Opportunities still exist for GaAs industry to prosper
In his talk entitled "What s hot and what s not", Entwistle identified automotive radar as a exciting future market for GaAs. Other opportunities exist in wireless local-area networks, according to Earl Lum, executive director of CIBC World Markets. "The best opportunities are for PA and switch sockets in the 802.11a band at 5 GHz," said Lum. "Standards are becoming fixed, and neither CMOS nor SiGe perform well to the specifications." He added that almost all chipset vendors use GaAs PAs, although pricing pressures are very intense. The market for digital radios in millimeter-wave communications suffered a serious setback in 2001/2, although there are signs of a possible recovery towards the second half of this year. Emerging millimeter-wave markets are one area where companies can succeed on performance, while cost is becoming a much more important factor for established markets such as wireless handsets.
Entwistle also said that InP won t directly compete with GaAs in mainstream markets. During the bubble, lots of companies said they were about to launch InP devices, and a great deal of R&D work was carried out. Now the markets have been pushed back three or four years. "InP looks likely to follow a similar path to GaAs," said Entwistle. "It will take 10 years to become an overnight success." Market numbersAccording to Entwistle, the total GaAs IC market (including MMICs, digital ICs and discretes) declined by about 6% in 2002 to a level of $2.72 billion, of which around 81% was merchant sales and the rest was for internal consumption. While the MMIC market showed 11% growth in 2002, the market for digital GaAs ICs declined by at least 50%. Facing stiff competition from both SiGe and CMOS, GaAs has already been largely driven out of 2.5 and 10 Gbit/s fiber-optic systems. "This marks the end of GaAs in mainstream consumer markets in the digital domain," said Entwistle.
Fortunately, despite the slump in 2001, the analog semiconductor market as a whole has not been subject to the cyclical behavior seen in the overall semiconductor market, which is dominated by digital ICs. Both Entwistle and Lum broadly agreed that the analog GaAs market was likely to grow at a fairly modest annual rate of about 15% over the next few years.
Lum presented figures for the analog GaAs market (table 1), showing a decline of 29% from 2000 to 2001, followed by a further drop of 13% in 2002. As shown in figure 1, the GaAs market is continuing to shift towards modules, which accounted for 35% of the entire market in 2002. Since modules can contain a range of technologies, not just GaAs, this makes accurate counting of the GaAs market increasingly difficult.
Lum also provided a run-down of the top analog GaAs suppliers, as shown in figure 2. Since 2001, Fujitsu Quantum Devices has been overtaken at the top of the list by both RFMD and Skyworks, which together accounted for over 36% of the analog GaAs market in 2002 (figures for Skyworks show combined Conexant and Alpha GaAs revenues). Fujitsu experienced a revenue drop of 47% in 2002, exceeded only by Motorola, which saw its sales fall by 65%. TriQuint s revenue fell by more than 28%, even though the numbers include Infineon s sales. The top 10 companies, 6 from the US and 4 from Japan, together accounted for more than 80% of the market.
Entwistle made a further series of predictions about the market, one being that 4 inch GaAs for microelectronics will continue in the mainstream. This seems reasonable given the industry slowdown over the past couple of years, and the fact that two 6 inch fabs, at Vitesse and Infineon, are being closed. Also, the market for semi-insulating GaAs substrates will grow to around 23 million square inches by 2006, at least two-thirds of which will be used in epitaxial processes. Entwistle also predicted that MOCVD will overtake MBE as the dominant epitaxial technique for GaAs microelectronics.
The foundry model is still being tested in the GaAs industry, and a shake-out among the several "pure-play" foundries in Taiwan and Korea is likely. In the meantime, some of these companies are running other processes (e.g. SAW filters) instead of GaAs to boost revenue until the upturn comes. One problem with the GaAs foundry model is the lack of standards compared with the silicon CMOS world.
