Technical Insight
GaAs sales set to plateau
Price erosion and a slowing of smartphone growth will lead to a very modest increase in the GaAs market, argues Eric Higham from Strategy Analytics in an interview with Richard Stevenson
Q: Will the introduction of 5G deliver a hike in GaAs sales? And how long might it be before sales of phones with 5G technology take off?
A: I think the answer to that question is being defined as we speak. We have done a number of 5G projects this year. We have talked to a lot of equipment guys, and a lot of operators, to get a sense of what they're looking for.
I'm kind of envisioning it as a multi-variable equation that is getting solved. The solution is going to determine the technology.
One of the things that is going to be fundamental is the definition of massive MIMO. How many of these beams do you have? The equipment guys tend to talk about transmit power in terms if EIRP "“ effective isotropically radiated power. Those numbers are going to be fixed, and they'll be a number for a macro-cell, or a micro-cell. But the more antenna beams that you are going to have in your massive MIMO network, the lower each transmitter needs to be for a fixed EIRP.Â
So, if you say my definition of massive MIMO is going to be 16 antenna beams, that implies a different level of power than if it's going to be 128. Depending on what the number of antenna beams you are going to have, you can run from GaN all the way to RF CMOS to meet the power requirements.
There are lots of other factors in this multi-functional, multi-variable equation that I'm talking about. What is going to percolate to the surface sooner or later is power consumption. I heard a company talking about having done a CMOS PA at millimetre-wave frequencies "“ it might have been E-band "“ but it has 8 or 10 percent power-added efficiency out of that PA. If you think that you are going to be doing a massive number of additional base-stations, and each one has a very low power-added-efficiency, how do you find all that power?Â
The other interesting thing is that as we start to talk about these massive MIMO antennas at 28 GHz or the true millimetre-wave frequency range, antenna physical spacing starts to become an issue.Â
As you go higher in frequency, the wavelength spacing between radiators gets to be a very small physical distance. How do you handle that? Is that running lines along different dimensions? That adds loss and denigrates performance.
With GaN, because for a given power output you can make the chip smaller than other technologies, that might give that technology a leg up, particularly on the packaging standpoint. As these antennas that are half-wavelength spaced tend to be cross-polarised, you tend to have two antennae at that location "“ and you can put two GaN amplifiers in a package, and maintain that spacing at 28 GHz right up to the antenna radiating element. That might not be the case with other amplifiers, because the amplifiers physically have to be bigger. Â
The GaAs community is very excited about what is going to happen. For the middle of the range, the number of antennae result in a power level that would be pretty easy for GaAs to handle.
So that's positive, but until we get more consensus on what that architecture is going to look like, I definitely wouldn't rule out GaN, even for the lower power. Nor would I would rule out silicon for some versions, but that's probably RF CMOS.Â
A: I think the answer to that question is being defined as we speak. We have done a number of 5G projects this year. We have talked to a lot of equipment guys, and a lot of operators, to get a sense of what they're looking for.
I'm kind of envisioning it as a multi-variable equation that is getting solved. The solution is going to determine the technology.
One of the things that is going to be fundamental is the definition of massive MIMO. How many of these beams do you have? The equipment guys tend to talk about transmit power in terms if EIRP "“ effective isotropically radiated power. Those numbers are going to be fixed, and they'll be a number for a macro-cell, or a micro-cell. But the more antenna beams that you are going to have in your massive MIMO network, the lower each transmitter needs to be for a fixed EIRP.Â
So, if you say my definition of massive MIMO is going to be 16 antenna beams, that implies a different level of power than if it's going to be 128. Depending on what the number of antenna beams you are going to have, you can run from GaN all the way to RF CMOS to meet the power requirements.
There are lots of other factors in this multi-functional, multi-variable equation that I'm talking about. What is going to percolate to the surface sooner or later is power consumption. I heard a company talking about having done a CMOS PA at millimetre-wave frequencies "“ it might have been E-band "“ but it has 8 or 10 percent power-added efficiency out of that PA. If you think that you are going to be doing a massive number of additional base-stations, and each one has a very low power-added-efficiency, how do you find all that power?Â
The other interesting thing is that as we start to talk about these massive MIMO antennas at 28 GHz or the true millimetre-wave frequency range, antenna physical spacing starts to become an issue.Â
As you go higher in frequency, the wavelength spacing between radiators gets to be a very small physical distance. How do you handle that? Is that running lines along different dimensions? That adds loss and denigrates performance.
With GaN, because for a given power output you can make the chip smaller than other technologies, that might give that technology a leg up, particularly on the packaging standpoint. As these antennas that are half-wavelength spaced tend to be cross-polarised, you tend to have two antennae at that location "“ and you can put two GaN amplifiers in a package, and maintain that spacing at 28 GHz right up to the antenna radiating element. That might not be the case with other amplifiers, because the amplifiers physically have to be bigger. Â
The GaAs community is very excited about what is going to happen. For the middle of the range, the number of antennae result in a power level that would be pretty easy for GaAs to handle.
So that's positive, but until we get more consensus on what that architecture is going to look like, I definitely wouldn't rule out GaN, even for the lower power. Nor would I would rule out silicon for some versions, but that's probably RF CMOS.Â
Smartphones penetration continues to rise, but the rate of growth is in decline.
Q: Will there be much growth in the global GaAs microelectronics market over the next five years?
A: It looks like a little bit of a kick up if you look at the five-year forecast. We will have a little dip in the next couple of years, followed by an upward trending line to get to that little kick-up. But these downs and ups are a percent, or less than a percent, from year to year, so relatively flat.
The cellular market is the largest single driver by far. If you expand that a little bit and call it the wireless market, which will pull in base stations, WiFi and VSAT, then that's 80 percent of the GaAs market. So the trend for cellular, and broader scale wireless, is driving the overall market. Right now, the cellular trend is price erosion, with other technologies capturing share. Smartphones are driving the cellular market, and the big negative is that we are starting to reach a time in history where smart phone penetration is starting to definitely slow down.
On the flipside, handset quantity goes up every year, and you've still got more complexity. The US just finished its 600 MHz auction, and now handset manufacturers are going to have to figure out how to squeeze that frequency band in amongst all the others.
Q: So is the GaAs content in mobile devices still on the up?
A: The complexity is increasing, definitely. I'm sure you'll have seen the announcements about speed trials and speed results as we trend towards 5G. Operators are doing that with bigger chunks of spectrum, but also with carrier aggregation.
What does that mean for the handset? More linearity in the power amplifier, and definitely more complexity and more sophistication in that front-end. Â
What makes answering your question a little bit more challenging is that device manufacturers have done a great job at being able to do multi-mode, multi-band "“ and the number involved in the multi portion of that is increasing. If you are covering five bands or eight bands, that not five or eight PAs. It hasn't been that way for a long time, but I think that the number of distinct PAs is decreasing.
When you started to aggregate, and two bands became one band, it was never twice the price or twice the die area. It was always a little bit less than that. The phone manufacturers are not going to say we like you guys, so we're going to give you more money. So it's conflicting forces. If you look at the bill-of-materials. I think that's going up, but perhaps slowly. That's a positive "“ one of the only positives in the cellular market.
Q: Several years ago, the leading players in GaAs HBT technologies were championing their BiFET and BiHEMT technologies. Are these now in production?
A: I think they are. There were being championed as a difference maker, but the reasons for the BiFET have gotten less compelling. However, the more hybrid approach has gotten to be less expensive. If you are going to have to make a couple more wire bonds, or a couple more part placements, I don't think that's significant any more.
Q: Is there much investment in GaAs technology. For example, are features sizes shrinking?
A: In the silicon industry, everybody is so interested in Moore's Law. That node size is decreasing pretty quickly. The claim is that the GaAs industry isn't doing anything similar. I certainly would say that the GaAs industry is not changing feature sizes as quickly as the silicon industry. But we are seeing a trend towards smaller gate sizes. I know that Qorvo announced a 90 nanometre process in the last year or so, geared towards at the time optical applications. But I'm sure they ported that over to millimetre wave, so point-to-point radio and emerging 5G. I think it's not unusual to see foundries that are in the hundreds "“ 120, 130 "“ or down below 100.Â
Q: Will there be much growth in the global GaAs microelectronics market over the next five years?
A: It looks like a little bit of a kick up if you look at the five-year forecast. We will have a little dip in the next couple of years, followed by an upward trending line to get to that little kick-up. But these downs and ups are a percent, or less than a percent, from year to year, so relatively flat.
The cellular market is the largest single driver by far. If you expand that a little bit and call it the wireless market, which will pull in base stations, WiFi and VSAT, then that's 80 percent of the GaAs market. So the trend for cellular, and broader scale wireless, is driving the overall market. Right now, the cellular trend is price erosion, with other technologies capturing share. Smartphones are driving the cellular market, and the big negative is that we are starting to reach a time in history where smart phone penetration is starting to definitely slow down.
On the flipside, handset quantity goes up every year, and you've still got more complexity. The US just finished its 600 MHz auction, and now handset manufacturers are going to have to figure out how to squeeze that frequency band in amongst all the others.
Q: So is the GaAs content in mobile devices still on the up?
A: The complexity is increasing, definitely. I'm sure you'll have seen the announcements about speed trials and speed results as we trend towards 5G. Operators are doing that with bigger chunks of spectrum, but also with carrier aggregation.
What does that mean for the handset? More linearity in the power amplifier, and definitely more complexity and more sophistication in that front-end. Â
What makes answering your question a little bit more challenging is that device manufacturers have done a great job at being able to do multi-mode, multi-band "“ and the number involved in the multi portion of that is increasing. If you are covering five bands or eight bands, that not five or eight PAs. It hasn't been that way for a long time, but I think that the number of distinct PAs is decreasing.
When you started to aggregate, and two bands became one band, it was never twice the price or twice the die area. It was always a little bit less than that. The phone manufacturers are not going to say we like you guys, so we're going to give you more money. So it's conflicting forces. If you look at the bill-of-materials. I think that's going up, but perhaps slowly. That's a positive "“ one of the only positives in the cellular market.
Q: Several years ago, the leading players in GaAs HBT technologies were championing their BiFET and BiHEMT technologies. Are these now in production?
A: I think they are. There were being championed as a difference maker, but the reasons for the BiFET have gotten less compelling. However, the more hybrid approach has gotten to be less expensive. If you are going to have to make a couple more wire bonds, or a couple more part placements, I don't think that's significant any more.
Q: Is there much investment in GaAs technology. For example, are features sizes shrinking?
A: In the silicon industry, everybody is so interested in Moore's Law. That node size is decreasing pretty quickly. The claim is that the GaAs industry isn't doing anything similar. I certainly would say that the GaAs industry is not changing feature sizes as quickly as the silicon industry. But we are seeing a trend towards smaller gate sizes. I know that Qorvo announced a 90 nanometre process in the last year or so, geared towards at the time optical applications. But I'm sure they ported that over to millimetre wave, so point-to-point radio and emerging 5G. I think it's not unusual to see foundries that are in the hundreds "“ 120, 130 "“ or down below 100.Â
Q: Is most of the world's production now on 6-inch lines?
A: Yes, but I don't think it makes sense to obsolete 4-inch lines. The premise that the mobile handset market drives the GaAs market, and price erosion there is strict, means that it makes sense to go to 6-inch to get to that cost reduction. One of the foundries told me, probably a couple of years ago, that they were seeing somewhere in the 10-15 percent per quarter die shrink, so that's cost saving. If it gets smaller, it gets cheaper. I think companies that have 4-inch are selectively using that 4-inch for some of the other 50 percent of the market that is not handset. This may not have that same negative slope on the price erosion.
Q: The number of makers of GaAs parts seems to be declining. Anadigics is no more, and TriQuint and RFMD have merged to form Qorvo. Does this entity and Skyworks have a stranglehold on the market?
A: Qorvo and Skyworks account for almost 60 percent of the market. And if you throw in Broadcom and WIN Semiconductor, you are almost up to 75 percent. The GaAs market is bifurcating into the top two heavyweights that are mobile device driven, and everybody else. Everybody else has a pretty wide range of customers "“ there are 50, or 60 or 70 companies that do something in the GaAs market. Not everyone has their own foundries "“ a lot of those guys are buying chips and building higher level assemblies. But it's a big universe of companies that will produce devices that have GaAs devices in them.
Q: The capabilities of Qorvo and Skyworks seems to have increased, with products that contain several different technologies. The days of thinking of them as simply GaAs HBT makersare long gone, aren't they?
A: Very true. Just look at the top-line revenues that both of them report, a smaller and smaller portion of that is GaAs these days.
If you look back in history, Skyworks got left out when Nokia was the biggest handset manufacturer, and Nokia was tied in tightly with RFMD. So Skyworks looked for other manufacturers, other markets and other technologies to remain in that position they had. History has moved forward, and throwing your rod in with Nokia for smartphones turned out not to be the best decision in the world. Â Some of the other guys that Skyworks looked at, such as Samsung, got to be much more important.
Skyworks has also done a great job of diversifying technology and end markets. They have got involved with the internet-of-things, smart metering "“ what you might not think of as mainstream markets if you think of Skyworks as strictly a GaAs company. That's in their DNA to find other markets for their technology. Pretty early on they acquired SiGe Microsystems, and had SiGe technology for a while. With Qorvo, with Green Peak, their IoT acquisition, those guys can now do everything on silicon. Now Qorvo can go to shows where you might not think a GaAs company would be. With both of those guys, they have really done a good job in having a technology portfolio. Part of the challenge has been getting comfortable with that outsource model, where they don't have the RF CMOS and the SiGe foundry under their roof.Â
Q: Has the merger of RFMD and TriQuint led to the cost savings that were hoped for?
A: Yes, but I don't think it has materialised quite the way they thought it was going to. We all know that synergy is another word for layoffs, and I think that when that $150 million synergy was first announced, most people looked it and thought that's got to be foundry. They would have thought that they've got duplication of foundry, too much capacity, and something is going to happen with that "“ but it hasn't.Â
The reason for that is that they have done a really good job in repositioning the foundry, and they have been helped by a big uptick in the filter market. I think that Qorvo, definitely more than Skyworks "“ but Skyworks to a little extent "“ have branched off into filters, and things that aren't active parts anymore.
For five or six years, growth of Skyworks' GaAs revenue has outpaced the overall market. That allowed them to increase their lead over Triquint and RFMD and then Qorvo. This last year, Skyworks, for the first time in my memory, underperformed the GaAs market growth, which wasn't much. Qorvo seems to have got things squared away and closed the gap on Skyworks a little. They are coming out with new products that are addressing new product segments, and the financial results seem to be forming in a positive direction. So we can assume that they are ready to move forward, and are not spinning their wheels. Â
Q: In general, are profit margins tight for suppliers of components for handsets?
A: Yes. However, even though price erosion has been pretty constant in that market, I think they've gotten better by using technology. They've been able to make the multi-mode, multi-band PAs incorporate more bands. So they retain a large portion of that overall revenue by creating a better mouse trap, if you will, to address the problem. They don't just have to compete on whether everything will be cheaper than the guys down the street. I think Qorvo has benefited from a strong filter capability, from the module standpoint. Perhaps they can capture more revenue by supplying a fully integrated module with a number of filters and PAs, rather than just selling them individually.
For 5G, the number of antenna employed for the MIMO architecture will determine the power requirements, and thus the suitable technologies.
Q: The GaAs HBT is under threat from silicon technologies, such as SOI. How strong is this challenge?
A: I think that there have been some interesting developments. About three-to-four years ago Qualcomm announced the RF360 platform that had CMOS PAs and was going to target the latest release of LTE. Shares for WIN, TriQuint, RFMD and Skyworks from before the Qualcomm announcement to the Qualcomm announcement underwent a step function. But as we see now, Qualcomm has just entered into an agreement with one of the Japanese companies to do PAs in GaAs for their higher-frequency, higher-performance offerings.Â
At the time, we were saying if anyone can get their foot in the door with CMOS PAs, it will be Qualcomm. They have a tremendous amount of expertise to throw at the problem, they've got the name. Having said that, for the lower feature phones, entry-level 3G and the 2.5G that still exists, there is a lot of CMOS PA content. It's just that the drive until recently was to have the latest and greatest smartphone. I don't believe CMOS has gotten much traction in those next-generation, higher-frequency 4G, 4.5G phones, but it is definitely taking market share away from GaAs in the lower side of the market. The good news there is a lot of people are being transitioned, and people are turning off 2 G networks and going to higher data-rate capability. Â Â
Q: In non-mobile GaAs technologies, is there a place for the little guy?
A: If you exclude WiFi, markets are all fairly small. Here you can get some better gross margins than the high volume markets. They end up being the targets of the ecosystem of 50 or 60 companies who aren't Qorvo or WIN or MACOM or Skyworks. None of these guys are big, or have a portfolio that addresses every single market, but they have all found niches. I think what keeps the little guy going is finding those niches where they can do a value added, performance-driven solution, and not have to worry about squeezing that last fraction of a penny out of the cost of the price.
For point-to-point radio, test-and-measurement, and other markets, you can almost count volumes on your hand, and you can get good gross margins. Hittite is the classic example. Those guys minted that 70 percent gross margin quarter after quarter. Â They would not set foot in the high-volume market, because there is nowhere to get the margins there. I think that's what keeps the second tier, in terms of revenue, going. They find that niche, and they do a good job of that niche.
Q: The GaAs HBT is under threat from silicon technologies, such as SOI. How strong is this challenge?
A: I think that there have been some interesting developments. About three-to-four years ago Qualcomm announced the RF360 platform that had CMOS PAs and was going to target the latest release of LTE. Shares for WIN, TriQuint, RFMD and Skyworks from before the Qualcomm announcement to the Qualcomm announcement underwent a step function. But as we see now, Qualcomm has just entered into an agreement with one of the Japanese companies to do PAs in GaAs for their higher-frequency, higher-performance offerings.Â
At the time, we were saying if anyone can get their foot in the door with CMOS PAs, it will be Qualcomm. They have a tremendous amount of expertise to throw at the problem, they've got the name. Having said that, for the lower feature phones, entry-level 3G and the 2.5G that still exists, there is a lot of CMOS PA content. It's just that the drive until recently was to have the latest and greatest smartphone. I don't believe CMOS has gotten much traction in those next-generation, higher-frequency 4G, 4.5G phones, but it is definitely taking market share away from GaAs in the lower side of the market. The good news there is a lot of people are being transitioned, and people are turning off 2 G networks and going to higher data-rate capability. Â Â
Q: In non-mobile GaAs technologies, is there a place for the little guy?
A: If you exclude WiFi, markets are all fairly small. Here you can get some better gross margins than the high volume markets. They end up being the targets of the ecosystem of 50 or 60 companies who aren't Qorvo or WIN or MACOM or Skyworks. None of these guys are big, or have a portfolio that addresses every single market, but they have all found niches. I think what keeps the little guy going is finding those niches where they can do a value added, performance-driven solution, and not have to worry about squeezing that last fraction of a penny out of the cost of the price.
For point-to-point radio, test-and-measurement, and other markets, you can almost count volumes on your hand, and you can get good gross margins. Hittite is the classic example. Those guys minted that 70 percent gross margin quarter after quarter. Â They would not set foot in the high-volume market, because there is nowhere to get the margins there. I think that's what keeps the second tier, in terms of revenue, going. They find that niche, and they do a good job of that niche.
