Technical Insight
Dissecting prospects for CS industry
Analysts at CS International claimed that 5G will be good for GaAs, opportunities in the infrared can diversify LED sales, and space applications are dominating shipments of multi-junction cells
RICHARD STEVENSON REPORTS
Our industry is united, but incredibly diverse. We share the processes for making chips, which involve the selection of the substrate, the growth and characterisation of epiwafers, and subsequent processing to create compound semiconductor devices. Yet the operation of the device varies tremendously, enabling our industry to sell products to a multitude of different sectors, from security and communications, to medicine, aerospace and consumer electronics.
This diversity of sales is so great that it prevents anyone from having an intimate knowledge of our entire industry. So, if you want to know the current state of our industry, and its prospects, you want to hear from a range of experts in their fields.
One way to do this is to head to CS International, where respected market analysts and industry veterans offer their take on the opportunities for compound semiconductors. At this year's meeting, held in early March and co-located with the inaugural PIC International, insights were given into: the impact that 5G will have on the sales of GaAs chips; the opportunities for infra-red LED sales in smartphones and smartwatches; the dramatic decline in multi-junction cell shipments for concentrating photovoltaics; and the most likely architectures for incorporating III-V channels in next-generation logic.
5G will deliver a rafter of performance improvements over 4G, making it better suited to serving mobile phone communication, the internet-of-things, and machine-to-machine communication.
5G: Good for GaAs
Speaking to delegates at CS International, Eric Higham, Director of Advanced Semiconductor Applications at Strategy Analytics, argued that the emergence of a 5G network should spur GaAs revenue to new highs. Higham began by explaining the great need for these new, superior networks, before stating the capability of GaAs devices for serving this application.
A major driver for the deployment of 5G is that it can accommodate the rocketing growth in mobile data traffic. Higham illustrated this point by referring to a study by Cisco Systems that projects that between 2009 and 2019, mobile data traffic will grow at a compound annual growth rate of 75 percent. That equates to a staggering 266-fold increase over that time frame. Cisco has also provided projections for the next ten years from now. During that time the increase in mobile data traffic will slow, but still undergo a 40-fold hike.
Higham pointed out that while this growth in traffic is one of the justifications for 5G, there are also other benefits related to its roll out. He summed this up by describing the 5G vision as: "Unlimited access to information and the ability to share data anywhere, anytime, by anyone and anything." Extreme mobile broadband will be combined with massive, incredibly reliable machine communication.
Goals that must be met for delivering extreme mobile broadband would include a peak data rate exceeding 10 Gbit/s, and coverage at 100 Mbit/s everywhere. Meanwhile, the other two pillars of the 5G vision − which could be summed up by the phrases "instant action" and "for everything", and relate to tactile internet and the internet-of things − would involve machine-to-machine communication.
In some cases, these lines would be critical, requiring a latency of less than a millisecond, and ultra reliability; and in other cases, it would be massive machine communication, involving tens or hundreds of incredibly low-cost devices with a 10-year battery life.
A key technology for delivering the high data rates of 5G is MIMO "“ multiple-in, multiple-out. It increases the capacity of a link by using multiple transmit and receive antennas. Higham highlighted the progress with this technology, including a 64 element array developed by a team from Bell Labs and Rice University that operates at 28 GHz, boosts capacity by a factor of 10, and is 2.5 feet by 2 feet in size. "It's large, but you could picture it being part of a base station."
Also discussed was a 64-element array block operating at 60 GHz. It featured 22 mm by 22 mm chips produced with the TowerJazz SiGe process. Higham questioned whether these devices were too big for making a phased array for a handset.
MIMO technology has already been used for 5G trials. In a high-rise complex, Nokia has demonstrated data transmission exceeding 2 Gbit/s at 70 GHz, while Samsung topped 2.5 Gbit/s from a vehicle travelling at 60 km/hour. Even more impressive results have come from trials at Ericson, where data rates exceeded 10 Gbit/s, and at Fujitsu, where researchers delivered more than 11 Gbit/s using mobile devices as a form of base station.
Higham views 5G as a huge revenue opportunity for the GaAs industry, because it plays into the strengths of this III-V − namely a high-frequency response, good linearity, and excellent efficiency. In 2015 global GaAs revenue stood just above $7 billion, according to figures from Strategy Analytics, with sales dwarfed by estimates for the compound semiconductor industry and the total semiconductor market, which are in the region of $57 billion and $336 million, respectively.
However, Higham admitted that there are factors that could put a damper on his unabashed enthusiasm. Commercial deployment is slated for 2020, and will probably be timed to coincide with a major global sporting event, such as the Olympic Games held in Tokyo. Showcasing this technology could aid its deployment, but if there is teething trouble, or slipped schedules, the uptake of 5G could suffer.
Higham reinforced the uncertainty over the roll-out of 5G by referring to a paper from Rethink Wireless and Lewis Insight. A survey of mobile network operators revealed that 76 percent of them currently have no plans for 5G deployment. The primary reason for this is the lack of a clear return-on-investment.
The worst that could happen, according to Higham, is some form of "4G on steroids". But in his view, even that would swell GaAs revenue.
Frank Dimroth from Fraunhofer ISE showed that in order for III-V-on-silicon to compete with silicon flat plate, there would need to a be a 100-fold fall in the costs of epitaxy, substrate preparation and waste treatment.
Figures shown by Frank Dimroth from Fraunhofer ISE reveal that the CPV industry is struggling, with deployments plummeting from over 100 MW in 2012 to below 20 MW in 2015.LEDs: Seeking higher margins
In the LED chip industry, the largest applications are visible LEDs for backlighting and general illumination. Here, competition is incredibly fierce, and margins are squeezed. So several chipmakers − including leading players Osram, Nichia and Epistar − are trying to generate healthy revenues with devices emitting in the infrared.
At CS International, Pierrick Boulay, Technology and Market Analyst for LEDs, OLEDs and Lighting Systems at Yole Développement, discussed the leading markets and the emerging applications that these infrared LEDs can serve.
In 2014, according to Boulay, the security camera provided the biggest market for the infrared LED. In this class of camera, infrared LEDs are positioned around the rim of the lens, illuminating the scene that is imaged by the camera.
Right now, consumer electronics is accounting for most of the growth in the infrared LED market. "Smart watches use infrared LED technology to measure biometric values, such as the heart rate," explained Boulay. According to him, infrared LEDs are also used in proximity sensors in cell phones, enabling the detection of the presence of the human ear. Armed with this technology, the touch screen is disabled when making a call.
Looking further ahead, Boulay believes that infrared LED sales will climb via deployment in the automotive sector. It could be used in systems to measure driver alertness via observations of the eye, and gesture recognition; but the most significant sales are tipped to come from night vision and LIDAR, which is a remote sensing technology that is the optical cousin of radar.
There are two common spectral ranges for automotive night vision systems. They can be passive, operate in the far infrared at typically 8 μm to 14 μm, and be based on microbolometers; and they can be active, using an LED or another source emitting at around 800 nm, to illuminate objects that are detected by a CMOS sensor. The latter offers superior resolution, but is not so good at detecting pedestrians and only works up to around 200 m, while the passive system can identify objects up to 500 m away, and is more pricey. Both types of infrared system have been available since 2000, but retail prices exceeding $2000 are hampering costs.
Boulay believes prices could fall, and the imaging systems become more attractive. He notes that in 2014, Autoliv starting shipping a night vision system that combined near infrared and far infrared capabilities, and was offered as an extra on the Mercedes S class. This option is currently $2,260, but prices could fall fast, driving up adoption.
The Yole analyst also briefly outlined a range of other applications for the infrared LED: in medicine, where sources emitting at 850 nm or 940 nm could be used for dermatology and neurology; in the military, where these emitters are used as tail lights that are only seen with night-vision goggles and devices; in smartphones and tablets, where they equip these devices with the capability to act as a universal remote control; and as a measure against movie pirating, with illumination behind the screen degrading the quality of illegal video recordings.
With so many opportunities for infrared LEDs, it is not surprising that more chipmakers are starting to manufacture these devices. However, Boulay warns that the production of an infrared LED is markedly different to that of a white-emitting sibling, so it is far from trivial to enter a market that has been led by the likes of Osram, Nichia and Epistar for at least five years. This challenge will be easier to overcome for the bigger players, which Boulay expects to penetrate this market in the short-to-medium term.
Yole Développement analyst Pierrick Boulay expects the market for infra-red LEDs to grow, as they are increasing used in smartphones and smart watches, and for night vision systems in cars.
The collapse of CPV
Delegates were offered a view of the prospects of multi-junction solar cells from the academic Frank Dimroth from Fraunhofer ISE. He made a compelling case that multi-junction cells will continue to succeed in the aerospace market, but will struggle to make an impact in the terrestrial sector.
To show how difficult it will be for the CPV market to grow, Dimroth highlighted the selling points of the main rival, silicon technology. Its price is now very low "“ by 2015, a square silicon wafer with sides of 156 mm cost just $1.48. The market is also incredibly well established, with global shipments last year totalling 51 GW, equating to a mind-boggling 10 billion 156 mm-square wafers.
"The failure of CPV, to a good extent, is correlated to [silicon] overcapacity that was not foreseeable," remarked Dimroth. He illustrated this point by showing data from Lux Research that showed that in 2012, 2013 and 2014, excess capacity was around 104 percent, 85 percent and 50 percent, respectively (note that the last two figures were estimates), and during that time, the price of modules plummeted, destroying the fledging CPV industry. Installations of high-concentration CPV peaked in 2012, when they totalled more than 100 MW, fell to just half that figure by 2014, and last year were below 20 MW.
During his talk, Dimroth also compared the size of last year's space and terrestrial markets for multi-junction cells. He estimates that global shipments for space applications totalled about 200,000 4-inch wafers per annum, compared to just 15,000 for CPV.
At first glance, it might appear that the solution for enabling III-Vs to make an impact on the terrestrial market is to deposit these materials on silicon substrates. Dimroth pointed out the performance benefits that can result, citing work from a research group in the US, involving Steven Ringel from The Ohio State University. That team fabricated a dual-junction cell with an efficiency of more than 30 percent.
It featured a 5 μm-thick, III-V epitaxial stack.
This efficiency is 12 percentage points higher than that for silicon. However, this increase in efficiency currently comes with an unacceptable hike in cost. Polishing a 156 mm-square wafer to make it suitable for epitaxy would cost €5-10, while depositing a 5 μm film of compound semiconductor materials would add €50-80, and the waste treatment of the III-Vs would command another €10. The harsh reality is that unless the cost for epitaxy, substrate preparation and waste treatment can fall 100-fold, this technology will fail to compete with conventional silicon.
Program manager for Logic at imec, Nadine Collaert, highlighted defect densities and suitable gate dielectrics as two of the major challenges facing developers of III-V on silicon transistors.
The logic of III-Vs
There are applications where it is not a battle directly between silicon and the compounds. Instead, the challenge is to get these materials to work really well together. That could well be the case in the silicon microprocessor industry. ICs made from silicon transistors are tipped to continue to the 7 nm node, but the insertion of higher mobility materials in the channels "“ most likely germanium for the pFETs and InGaAs for the nFETs "“ may be needed to maintain the reductions in power consumption per transistor that are associated with scaling.
Discussing the pros and cons of different transistor architectures incorporating non-silicon channels was imec program manager for logic, Nadine Collaert, who began by discussing some of the trends outlined by Higham, such as the impact of machine learning and the internet-of-things on requirements for ICs. These wish lists vary, with ICs for data centres operating with wired connections and needing to deliver an increased performance at a constant power density; while smart mobile devices and sensors will have to operate wirelessly, with the former providing increased performance at constant leakage and the latter operating at ultra-low power.
Introducing III-Vs and silicon will "need compatibility with 300 mm silicon", argued Collaert, adding that this included the use of silicon-based processes and toolsets.
She went on to outline the fin replacement process that has been pursued by imec. Growing high-quality III-Vs on silicon is challenging, because the 8 percent lattice mismatch between these materials spawns numerous defects. They can be reduced substantially via growth of the compounds in a V-shaped trench, as many of the defects annihilate at the sidewalls. However, this cannot prevent lines of defects propagating along the channels.
Another major challenge for the developers of III-V transistors is finding a suitable material for the gate dielectric. Unlike silicon, there is no suitable native oxide, and many alternative materials lead to a high density of interface traps, which are to blame for reliability issues.
Recently, engineers at imec have switched from a finFET architecture to a gate-all-around device that improves electrostatic control of the channel. This move has proved a great success: transconductance, which reflects how quickly charge carriers move in the channel, has leapt from 500 μA/μm to more than 2000 μA/μm; and the sub-threshold swing, which is related to the switching capability of the device, has fallen from around 190 mV/decade to 110 mV/decade.
The latest improvements to the gate stack may lead to even lower values for the sub-threshold swing. However, with any form of traditional FET, the sub-threshold swing cannot extend below 60 mV/decade. If lower values are needed, the tunnel FET must be considered, according to Collaert, who explained that with this type of device, low currents can be an issue "“ but can be increased by replacing silicon with lower bandgap materials, such as InGaAs. imec is active in this area, having managed to increase currents by replacing In0.53Ga0.47As with In0.70Ga0.30As (in the next issue, researchers from imec will detail their successes).
There is clearly far more work to do before III-Vs can make an impact in the silicon microprocessor industry. But they are playing a key role in a wide variety of industries, as shown by speakers at CS International, and it will be interesting to see where these compound semiconductor chips are being deployed towards the end of this decade and beyond.
