Bright future awaits the compound semiconductor industry
Despite Soitec's decision to exit the concentrating photovoltaic business, deployment of this technology is tipped to rise and should spur sales of multi-junction III-V cells.
RICHARD STEVENSON REPORTS
One of the strengths of CS International is its broad coverage of our entire industry. Thanks to this, delegates attending this two-day meeting are given an accurate snapshot of the state of, and prospects for, the compound semiconductor industry.
And the good news is that the outlook is very positive "“ after hearing a range of talks from analysts on GaAs power amplifiers, III-V CMOS, LEDs, concentrating photovoltaics and wide bandgap devices, delegates would have concluded that sales of all forms of compound semiconductor chip should soar throughout this decade.
Wide bandgap devices are tipped to be deployed in electric and hybrid electric vehicles by the end of this decade.
The most upbeat talk came from Mike Corbett, Managing Partner at Linx Consulting, who argued that the CMOS industry will turn to III-V MOSFETs to maintain the march of Moore's Law. "We believe that this is likely to be used at the 7 nm node, first by Intel," explained Corbett.
This would be a radical step, but maybe the silicon industry is more accepting of change than ever before, given the recent measures to improve IC performance while shrinking the dimensions of the transistor. As Corbett pointed out, innovations that have already been used include straining the channel to increase carrier mobility, and preventing a rise in electron leakage at the 45 nm node with the introduction of a high-k dielectric and a gate-last process. The latter was a "big deal," according to Corbett, because it required three-to-four new processing steps.
Since then, the most radical move has been the switch from a planar transistor to an architecture with a fin protruding out of the plane of the substrate. Intel has pioneered this three-dimensional alternative that enables greater control over the carriers in the channel. The US chipmaker introduced the finFET at the 22 nm node in 2011, and three year's later it launched a variant at the 14 nm node. "So far, no-one else is using the finFET, but foundries are pursuing this," explained Corbett, who believes that the introduction of high-mobility materials will come next, probably beginning with a switch from silicon to either germanium or SiGe in the channel in the pFET, a step that Intel might introduce at the 10 nm node.
A number of challenges will have to be overcome to produce this device, before overcoming additional obstacles that will allow insertion of a III-V nFET at the 7 nm node. Forming high-quality materials on silicon is a tremendous challenge, but Corbett did highlight some promising approaches, including imec's growth of III-Vs and germanium in the trenches of silicon wafers.
After Intel introduces transistors sporting new channel materials, the three foundries with the highest incomes are expected to follow suit: TMSC, the world's biggest foundry, which produces 100,000 wafers per month; Global Foundries, which is buying IBM's microelectronics business; and the independent device manufacturer Samsung. According to Corbett, these are the only three foundries keeping up with the The International Technology Roadmap for Semiconductors - smaller rivals are off the pace by one or two nodes.
The new technologies employed for scaling will drive up wafer costs. Calculations by Linx suggest that the transition from planar transistors at the 20 nm node to 16 nm finFETs adds 13 percent to wafer costs, while the move to germanium and III-V FETs at the 5 nm node will add another 39 percent. Higher costs results from additional steps for chemical mechanical polishing, cleaning, MOCVD growth, and formation of the fin "“ the latter will be the largest fab processing cost, accounting for 2 percent of the total wafer cost.
Production of wafers incorporating III-Vs should begin around 2018, with Intel launching a new generation of transistors at the 7 nm node, and high-volume manufacturing of such wafers lasting for two nodes, until the mid 2020s. Foundries will follow, introducing III-V channels at the 5 nm node in 2022, with production continuing throughout that decade.
The growth of InP in trenches in a silicon wafer offers one promising route to increasing channel mobility, and ultimately maintaining the march of Moore's Law at the 7 nm node and below.
Sunny outlook for CPV
Another positive outlook for III-Vs, in this case as a material system for making cells used in concentrating photovoltaic (CPV) systems, was given in a presentation by Karl Melkonyan from IHS Technology. Although Soitec, a leading maker of multi-junction III-V CPV cells, as well as modules and systems based on them, has decided to exit this business, Melkonyan still believes that revenue for CPV will increase substantially throughout this decade.
He kicked off his talk by discussing the entire PV industry, arguing that understanding this is an essential pre-requisite to grasping the opportunities for CPV. According to him, 2014 was a tough year for PV, with many companies exiting the business. However, despite this, installations increased by 20 percent to 45.1 GW last year, and total deployment in 2015 could hit 52.7 GW.
This growth is not universal. In continental Europe cuts in incentives have contributed to a reduction in installations between 2013 and 2014, but during that time there has been increasing uptake in Asian and US markets. This is driving an increase in PV revenue from areas that are sunny and dry "“ they are said to have high levels of direct normal irradiance "“ and in such areas, CPV has its best chance of competing with other solar technologies. Here, just over 3 GW of ground-mount solar was installed in 2013, and by 2018 this is tipped to rise to almost 8 GW.
How much of this will go to CPV depends on its cost, and how this compares to other technologies. The incumbent technology, silicon, became far more competitive in 2011 and 2012, with prices plumetting to such an extent that gross margins fell to less than 5 percent. Since then this margin has increased slightly, while the cost-per-Watt has gradually declined. "Cost declines of conventional PV will continue," claimed Melkonyan, "but there will be no dramatic drop in prices." This is because there is no significant overcapacity, and margins are already very small for companies operating in this sector.
Installations of CPV hinge on winning funding from investors, with decisions based on the levelised cost of energy "“ this takes into account the cost of deploying the system and its lifetime.
On this basis, CPV already undercuts PV by about 20 percent in regions with high irradiance, and generation costs are tipped to fall from 0.1 $/kWhr in 2013 to about 0.07 $/kWhr in 2018. This has led Melkonyan to forecast a rise in all forms of CPV "“ that includes cells made from III-Vs and from silicon "“ from less than 200 MW in 2014 to more than 450 MW in 2018.
The best-case scenario, argues Melkonyan, is that CPV will outperform silicon in the levelised-cost-of-energy stakes by 30 percent. This could change the perception of investors, spurring CPV deployment to 700 MW in 2018.
Taking a liking for LEDs
Another chip that will be shipped in greater numbers throughout this decade is the LED. Its growth in lighting was highlighted by Will Rhodes from IHS Technology, while Pars Mukish from Yole Développement outlined a wide variety of opportunities for the ultraviolet LED, including water treatment and curing adhesives.
Rhodes began by reviewing the state-of-play in the light bulb market. The biggest seller of the twentieth century, the incandescent, now accounts for only one-fifth of the installed base, and this figure is falling fast as many countries outlaw this inefficient source.
"The vast majority of lamps go to consumers, so price is the number one [issue]," added Rhodes, which explains why the most common purchase today is a form of fluorescent bulb. This is much cheaper than that based on LEDs, and now accounts for more than half of the working bulbs in the world. It will continue to dominate for several years, because it is competitive, combining a low retail price with lifetimes of up to 25,000 hours and reasonable efficiency. Although the cool white light that most brands produce is not favoured in Europe and the US, it is popular in China and India.
In the long term, fluorescent bulbs face the greatest competition from those made from LEDs. Sales of these solid-state sources will continue to rise over the coming years, with installations ramping from just a few percent of all lamps today to around one-quarter by 2022. By then, about a third of all the lamp sales will be LEDs, and they will account for about two-thirds of total revenue, due to their relatively high price tag. This is declining, however, and during the last year it dropped by about 20 percent.
Driving down prices will require a reduction in the cost of the LEDs, which typically account for more than 40 percent of the bill of materials in high-quality bulbs. The good news is that total chip costs should fall by 30 percent by 2019, thanks to increasing use of mid-power LEDs, a hike in GaN-on-silicon chip production, and oversupply within this industry. By the end of the decade there will also be a considerable market for chips emitting at shorter wavelengths. "The UV LED market will increase from $90 million in 2014 to $500 million in 2019," claimed Mukish. "This market is really going."
The opportunities for the UV LED depend on emission wavelength. Devices operating in the deep UV, which is also known as the UVC region (200 nm to 280 nm), have opportunities in air, water and surface purification; while those emitting in the UVB (280 nm to 315 nm) can find deployment in medical phototherapy; and cousins operating in the UVA (315 nm to 400 nm) can be used for curing, counterfeit detection, tanning and photocatalytic purification.
At present, many of these applications use traditional UV lamps. A high proportion of these are low-pressure mercury lamps, which generated sales of $220 million in 2014, and can be used for tanning, counterfeit detection, and residential disinfection and purification. There are also high-power mercury, deuterium and xenon lamps, which netted total sales of $370 million, finding deployment in curing, analytical instruments, and municipal and industrial disinfection and purification.
Mukish made a very strong case for the superiority of UV LED sources over the traditional lamp. Switching to solid-state allows fragile mercury-containing tubes, which last for an average of 2000 hours and take up to 10 minutes to warm up, to be replaced by a more robust source that hits full power in an instant, promises to last for tens of thousands of hours and has a size of up to 1 cm2. The only area where the traditional lamp has the upper hand is efficiency: It ranges from 10 percent to 45 percent, while that of the UV chip can be just a few percent in the UVC and UVB, and up to 35 percent in the UVA.
These strengths have driven the incorporation of the UV LED into quite a few curing systems. Mukish highlighted examples of this, showcasing a total of 11 products from US, German and UK companies that had been brought to market since 2011. Mukish expects UV LEDs to be deployed in a wide variety of applications by the end of this decade: UVA sources are already adopted for nail gel curing, will be used in general lighting from 2016, and the year after that will appear in biomedical devices; UVB LEDs have just started to speed plant growth, and by the middle of this year will be used in medical phototherapy; and at around that time, UVC LEDs might start to be deployed in cell phone disinfection systems.
Many countries have outlawed the incandescent bulb, due to its low efficiency. Fluorescent sources are now popular, due to their long lifetimes, low retail price and high efficiency. LED lamps will be the long-term winner, however, with sales rising as the price of these efficient, instant-on bulbs fall.
Packaging problems
While the UV LED is still in its infancy, the compound semiconductor sector based on SiC and GaN power electronics is far more advanced. Pierric Gueguen from Yole Développement even went so far as to say that the era of wide bandgap power electronics is well underway, having been going for a couple of years.
Before detailing the successes to back up that claim, Gueguen considered the big picture, reviewing changes in the power electronics market and offering a prediction of where it is heading. Global revenue for the combination of power ICs, power modules and discretes totalled just over $10 billion in 2012 and 2013, but grew by 8 percent in 2014, and will steadily climb to around $17 billion by 2020. Although silicon dominates this market, chips made from wide bandgap materials are playing an increasing role. The SiC diode has now been available for more than a decade, and it has been joined in the last few years by various forms of transistor, including the SiC BJT, the SiC JFET, the SiC MOSFET and the GaN HEMT.
Many of these devices are already available from more than one supplier, and buyers will have even more choice in future, as chipmakers launch products now in development. The competition between these makers of wide bandgap devices will be fierce, and include many formidable firms that have established themselves as silicon chipmakers "“ that's the case for Infineon, Rohm, Panasonic, Toshiba, Hitachi, and Furukawa Electric.
The battle between these silicon giants and some specialists in SiC and GaN will fuel improvements in chip performance, but widespread adoption of wide bandgap products is not guaranteed to follow, argued Gueguen, unless packaging improves so that it can ensure reliability at high temperatures and deliver low stray inductance.
If devices can be armed with these attributes, engineers can then build products operating at higher frequencies and temperatures, and deliver benefits at the system level "“ such as lower cost, smaller size, and reduced weight "“ that can offset the research and development costs associated with introducing new forms of diodes and transistors.
Gueguen went on to discuss the packaging of power modules, revealing that the majority of failures are associated with interconnection, die attach and substrate attach. To address this, efforts are being directed at: connecting die using innovative wire-bonding approaches such as aluminium ribbon bonding, or techniques that use no wires; and attaching the parts that form the module using new materials, such as copper-tin eutectics.
Although improvements in packaging would be welcome, SiC devices are already being used in PV inverters and rail traction, and in the next couple of years they will start to be deployed in uninterruptable power supplies, industrial motor drives and wind turbines. But it will be 2018 before SiC is seen in electric vehicles. Meanwhile, for GaN, its use in DC-to-DC convertors and wireless charges is underway "“ and this material is about to be used in power supplies and PV invertors, and it should see deployment in electric vehicles from 2017. However, it is not yet clear whether chips based on GaN will be used for uninterruptable power supplies and industrial motor drives.
These developments are good news for substrate and epiwafer manufacturers. According to Gueguen, global sales of SiC substrates should grow from just over $100 million in 2015 to almost $300 million by 2020, while annual revenue for GaN-on-silicon epiwafers should rocket from $30 million for this year to $280 million by the end of the decade.
The fifth CS International, held at Sheraton Frankfurt Airport Hotel and Conference Centre on Wednesday 11th and Thursday 12th March, attracted over 300 delegates and featured more than 30 presentations, including seven from leading market analysts.
Opportunities in defence
Growth is heading in a similar direction for RF compound semiconductor chips, although increases in sales will be more modest, according to projections presented by Eric Higham and Asif Anwar from Strategy Analytics.
Higham revealed that growth of global revenue for GaAs microelectronics has been increasing at around 11 percent per year since the birth of the wireless era in 2014, and last year it might have been worth as much as $7 billion. The majority of this is associated with sales of power amplifiers, which are increasingly multi-band, multi-mode products that will face increasing competition from those based on silicon "“ a detailed report of Higham's presentation, and those by CMOS PA maker Peregrine and GaAs-based rivals Skyworks and Qorvo is provided in the feature What future for the PA, on p. 32 of this issue.
Meanwhile, Anwar considered the opportunities in the defence and military markets for compound semiconductor devices, and began by reviewing global defence spending. This is growing at just over 2 percent per year, but thanks to an emphasis on improving capability at the system level, spending will grow much faster in areas where III-Vs have a role to play, such as radar and communication.
One example of this is the growing deployment of active electronically scanned radar. This form of radar, which may contain GaN transmit modules, is being fitted in warplanes and military vessels, and is helping to increase worldwide spending on radar from just under $14 billion in 2015 to more than $18 billion in 2023. Spending on electronic warfare is expected to follow a similar trend, rising from just over $8 billion this year to almost $11 billion by 2023, and there are also opportunities for III-Vs in military communication, which is growing at more than 3 percent per year, and will be worth $35 billion in 2024.
Thanks to growth in all these areas, Anwar predicts that sales of GaAs devices, which account for about three-quarters of all III-V revenue related to defence systems, will increase at 11 percent per year through to 2019. Meanwhile, GaN, accounting for all of the remaining compound semiconductor sales over that time frame, will increase in revenue at 28 percent per annum.
Success in these markets, plus the likes of solar and mobile devices, indicates that revenue for many III-V chipmakers should rise throughout this decade and beyond. While it's hard to predict the future, it clear that it is going to be underpinned by compound semiconductor devices.
