The recent International Microwave Symposium (IMS) provided a good opportunity to judge the state of GaN and its progress toward becoming a mainstream RF technology.
With its high power density, GaN was envisioned for defense applications in the 1990s. The technology was first fielded in the mid 2000s for what was called IED (improvised explosive device) defeat. Cree and Sumitomo were early suppliers of GaN on SiC devices for the IED jammers used in Iraq and Afghanistan. The technology was also used in military communications; Nitronex shipped around a million GaN on Si devices for the Falcon multi-band tactical radio.
In commercial markets, RFMD successfully inserted GaN into CATV amplifiers, where the technology provided the needed output and linearity with the advantage of lower power consumption. Now Qorvo, RFMD's and TriQuint's GaN plans have merged into a single product roadmap, leaving Anadigics the primary competitor for GaN line amplifiers. The major CATV opportunity is the migration to DOCSIS 3.1; however that market isn't large enough — even with China's population and network upgrade — to load the industry's GaN fabs.
Judging from the IMS exhibition in Phoenix, GaN has been widely adopted as the core power technology for kilowatt-level power amplifiers (PA) that are used for defense, electromagnetic compatibility and other industrial applications. Multiple products from Analog Devices (formerly Hittite and Keragis), Empower RF Systems, IFI (Instruments for Industry), Teledyne Microwave Solutions and others indicate the industry's confidence in GaN's performance and reliability.
The Battle for the Base Station
That leaves the PA in the cellular base station (BTS) as the major market opportunity and battleground for GaN, a market worth “just over $1 billion,” according to Lance Wilson, analyst at ABI Research.
Sumitomo was the pioneer, supplying GaN on SiC devices for the Japanese cellular market. Cree was also an early mover into base station PAs, working through various foundry engagements and strategic partners. Most visibly, in June of 2009, Cree announced a relationship with RFHIC that combined Cree's GaN on SiC device and process capabilities with RFHIC's amplifier design, packaging, assembly/test and channel to market. According to Ryan Baker, Cree's marketing manager for RF components, they still participate “very heavily” in the BTS market.
ABI's Wilson estimates that GaN has captured only around a 10 percent share of the cellular infrastructure market, with most of that split between Sumitomo and Cree. In Wilson's view, the high cost of GaN has limited its growth.
According to Paul Hart, senior vice president and general manager of Freescale's RF business, GaN remains twice as expensive as LDMOS, using the standard metric of $/watt. Hart should know, since Freescale has the largest market share of LDMOS PAs and just introduced their “first” GaN transistor for BTS. The GaN device supports 30 to 40 watt PAs in the bands from 1800 to 2200 MHz.
Actually, Freescale's very first GaN PA for cellular infrastructure was announced at the 2012 IMS in Montreal. A 350 watt, 2:1 asymmetric device covering 2300 to 2700 MHz, that product quietly disappeared from Freescale's website. Hart said GaN technology wasn't really mature then. This new device is based on their fourth generation of GaN, which they source from an external foundry. He also said the trend is for new base stations to be LTE-only, which makes linearizing the GaN PA much easier and the performance better.
It's interesting that Freescale chose the 1800 to 2200 MHz bands for their initial product, rather than 2300, 2700 or higher. While the middle cellular bands offer the greatest volume, GaN's efficiency advantage over LDMOS is stronger at higher frequencies. Presumably, that advantage would better justify GaN's premium price.
NXP's High Performance RF (HPRF) segment holds the #2 position in supplying LDMOS PAs for wireless infrastructure. They seem to be taking tentative steps to develop a GaN portfolio. Their most recent product catalog, published in May, listed eight GaN PAs, of which six were not yet released for mass production. These devices provide 10 to 100 watts of output power, with 3.5 GHz the upper frequency of the higher power devices.
HPRF's pending sale to Jianguang Asset Management Co. Ltd. (JAC Capital) may create some uncertainty or delay in the development of additional GaN products. On the other hand, in a letter to customers, NXP stated JAC Capital will increase investment in R&D and manufacturing to strengthen their market position.
Perhaps the most interesting plays in this space will be from Qorvo and MACOM, neither with a position in BTS power and both challenging the incumbent LDMOS to make GaN mainstream.
Qorvo is leveraging TriQuint's GaN on SiC technology that was largely developed with U.S. DoD funding. At IMS, Qorvo announced their first GaN PAs packaged in plastic, a critical step to reduce cost. Their initial product development seems targeted at the higher frequency cellular bands. See, for example, the article on their 3.5 GHz PA reference designs in the June issue of Microwave Journal.
Rather than compete head-on with the rest of the industry, MACOM has defined a blue ocean strategy by choosing GaN on Si — their motivation for acquiring Nitronex. With their latest epitaxial tweaks, which yielded Gen 4, MACOM claims their GaN on Si has equivalent power, efficiency and gain as GaN on SiC. MACOM argues that a GaN on Si wafer populated with RF devices looks the same to an 8 inch, high volume wafer fab as a GaN on Si wafer carrying power electronics devices. By extension, the RF wafer will cost the same as the power electronics wafer. Additionally, MACOM says that the higher power density of GaN on Si, compared with LDMOS, yields four times more die per wafer, making the die cost of GaN cheaper than LDMOS for the same output power.
Puts and Takes
How much of the BTS PA market GaN captures will depend upon that price-performance balance between GaN and LDMOS. Assembly and test cost and capacity will be important, as well as the supplier's understanding of the application and the support they provide to their customers.
I think it's improbable that the LDMOS incumbents will lose this battle, assuming they embrace GaN and use it where it provides differentiated performance. Their longstanding assembly and test capabilities and knowledge of the BTS market give them an advantage.
However, if MACOM's arguments prove true, GaN on Si will be disruptive and a threat to LDMOS. MACOM's challenge will be adding the assembly and test capability and application intimacy that Freescale and NXP do so well.
Cree will avoid having to create a back end by continuing to serve the market as a foundry. Their challenge, though, will be providing attractive wafer pricing that supports stacked margins (theirs and the PA manufacturer's), while generating margins that meet the profitability expectations of their soon-to-be IPO investors.
Qorvo may face the steepest hill, executing a business model to provide an assembled and tested GaN PA, with applications support, directly to the BTS manufacturer. Like MACOM, they need a sophisticated and low cost assembly and test capability as well as strong applications support. Unlike MACOM, they don't have a disruptive cost strategy. Their products must strike the right balance between price and performance to win designs, generate production wafer volume and gain market momentum.
Watching MACOM and Qorvo try to unseat Freescale and NXP, while negating each other, promises to be more interesting than a reality show. Seeing how Cree, Sumitomo and Infineon — the #3 LDMOS player — fare will add interest to this complex story.
Keeping Track of the Players
GaN Going Mainstream Webinar
You can get a firsthand sense of the GaN capabilities and strategies of Freescale, MACOM and Qorvo at the GaN Going Mainstream webinar on June 16 at 11:00 am EDT. It will be moderated by industry veteran Ray Pengelly, now retired from Cree. Register to attend or view the replay.
Updated June 12 to add the link to the Qorvo GaN PA article.