a designer needs to build an entire system solution to help customers adopt SiC, Speer said. Today, the cost of substrate plus epi is just 10% for a finished silicon device, compared with 50% for the SiC. But as Veliadis said, even with 50% of the cost of the device stemming from the wafer, in some applications, the total SiC system solution will cost less. And that’s due to major simplifications at the system level and better thermal management. Such benefits could be enough to promote the deployment of SiC on a wider scale, but reducing the cost of the substrate will give a boost to the whole SiC sector.
SiC killer applications and future prospects
If lower-voltage SiC (650 V to 1,200 V) had been a killer app in automotive, what is a killer app for high voltage? And what are the biggest barriers to get into high voltage to fabricate them reliably? Despite a temporary slowdown in EV sales, the trend will continue to go up. The killer app is probably going to be DC fast chargers in the medium-voltage grid, where operating voltages typically fall in the range from 3 kVAC to 40.5 kVAC. Chargers are not just for automobiles; they could also be for ground power in aviation and energy storage systems. High-voltage SiC can therefore represent an opportunity to modernize the power grid and be used for protection purposes in medium and high voltage, provided that SiC resources, mostly dedicated to automotive, can be rebalanced.
Fully functioning prototypes of solid-state transformers, mentioned at the panel, could be evaluated for commercial use. These 400-kVA machines, delivered to the U.S. Navy, are transforming 4,160 V to 480 V and are implemented with 10-kV SiC MOSFETs in a two-level topology.
In the wrap-up, Veliadis remarked that the government’s involvement was successful because they came in early, making the right investments at the right time and with the right people. But it is also important that the private sector help push the technology forward.
“When it comes to silicon carbide, we do have supply chain resilience,” Veliadis said. “We’ve got several companies that offer substrates. We’ve got several companies that do epitaxy. There are plenty of companies from foundries to fabs that will do the silicon carbide fabrication, companies that will put SiC into modules, and certainly an infrastructure of doing the actual applications.”
CLAWS vs. PowerAmerica
Commercial Leap Ahead for Wide Bandgap Semiconductors (CLAWS) is a semiconductor research hub headquartered and led by NC State University and formed by BluGlass, Coherent, Wolfspeed, General Electric, Adroit Materials, Kyma and North Carolina A&T State University, with the federal government financing the group with a $39.4 million grant. A part of the investment ($14 million) will be used to purchase equipment for the cleanroom at NC State to help set up a pilot line.
The CLAWS project aims to accelerate the development of wide- and ultra-wide-bandgap compound semiconductors. In terms of the motivation, CLAWS is broader than WBG chips and power electronics as a focus, versus PowerAmerica. In terms of applications, it is also looking into the RF space. That means it is seeking to develop avionics and satellite chips to support power supplies and RF and looking into how a chip can handle light in the UV and visible part of the spectrum for quantum applications.
Both CLAWS and PowerAmerica share a common philosophy when it comes to U.S. Department of Defense (DoD) applications that are often in harsh environments. With the Microelectronics Commons initiative, the DoD aims to bridge the gap between microelectronics research and commercialization by fostering partnerships among emerging technology sources, manufacturing facilities and interagency partners. An example is GaN for power electronics, for which an ecosystem approach was worked out that included substrate, epitaxy, device and application. And in this case, for avionics and satellites, EPC Space and Lockheed Martin were put together.
It is necessary that this research foundry concept have flexibility. In other words, some clear benefits must be taken out of a process design kit (PDK), a set of files used within the semiconductor industry to model the fabrication process, to gain control of processes and provide some flexibility of innovation, normally difficult to get from a rigid PDK.
Synergy between CLAWS and PowerAmerica
One way to achieve the best synergy between the two hubs is to be co-located on campus, where continuous dialogue and collaboration are conditions for success. But how can a smaller-volume application help have a more significant impact on a larger-volume one? Let’s take, for instance, an aerospace or satellite application, a niche application with very special requirements. By learning from niche applications, one could infer new rules for making automotive parts better and more robust.
Another example is SiC work at General Electric. Matching certain requirements for the DoD can give new ideas for even higher-voltage applications, such as those found in power grids.
There are also big opportunities for high-voltage packaging, especially in critical applications in which electronic parts not only have to work today but must do the job for at least 20 years. On large vessels and submarines, where a 13.8-kV bus represents the backbone for electrical power distribution, components need to be housed in special packaging. Additionally, such devices ideally need to last for the entire life of the ship or at least be replaceable. And by looking at how the lifecycles of submarines have been expanded, designers have all elements to push lifetime to 30 and even 40 years.
On the technology side, there is a lot of expertise, not just in the packaging but also on how to use the packaging in circuits at NC State University, and electronic design automation is an important part of that, in terms of both the thermal and electro-thermal interactions.
GaN opportunities in data centers
Navitas’s Vaughan-Edmunds gave an interesting overview of GaN’s impact on data centers due to the ever-increasing use of chips for AI processing requiring power supply units (PSUs) to handle higher power densities. Typically, an AC/DC unit in a cabinet delivers 3 kW of power to supply about eight to 10 GPUs. GPUs released a couple of years ago by Nvidia consumed about 700 W. The Nvidia Blackwell GPU being released later this year burns 1 kW, but it has to use the same rack space. So the next-generation PSU needs to manage 10 kW, and GaN can allow for that power density, a job that silicon just cannot do.
There will be a massive growth of GaN in data centers as the demand for services grows exponentially, driven by the unstoppable expansion of cloud computing, IoT, streaming entertainment and big data analytics. As an example, a single generative AI session like ChatGPT consumes up to 100× more energy than a typical Google search. According to some estimates, more than 13 million servers are shipped each year, and each one can have over $75 of GaN content, so the potential market for GaN in data centers represents a business worth $1 billion.
[ad_2]
Source link
