Positive Vibes Abound At CS Europe
It’s easy to get your head stuck down in a project and never look up. After all, the pressure to hit deadlines, meet targets and help your company to keep pace with the competition can take up all the time you have. But if you never break out from that cocoon, you will never see the bigger picture – how your efforts form part of the growing success story of the compound semiconductor industry.
This wider view was one of the hallmarks of the plenary talk given by legendary MBE pioneer Klaus Ploog to the 150 delegates that flocked to the inaugural CS Europe conference, which was held on 22 March in Frankfurt, Germany.
Keynote speaker at the inaugural CS Europe conference was MBE pioneer Klaus Ploog, who pioneered the development of GaN structures on non-polar planes
Ploog, who is a former Director of the Paul Drude Institute for Solid State Physics, kicked-off the meeting by looking back at the very beginnings of compound semiconductor technology – in 1952, the German researcher Von Welker, wrote a paper discussing these materials as promising ones for making Hall sensors. Ploog then gave a rapid review of the fundamental properties of the compounds, before moving on to first discuss some of the challenges associated with nitride light emitters and then the opportunities that III-V devices have to cut the power consumption in nextgeneration supercomputers.
Although Ploog believes that there are three big challenges facing white LEDs for solid-state lighting – how to further increase efficiency, improve colour rendering, and cut costs – he did not dwell on these issues, preferring instead to focus on a problem known as the ‘green-gap’, the very low efficiency of LEDs in the green region of the visible spectrum. He explained that the InGaN material system can yield devices with wall-plug efficiencies of above 20 percent in the blue, but at 530 nm this efficiency is only 10 percent and falling fast. The AlInGaP family of materials cannot help. Although they can be used to make red LEDs with efficiencies in excess of 40 percent, this III-V alloy switches from a direct to an indirect bandgap as its wavelength is shifted to the green, causing its emission efficiency to plummet.
According to Ploog, pushing the emission of blue LEDs to longer wavelength is the most promising approach to making a green emitter. This requires an increase in the indium content of the InGaN quantum wells, which lie at the heart of the device, from a few percent up to more than 40 percent.
Producing such a material is tough, says Ploog, for three reasons: InN suffers from thermal instability; there is a high degree of lattice mismatch between nitride layers in a long-wavelength green LED; and strong internal electric fields in an InGaN LED with a high indium content pull apart electrons and holes in the quantum well, hampering high radiative efficiency.
To overcome problems associated with the thermal instability of InN, the InGaN films can be grown at lower temperatures. However, alloy clustering and a high point defect density can then degrade material quality. It is also possible to combat the strong internal electric fields with the formation of epilayers on a different nitride plane. “We did this more than 10 years ago, but no-one was interested," quipped Ploog.
He finished his discussion of visible LEDs by highlighting the problem of droop, the decline in nitride device efficiency at high current densities. Possible explanations for droop include electron leakage or overspill, Auger recombination, polarization mismatch in the active region, defects, carrier delocalization, junction heating and a lack of hole injection. “There are numerous explanations in the literature, and [the cause of droop] is still a big debate," said Ploog. He pointed out that some designs of LED were able to reduce droop, but none had abolished it.
Ploog also touched on some of the challenges facing developers of nitride VCSELs. Higher reflectivities and lower series resistances are needed in the mirrors, along with higher hole injection efficiencies.
The latter part of Ploog’s talk focused on the role that III-Vs could have in scaling silicon CMOS. He began by pointing out that the current supercomputer champ, China’s Tianhe-1A, needs 5 MW to churn out calculations a rate of 2.57 petaflops/s (2.57 quadrillion mathematical operations per second). The US Defense Advanced Research Projects Agency has a far faster machine in mind, and has been considering what technologies would be needed by 2015 to build a supercomputer capable of a quinitillian (1018) flop/s. One alarming issue is that such a machine would consume a colossal 1.5 GW if it were built by simply scaling the Tianhe-1A. Such a power requirement – more than the output of a typical nuclear power station – is unacceptable. Fortunately, according to Ploog, the compounds have the potential to step in and help out in three different ways: lowering the operating voltage of processors, enabling novel memory and powering optical interconnects.
One of the great strengths of the III-Vs is that they can yield transistors operating at just 0.5 V. Electron mobilities are also far higher than those in silicon, but hole mobilities are relatively low, which is why germanium is the leading candidate for pMOS structures. Scaling the transistors to 10 nm so that they can maintain the march of Moore’s Law may also require the development of FinFET-like structures to combat short-channel effects, according to Ploog.
He also outlined a promising opportunity for a rather esoteric class of materials, known as the chalcogenides, that could help to improve the energy efficiency of computer memory. The state of this material family that is based on germanium, antimony and tellurium can switch between a crystalline conductive phase an amorphous resistive one under current-induced heating. This enables the development of low-voltage, non-volatile, high-density memories capable of 1012 write/erase cycles per second.
Another family of materials – InP and related alloys – could also help to spur development of more efficient supercomputers, in this case by improving data transfer rates. According Ploog, Intel has a vision for a 1 terabit/s transmitter containing 25 silicon evanescent ‘hybrid’ lasers, each emitting at a different single wavelength, coupled to 25 modulators and multiplexed into one output fiber. The key building block to this structure is the ‘hybrid’ laser that features an InP laser diode, developed by John Bowers from the University of California, Santa Barbara, bonded to a silicon waveguide. “These lasers will always have to be made out of III-Vs, because we need high efficiency and low power consumption," explained Ploog.
Klaus Ploog pointed out that optical approaches could help to improve data communication within computer hardware. Intel has been at the forefront of efforts to do this, developing InP-on-silicon hybrid lasers. In this image of a 50 Gbit/s silicon photonics link, the transmit module (left) sends laser light from the silicon chip at the center of the green board, which then travels through optical fiber to the receiver module (right), where a second silicon chip detects the data on the laser and coverts it back into an electrical signal. Credit: Intel
Osram target picoprojectors
One of the themes of Ploog’s presentation – improving light emission efficiencies of nitride devices in the green - was picked up in a talk by Marketing Engineer Alexander Bachmann from Osram Opto Semiconductors. His presentation outlined Osram’s development of high-power blue and green lasers for picoprojectors and began by referencing a report by the market analyst Yole Développement, which claimed that the light source for picoprojectors would be a $100 million market by 2015. Initially LEDs will dominate, before lasers take over in 2016.
The reasons for the switch, according to Bachmann, relate to the superiority of laser based scanning beam projection technology. This can deliver high-contrast, high definition images that are always in focus and have a vast colour gamut (200 percent NTSC) from a very efficient, tiny laser-based source with a height below 6 mm.
Bachmann detailed the laser requirements for a projected image with a brightness in excess of 10 lumen: a 90 mW red chip emitting at 635-640 nm, a 50 mW green cousin emitting at 515-530 nm, and a 40 mW blue variant emitting at 440-460 nm. All three diodes must be “kink-free" and produce a good beam profile, added Bachmann.
Alexander Bachmann from Osram Opto Semiconductors revealed that the German outfit will be launching 515- 530 nm green laser diodes with an output of at least 50 mW in 2012. Credit: Osram
He then went on to talk about the single-mode blue, 450 nm lasers Osram produces today that deliver 50 mW or more, have an efficiency of 0.9 W/A and a median lifetime of 5000 hours. According to him, nextgeneration variants will deliver 80 mW or more, median lifetime will increase to 7000 hours and wall plug efficiency will go up from 11 percent to 14 percent. Osram’s engineers have also developed 500 mW kinkfree monomode blue lasers in the lab, revealed Bachmann, with good beam quality and a peak wallplug efficiency of more than 20 percent.
He concluded his presentation by discussing green lasers sources, which are less mature than their blue counterparts. Today, says Bachmann, the most common sources emitting at around 530 nm are either diode-pumped solid-state lasers or second-harmonic generation lasers, which are both relatively complex contraptions employing an infra-red source and a frequency doubling mechanism. Direct green lasers are sought after, because they are smaller and potentially cheaper.
Osram has been developing a green laser for several years, and Bachmann revealed some of the latest lab results: 522 nm lasers with 80 mW continuous-wave output and a wall-plug efficiency of 5-6 percent. Commercial production of 515-530 nm lasers with an output of at least 50 mW is set for 2012: “I’m confident that the price can reach $5-10 per piece," said Bachmann in response to a price inquiry from a delegate.
GaAs and GaN for wireless
Talks outlining the future for compound semiconductor technologies in the wireless sector were given by several speakers, including TriQuint’s Senior Director of Corporate Advanced Technology Development, Otto Berger, and Jeffrey Shealy, Vice-President of RFMD’s Mobile Products Group (Defence and Power).
Berger began with some jaw-dropping statistics: smartphones will account for more than 50 percent of the handset market in 2013; and over the next four years mobile data transfer will increase at a compound annual growth rate of 92 percent, leading to an exchange of 75 million terabytes of data in 2015. To cope with all this traffic, 3G and 4G networks will use complex modulation schemes to transmit and receive information over more than 40 frequency bands, including former TV bands and those currently used for GSM. The changes will have implications on the front-end technology in handsets, which must support multiple communication systems, including Bluetooth and WLAN, in parallel. Only power amplifiers with excellent linearity and efficiency will be suitable, says Berger, which will have to operate over multiple bands.
According to him, a typical 3G handset currently combines a quad-band GSM / EDGE and 4 band WCDMA. In future, handset makers will be looking for products that are cheaper, take up less space, and offer improved performance alongside higher efficiency at good linearity. Berger believes that the first step down this road will involve either the repackaging of two die into one module, or the monolithic integration of PA die. A “merged" power amplifier providing multimode and multiband operation on a single die will follow that, and even further ahead new technologies will be introduced to boost performance, such as envelope tracking. Berger is adamant that GaAs will continue to be the material of choice for the handset, thanks to its combination of strong RF performance, low-noise, high-voltage capability. “Silicon offers low cost and low performance," says Berger, “and it will not enter the smartphone market."
In Berger’s opinion, GaAs is also an attractive option in base-stations: compared to the incumbent silicon LDMOS technology, a TriQuint TriPower lineup with Doherty driver uses 15 percent less energy. Cutting energy consumption in basestations can have a big impact environmental impact, because network infrastructure accounts for 71 percent of the carbon dioxide emissions produced by the mobile industry, according to the GSMA Green Manifesto. Berger touched on the GaN effort at TriQuint, highlighting the company’s leading role in the DARPA Wide Bandgap Semiconductor program and its involvement in the Nitride Electronics next generation Technology (NEXT) project that is run by the same organization. The latter program is targeting development of high dynamic range logic operating at 300-500 GHz and based on E/Dmode GaN.
The role of power devices made from this wide bandgap material in commercial and defence applications was the focus of Shealy’s talk. He began by listing the various markets where GaN can play a role: military communications; electronic warfare; military and civilian radar; cable TV line amplifiers; digital video broadcast; cellular base stations; and industrial, scientific and medical markets. According to him, GaN is attractive in all these applications, thanks to its ability to deliver high powers at high frequencies.
Shealy said that RFMD’s GaN-on-SiC products combine a wide bandwidth with high efficiency and high power. According to him, tremendous bandwidth stems from high impendence and enables multi-band radios; a low capacitance holds the key to high efficiency circuit techniques that enables a smaller heatsink; and a high breakdown voltage supports the high power density that unlocks the door to production of high power amplifiers. “We tell our customers, if you can do it with silicon, you should," says Shealy, who by implication knows that GaN fits the bill when great performance is the key criteria.
He explained that the AlGaN/GaN HFETs that RFMD makes have 0.5 μm gate length, feature a dual field plate technology with the gate and source connected, and have a Ti/Al/Ni/Au ohmic contact and a Ni/Au gate. Production of these transistors that are grown on 3-inch SiC substrates kicked-off in 2009 using a process technology described by the company as GaN-1. This can be used to build devices that have an operating voltage of 28-65 V, a peak efficiency of 65 percent and a breakdown of 400 V. These are suitable for electronic warfare, military communication, radar and cellular basestation infrastructure.
Commercial launch of a GaN-2 process occurred last year. This process yields devices with an operating voltage of 15-48 V and a 6dB improvement in linearity over their predecessors. According to Shealy, broadband handheld radios, cable TV and industrial, scientific and medical applications could all benefit from devices made with this process. GaN-3 is in the pipeline and promises to deliver operating voltages in the range 36-65 V, improved linearity and high peak efficiency. Shealy says that the target applications are cellular infrastructure, public mobile radio and next generation military radio.
RFMD manufacturers it GaN chips in-house in a highvolume GaAs fab, and this leads lead to lower production costs, says Shealy. He revealed that these are falling all the time, with every doubling in cumulative area shipped driving down manufacturing costs by 21 percent. Shealy also spoke about the company’s GaN wafer production plans for their 2012 financial year, which began on April 4, 2011. Die manufacture should hit 2.3 million for the fiscal year, with monthly wafer starts increasing by 39 percent during that timeframe. RF amplifiers and switches will account for about 45 percent of production, cable TV line amplifiers another 30-35 percent and the remainder is expected to be devoted to foundry services.
According to RFMD’s own research, the total addressable market for GaN RF power devices in 2011 is about $1 billion. Cellular basestations account for 49 percent of this, with the remainder split between radar (22 percent), military communication (9 percent), industrial (7 percent), broadcast (5 percent), electronic warfare (4 percent), and security (4 percent). Shealy revealed that the company’s estimate of the value of the served available market for GaN high power amplifiers is $300 million, with four fifths of that in the defence sector.
Shealey rounded off his talk by detailing arguments for the inclusion of GaN in various applications. For example, he believes that employing these wide bandgap devices in point-to-point backhaul radio simplifies the power-combining network, thanks to the high operating voltages that lead to high impedances, and it also cuts the number of components required – switching from GaAs to GaN cuts the number of amplifiers that are needed for a 10 W output at 6-9 GHz from three to two. Another example that he gave was the remote radio-head output transmitter for 4G and LTE networks, which requires a high-power amplifier. GaN amplification leads to small, lightweight units that have wide bandwidth, a peak efficiency of 65 percent and can operate at high temperatures, reducing cooling requirements. Shealey’s abiding message, like many of the other speakers at the first CS Europe conference, was that the III-Vs are already playing a key role in many applications and many more opportunities lie in store over the coming years. Grab these, and sales will rise, propelling the industry into an even healthier state and giving us all an even greater incentive to lift our heads and see our collective progress.
© 2011 Angel Business Communications. Permission required.