Positive Vibes For 2035
In 1995, the year Compound Semiconductor made its debut, our industry was far smaller than it is today. Back then the white LED and the GaAs transistor were yet to uncover their first killer application, and it would be more than five years before the first wide bandgap devices hit the market.
Fast-forward to today and it is clear that the last 20 years have been a tremendous period for compound semiconductors, with annual chip sales rocketing and now worth billions of dollars. And in this twentieth anniversary issue, many pages of this publication are devoted to recounting these successes and the reasons behind them. But will this continue, and where could the compound semiconductor industry be in 20 year's time?
Looking ahead to 2035 is a task that analysts tend to avoid, as predicting revenues for the mid 2020s and beyond involves too much uncertainty. But as Gordon Moore showed in his seminal paper from 1965, progressed can be predicted. There is no doubt that compound semiconductor chips will continue make strides in their bang-per-buck. And in sectors where change is slow and new technologies have to deliver proven reliability, it is not beyond us to see that the compound semiconductor chips that are delivering great promise today will be those that are deployed in a decade or so.
To gain a more detailed view of where the compound semiconductor industry might be heading in the long-term, here we talk to three leaders from our community: Fred Schubert, an academic from Rensselaer Polytechnic Institute, who gives his opinion on the future for LEDs; Cree co-founder John Palmour, who provides his take on the opportunities for wide bandgap power electronics; and Douglas Reep, Qorvo's Senior Director of Research, Infrastructure and Defense Products, who shares his views on the future of compound semiconductor devices for wireless communications.
In the LED industry, chip production is certain to increase in the coming years, spurred by the uptake of solid-state lighting. "For LED lighting, the cost is already below any comparable technology, if we take into account the purchase price, electricity price, environmental costs, replacement costs and environmental impact," says Schubert.
Convincing the public to part with their cash, however, hinges on the retail price, as many customers are not swayed by calculations considering the return-on-investment. This hampers sales today, as solid-state 60 W equivalent bulbs are currently far more expensive than the incumbents. However, prices should fall fast. "The LED cost of the bulb is $1, we add another $1 for the power supply and another 50 cents for the bulb and the sockets and the screw-in plug, so we are at a manufacturing cost of $2.50," explains Schubert. "So a selling price of $5 is very realistic âˆ' and we are going to have that over the next few years."
Cheaper LEDs should result from improvements in manufacturing efficiency, primarily related to refinements in packaging. "˜Traditionally, 50 percent of the cost has been in the packaging field, because it has been a serial process," says Schubert, who expects costs to fall through the introduction of a parallel process involving the placing of LEDs on a foreign substrate âˆ' probably silicon âˆ' that is processed and diced up.
Improvements in LED performance are set to continue, and should lead to the availability of efficient devices that span the infrared to the ultraviolet. "I think blue LEDs will help green LEDs," says Schubert, who expects that in time there will be a reduction in the depth of the valley associated with the green gap. He is also predicting an increase in the efficiency of ultraviolet LEDs. "But I think it is going to be more difficult there, because of the high energies involved."
One area where Schubert is not expecting a major change is in the architecture of the LED: It will have a p-n junction at its heart. However, there will be refinements. "Silicon technology has come a long way, and every detail in silicon technology "“ the gate, the dielectric, the interconnect, and so on âˆ' has been improved. We will see the same thing in LEDs: p-type doping, n-type doping, quantum wells, phosphors, packaging "“ all those things will improve." And thanks to this, by 2035 devices produced in a cost-effective manner should be delivering ultra-high efficiencies, such as 90 percent or more.
The foundation for the chip is unlikely to change, according to Schubert: "I'm a big fan of GaN substrates, because they appear the better substrate, with fewer dislocations "“ but the price competition is so tough that GaN substrates cannot compete, at least for blue LEDs." It might be a different story for ultraviolet LEDs, however, which retail for more and can derive more benefit from a native platform.
Over the coming years we should expect to see increased deployment of systems that combine the selection of colour temperature with automation and colour-temperature dimming. These solid-state sources might also be used for wireless light communication. "The IEEE is working on that," says Schubert, who thinks this technology is promising, but more research is needed.
Predicting additional new applications for the LED is not easy, with Schubert warning that we should be ready for some surprises. He points out, for example, that large-scale photo-chemical reactions could be driven by LEDs.
In the long-term it is possible that the future of the LED is not actually that bright: Research at Sandia shows that it makes sense for solid-state lighting to be based on lasers, rather than LEDs, due to the lower cost-per-lumen associated with a droop-free device delivering high efficiencies at very high current densities.
Schubert believes that this work is a very interesting avenue, but one that may not be that successful. He argues that laser lighting was proposed in the 1990s, because lasers at that time were more efficient than LEDs. However, they did not have any success, due to the challenges of producing reliable devices operating at very high current densities. GaN in smartphones?
In the wireless arena, the next few years are very easy to predict: The makers of smartphones will launch ever more sophisticated models, featuring more complex components that cater for the 4G and 4G LTE standards. Switches in these mobiles will be based on silicon-on-insulator devices, while the amplifiers will probably be made from GaAs.
"We will see GaAs amplifiers there for some time, due to their ability to do good power-added efficiencies "“ it's all about battery life "“ and very good linearity characteristics," argues Reep. Looking further ahead, there will be the introduction of mobile communication technologies that support ever-increasing data rates. This will support the seemingly insatiable appetite for greater global data consumption, which will initially allow improved streaming of video to mobile devices, before underpinning the growth of the Internet-of-things.
The first step will be the introduction of 5G, which is tipped to start its roll-out by the end of this decade. "The standards are very immature and currently being worked on," says Reep. "My personal belief is that we will see some demonstrations, if not standards, start to emerge in the Ka-band." Although there is no official allocation for mobile wireless in this spectral range, Reep believes that this situation could soon change "“ and he hopes that 5G could get a further boost from a successful demonstration of this technology at the next winter Olympics, which will be held in Pyeongchang, South Korea, in early 2018.
Next-generation smartphones sporting 5G capability will also have to be compatible with many existing standards. "It is likely that we'll see the older standards disappear within a relatively modest timeframe "“ perhaps that includes GSM. But it is likely that 3G and 4G will be around during the deployment of 5G "“ so the problem with multiple channels doesn't get easier."
Reep believes that this scenario can open the door to revolutionary product development. "I think that this is where GaN will really shine for us "“ in compact, highly-efficient power amplifiers." These devices will be very attractive to the designers of smartphones, who will need to cram more radios into a fixed space, and are looking for amplifiers capable of operating at high power densities.
If the makers of GaN transistors are to grab this opportunity, they will have to develop new devices operating at lower voltages. "The devices that we have been working on today for infrastructure or defence are high-voltage devices, and operate between 28 V and 48 V
quite typically. The handheld needs 3 V, maybe a little less, up to 5 V, so we will need to see the emergence of a different class of GaN device that operates efficiently at low voltage."
Carrier aggregation is viewed as another approach to handling higher data rates. This involves transmission over multiple channels, which can be managed with a software-defined radio.
"If we have the ability to, under software control, select the channel to support, then we can certainly simplify some of the hardware implementation within the phone," says Reep.
However, this approach is not easy, with those trying to develop software-defined radio for use by first responders having struggled for some time to find a solution.
To reach data rates that are beyond those that can be realised by 5G, one must go to even higher frequencies.
One option is V-band. "It's in a bad spot from an atmospheric attenuation standpoint, but if I only want to communicate around my conference table, and not interfere with the folks in the conference room next door, then V-band can have some attributes," says Reep.
Head to even higher frequencies and the antenna needs to be pointed to the source. However, that inconvenience may not be a show-stopper, according to Reep: "I believe some of the researchers in Germany are convinced that for the data rates we need perhaps 300 GHz is the range for some of these short-hall communications."
"At that point I think that the competition will be light-based communication, because we will be in the sub-terahertz, as we would be taking a real beating on the atmospheric attenuation of the signal." The optical sources that could be used for this are either lasers or LEDs.
Underpinning the migration to faster data rates for smartphones will be the deployment of base stations operating at higher frequencies "“ and this spells good news for GaN. But this might not be the only use for this wide bandgap device in infrastructure, as it could also be used to provide connectivity for cars, trucks and other vehicles. As they move, links could be handed off between multiple access points, allowing adaptive traffic control. "If that vision occurs fairly rapidly, we'll have a proliferation of multiple platforms, and from an access standpoint, huge bandwidth demands," says Reep.
One area where he is predicting stability is in the material combination used for making GaN transistors for communication infrastructure and defense applications. He points out that GaN-on-silicon for high-power devices is not as appealing as it may seem initially, because the substrate is relatively expensive, due to the need for a high-resistivity foundation. "If one buys a GaN-on-silicon epiwafer from one of the current suppliers, it will cost more than a GaN-on-SiC wafer that we have in production today."
Although that may change with economies of scale, GaN-on-silicon may still fail to deliver significant success. Silicon has a significantly inferior thermal conductivity to SiC, and to prevent an increase in junction temperature, the devices must be spread further apart, which impairs cost-competitiveness.
Where GaN-on-silicon might have a role to play is in the power amplifiers of 5G phones. "We don't have the luxury of large current, because of the battery "“ and so in that arena, if we see a low-voltage device emerge, it will likely be a GaN-on-silicon technology."
GaN-on-silicon is already establishing itself in another market, by making an impact in power electronics. "The performance has always been there," says Palmour, who believes that progress has been held back by uncertainty over how to use this material effectively. Engineers have spent time deliberating on whether the GaN transistor should either be a normally-on, cascaded device, or a normally-off device; there are still concerns over reliability; and widespread adoption of a repeatable manufacturing processes is still needed. But all these areas are being addressed, and GaN transistors are destined to be increasingly deployed in products requiring power devices operating between 30 V and 600 V.
At higher voltages, it is SiC that will displace the silicon incumbent, according to Palmour. These devices will become increasingly attractive as on-resistance falls, because this trims costs and increases switching speeds.
Palmour does not expect the voltage ranges where GaN and SiC products serve to shift with time, reasoning that although GaN-on-silicon can move to much larger wafers to cut costs, so can SiC, which could easily move to 8-inch lines. And although GaN-on-silicon is cheaper at the wafer level, that does not mean that it is superior when the cost of high-voltage devices is taken into account. "If we can get three times more amps per wafer than you do with GaN-on-silicon, it's a win," argues Palmour. "We're a GaN company, we do lots of transistors in GaN "“ we just don't think it's the best way for the high-voltage market."
Even more promising than GaN and SiC are devices made from ultra-wide-bandgap materials, such as AlN, Ga2O3 and diamond. However, unlocking the potential is challenging, because as the bandgap gets wider, it get harder to identify shallow dopants that enable acceptable on-resistances at normal operating temperatures. Another issue is that even for GaN and SiC, it is the package that limits the device performance. Areas to address include improving heat extraction and the current-handling capability of the wire bonds.
It is worth noting that there is still a considerable opportunity to improve the performance of today's GaN and SiC power devices. "Most of the designs that are around today are designed around the capabilities of silicon," argues Palmour, pointing out that the multi-level technologies widely used today create big, bulky, complex devices that have been constructed to get around the inherent slowness of the silicon IGBT.
"If you multi-level a bunch of lower-voltage IGBTs where you would have used a single level of a higher voltage, you get faster and simpler performance," explains Palmour, adding that this trade-off does not exist with the SiC MOSFET, due to its unipolar nature. "I think there are a whole lot of different circuit topologies that can start being explored that did not make sense for silicon and will make sense for wide bandgap devices."
Spurring investment in these new designs will be the increasing sales of wide bandgap devices. SiC chips are already being used in switched-mode power supplies, solar inverters, high-frequency power supplies and LED lighting power supplies, and they will see increasing deployment in uninterruptable power supplies and motor drives.
In the next decade, the big breakthrough for SiC, and possibly GaN too, will be its incorporation into electric vehicles. The relative sales of these two classes of wide bandgap devices will depend on the bus-voltage used in the vehicle. "There is also a lot of play as to whether it is an all-electric vehicle, a plug-in electric, or a hybrid," says Palmour. "But it's a pretty good bet that you'll see wide bandgap [devices] in electric vehicles past 2020, and obviously it would be a huge new market."
By then, SiC devices operating at very high voltages should have an opportunity to win sales in electrical grid infrastructure. "It is a long path to actually get it to market, because you have to convince a power company that this is a good thing to do "“ and that is going to take many years," says Palmour.
But he believes that this could happen in the 2020s, beginning with the deployment of SiC MOSFETs and Schottky barrier diodes with 10 kV breakdown voltages.
Wind power is another potential market. Extracting power generated out of the tower requires a tremendous amount of copper, but, according to Palmour, this can be reduced by going to higher voltages into the nacelle, a move that is possible with SiC devices. To win adoption will require proving the reliability of the diodes and transistors, because any wind turbine downtime impacts its generating costs.
While silicon will not go away, by 2035 we can expect to see far greater adoption of wide bandgap devices than we do today, with devices deployed in electric cars, various forms of renewable energy systems and myriad power supplies. Compound semiconductors will also be serving mankind in many other ways "“ they will also be lighting our world and enabling communication at breath-taking speeds.