Shifting Landscapes In The GaAs Industry
It’s tricky to accurately describe the mood at this year’s CS Mantech. Quiet optimism is probably the best phrase to capture the feeling of the delegates that made their way to Boston between 23 and 26 April. They knew that the incredibly strong, recent growth in GaAs device sales is over, but they could take comfort from the steady, slight increase in shipments that should continue for the next few years. Figures supporting these views were provided by Eric Higham from the Newton, Massachusetts’ office of Strategy Analytics. According to him, the GaAs device market leapt in value by a staggering 35 percent in 2010 to almost $5 billion. Such a large hike in global sales of GaAs chips has not been witnessed since 2000, and back then trouble followed. Revenues tumbled when the Internet bubble burst at the turn of the millennium, excess inventory went through the roof and it took until 2008 for global GaAs device revenues to return to similar levels to those of the 2000 peak.
This time round, however, boom is unlikely to be followed by bust. Healthy handset sales that account for the majority of GaAs device revenues are tipped to rise, and the content of this class of chip in the phones is predicted to increase. These conditions have led Higham and his colleagues to produce a preliminary forecast that suggests that GaAs device revenues will increase at a compound annual growth rate of 6 percent through to 2015 from a starting point of $5.2 billion in 2011.
For many years, RFMD and Skyworks have been vying for the top spot for GaAs device revenues. Previous data from Strategy Analytics suggested that Skyworks pushed ahead in 2008, and recent figures revealed by Higham in his talk indicate that this chipmaker now has a significant lead. According to him, in 2010 Skyworks had 19.1 percent of the market, compared to 16.3 percent for RFMD (see Figure 1). Fast-forward to last year and there are significant changes on this leadership board. Skyworks has surged to 22 percent, while TriQuint has overtaken RFMD, grabbing a 15.1 percent share of the market, compared to just 12.4 percent for the Greensboro headquartered outfit. During his talk, Higham also highlighted the growing trend for outsourcing epiwafer production. In 2011, this market, which is now dominated by WIN Semiconductors, grew to $600 million.
Within the handset sector, by far the biggest change has been the phenomenal growth of smartphone sales. In 2010 and in 201, shipment of these advanced handsets leapt by 68 percent, with quarter-over-quarter changes varying significantly.
Figure1.Data produced by Strategy Analytics shows that the three big US GaAs chipmakers – Skyworks, RFMD and TriQuint – had more than half of the total market in 2010, which was valued at $5.45 billion, when revenue from foundries is included.
“In 2011, it was really 80 percent growth in the first half, dropping to only 40 percent in the second half," explained Higham. Although such rapid growth is unsustainable, Higham expects double-digit percentage gains in smartphone shipments to continue through to 2014. A year after that, these sophisticated handsets will account for more than half of all mobile phone sales.
Wireless infrastructure offers another opportunity for the growth of GaAs device sales. Higham and his colleagues have tracked a rise in the deployment of micro- and pico-cells, which have transmit powers ranging from 0.25 W to 10 W. Silicon LDMOS dominates the traditional base station market, which uses transmitters with an output of 40 W to 80 W. However, at lower powers, Higham claims that GaAs devices are more attractive.
Cramming it all in
President and CEO of Skyworks, David Aldrich, offered his perspective on the handset market at CS Mantech. He agrees with Higham that the average GaAs content in handsets is increasing - a 2G ‘feature’ phone can have less than $1 of GaAs content, while a smartphone can feature more than $10 of GaAs chips. According to Aldrich, increasing uptake of the latter class of handset is driving up the total addressable market for the RF content at a rate of 15 percent per year.
This growth in the market is not fueling any convergence in standards. “Each baseband partner has a different view of what they want the front-end to look like," explained Aldrich.
Handset makers, meanwhile, want what they’ve always wanted: Products with more functions that take up smaller footprints. And they seem to have been getting their wish, according to figures presented by Aldrich and produced by Prismark Partners. The PA footprint for a 3G handset has shrunk by a compound annual rate of 13 percent since it was launched 8 years’ ago, while the size of the based band has diminished even faster – falling by 16 percent per year.
Aldrich argued that one of the products that we will see more and more in smartphones is the multi-band, multi-mode PA. Skyworks’ has developed one of these that can work with up to 17 bands and support new features, such as envelope tracking. This technology, which can boost battery life, involves the application of dynamic adjustments to the voltage supplied to the final RF stage of a power transistor, so that it tracks the signal envelope. Battery power savings of 200 mW to 500 mW are expected, plus lower heat dissipation and improved 3G/4G coverage per base station.
Taiwanese foundry WIN Semiconductors detailed two new processes at CS Mantech: One for making rugged InGaP/GaAs HBTs for multi-mode, multi-band PAs, and another for producing Bi-HEMTs that feature a low on-resistance pHEMT and are suitable for LTE applications.
The multi-mode, multi-band PAs that WIN is developing will target next-generation handsets and tablets, which can support various different wireless standards. They will work with the higher data rates promised by 3G WCDMA/HSPA and 4G LTE, while also being compatible with existing 2G GSM and 2.5 G GPRS/EDGE standards. In addition, they will feature rugged HBTs that can withstand the stresses associated with high-voltage, standing-wave-ratio mismatches.
WIN has improved the ruggedness of its HBTs by tweaking the collector design so that it combines a high off-state breakdown voltage with a significant on-state breakdown voltage. To provide better protection from moisture and vibration, SiN films have replaced thinner ones made from polyimide.
An additional change has been a switch from a conventional layout, where the metal finger surrounds the emitter mesa, to a configuration involving the removal of outer base metal fingers. This trims base-collector capacitance by shrinking the base mesa area and increasing power gain (see Figure 2).
Figure 2.WIN can increase the power gain produced by its HBT by removing the outer base metal fingers, a step that reduces base-collector capacitance and trims base mesa area. The traditional HBT architecture is on the trims base mesa area. The traditional HBT architure is on the left; the advanced, higher-power version is on the right.
Tests reveal the benefits of all these changes. Refinements to the collector did not impact power gain or power-added-efficiency, but enabled a 50 percent improvement in the collector-emitter voltage during a ruggedness evaluation (the voltage standing wave ratio was fixed at 10:1 for 360 degree all-phase rotation, and the collector-emitter voltage was increased from 3.6 V to device failure – 7.5 V for a HBT made with the new process, compared to 5 V for a device made with the standard approach). Meanwhile, the new architecture with a reduced base-finger-area led to a slightly higher gain at frequencies below 3 GHz, the spectral range used for handsets.
WIN’s latest BiHEMT process, which unites pHEMTs and HBTs, is designed for a new generation of PAs that have two states: A high power mode; plus a low power mode, the most common operating condition for the handset. The power-added-efficiency of the PA can be improved by optimizing performance in both modes.
The Taiwanese pure-play foundry’s two commercially released BiHEMT processes, which are known as H2W technologies, produce pHEMTs with a significantly higher insertion loss than stand-alone equivalents. This inferiority must be addressed in order to capture the market for two-power-state PAs – a high insertion loss can drag down the amplifier’s power-added-efficiency in low-power-mode operation, and in turn, prevent the H2W process from fulfilling the high-linearity demands in emerging standards, such as 4G and LTE. To trim insertion losses, engineers at WIN have developed a H2W process that employs the company’s latest BiHEMT structure and its HBT4 process, and uses these in combination with a signal gate recess process and an optimized pHEMT epi-structure design. Like the improved standalone HBT previously described, the new process features a thick SiN layer in the place of a thinner polyimide film, plus the opportunity to define multiple gates, which can be used for high-isolation, high-linearity switch applications.
Characteristics for the HBT part of the BiHEMT include a typical turn-on voltage of 1.27 V, a DC current gain of 130, and breakdown voltages of 28 V and 14 V between base-and-collector, and emitter-and-collector, respectively. Meanwhile, the pHEMT has an on-resistance of 0.95 Ω.mm, 50 percent less than that produced by the previous H2W process. To evaluate RF switch performance, WIN’s engineers have constructed a full periphery design featuring a single-gate, 9 µm by 125 µm device. Insertion loss was just 0.1 dB, 0.05 dB lower than that for the original H2W process.
pHEMT’s bleak future
A rump session discussing the future of pHEMTs failed to find any silver lining within a looming black cloud. Everyone at that gatering believes that switches in handsets will be increasingly made from some form of advanced silicon, rather than GaAs, and nothing can be done to stop this. Although GaAs switches offer the best levels of performance, they are more expensive, and the current trend towards higher throw switches plays into the hands of silicon-on-insulator devices.
New opportunities for pHEMTs were considered during this discussion, but none seemed tremendously promising. For example, pHEMTs could be used to make WiFi products, but they would have to compete with silicon variants – and in battles between the compounds and silicon, the latter tends to come out on top. It’s a similar story in the automotive radar market, where pHEMTs have been used in reasonable volumes to make transceivers operating at 77 GHz, but are likely to loose market share to SiGe-based chips offering higher levels of integration.
Fortunately, handset manufacturers will not stop making models that use pHEMT switches overnight, so sales from these transistors will slowly decline over several years. In many cases, these chipmakers also have HBT processes, so hopefully any revenue losses incurred can be compensated by growth in power amplifier sales. Over the next few years, growing sales of products made with this class of transistor seem a sure bet.