Armadillos and BiFETs light up Austin
As well as the armadillos (whose races looked more like harassment than anything else), there was plenty for delegates to enjoy at the conference itself. In the customary "state of the nation" plenary, delivered this year by TriQuint Semiconductor CEO Ralph Quinsey, the upbeat but watchful tone of the week was set. "Business is good," he said. "Long term, it looks like it will be even better."
Quinsey s central message was that the GaAs IC portion of the compound semiconductor business reached a watershed in 2006, with the overcapacity of recent years finally being shaken off and supply matching demand. This is a healthy situation that should help to reduce the pricing pressure from powerful chip customers like Nokia and Motorola, something Quinsey would be very happy to see.
The key to the new-found equilibrium between GaAs device makers and their customers is largely the widespread acceptance of 3G cell phones. "3G is finally here in a material way [in the form of wideband-CDMA]," said Quinsey. "Between 150 million and 170 million 3G phones will be sold in 2007."
What with W-CDMA, WiMAX and LTE (long-term cellular evolution) already in the wireless lexicon, you would think that there were already enough acronyms to go around, but Quinsey introduced yet another – UMB, also known as "ultra-mobile broadband". It s a term that the CDMA development group is using to sum up the latest "next generation" in wireless communications. Scheduled for commercial roll-out in early 2009, one of its key features – as far as the GaAs community is concerned – is the use of multiple radios (multiple input, multiple output, or "MIMO") to support mobile broadband in the home, car, office, street and everywhere else you can think of.
With GaAs having seen off the threat of silicon-based radio power components thus far, it s reasonable to think that UMB, along with LTE, WiMAX and the rest of the wireless future, will be dominated by the compound semiconductor at the power amplifier stage. The bottom line is that more radios and amplifiers will be needed to support all those frequencies and MIMO designs, which is good for GaAs chip manufacturing companies.
"Value is moving back into the compound semiconductor space," Quinsey elaborated, comparing the current period of stability with the miserable five years from 2000 onwards, where commoditization and over-supply drained any value from the industry and limited investment opportunities. That said, he still doesn t expect to see major expansion in global chip capacity from the current level of just over 800,000 6 inch wafer equivalents, which should be enough to support requirements through 2011.
The future will not necessarily be plain sailing for everybody involved in GaAs production, however, with Quinsey predicting that mid-tier players could be forced to consolidate as the major volume suppliers, such as his own company and Skyworks and RF Micro Devices (RFMD), take increasing control of volume GaAs applications. Many of the lower-tier contenders have lost market share and become increasingly sidelined (one exception being a fast-improving Anadigics). As a result, the TriQuint CEO now envisages a scenario in which the industry increasingly splits into two distinct supplier bases that require different business models.
One of these will be for relatively low-end components used in the highest-volume market of personal voice and data transmission, mostly cell phones and laptop computers. The other, which will serve a much smaller niche in terms of chip and wafer volumes, will support unique applications and strategic needs, and perhaps be embedded within much larger organizations as part of a vertically integrated business.
When Quinsey said he saw no pressing need for major capacity expansion in the coming years, it s possible that he was taking into account the effect of BiFET innovations at GaAs manufacturers, who are shrinking more complex chip functions into smaller die sizes. Anadigics and Skyworks have pioneered this technology but as the Mantech session on the subject confirmed, it is spreading fast.
After Bill Peatman had shown the extent to which Anadigics has implemented its InGaP-plus approach to BiFET fabrication, launched in 2003 and now featuring in most of its circuit designs, the Taiwanese foundry WIN Semiconductors revealed details of its own new HBT-PHEMT process.
Called H2W, short for HBT and HEMT at WIN, this approach involves an InGaP HBT grown on top of an E/D-mode PHEMT, separated by 25 μm in the device design. Using MOCVD on 6 inch GaAs, the H2W process has now passed internal qualification, said WIN s C K Lin, and has shown good early reliability data and pulse I-V characteristics.
TriQuint is also taking integration one step further. Tim Henderson from the company s Hillsboro, Oregon operation stressed that using less GaAs and a smaller die size ultimately translates to a lower total system cost for the customer. His team s approach is to combine an HBT with a D-mode PHEMT and an E-mode PHEMT that sit side by side. The difficulty with this so-called BiHEMT is that, with the HBT structure on top of the PHEMT, the PHEMT mobility is affected. TriQuint engineers have adjusted the PHEMT epi-structure accordingly (see diagram).
Although still in the development phase, with plenty of qualification to be carried out, the approach could lead to an HBT-power-amplifier circuit combined directly with E/D-mode logic and control circuitry, RF switches and low-noise amplifiers. TriQuint is now working on improving performance with the development of a 0.5&thinspμm gate process and ramping the overall scheme to volume production.
Skyworks Ravi Ramanathan echoed the main reason behind BiFET development, however, "Everyone wants better efficiency," he said. "But the bottom line is cost, cost, cost."
One firm not yet pursuing the BiFET or BiHEMT route is RFMD. While there does not seem to be any likelihood of that changing in the near future, the firm is rumored to be looking at acquiring an additional fab – possibly Avago Technologies Fort Collins site, currently used for E-PHEMT production. GaN reliability proves popular
One new technology that RFMD is actively pursuing is GaN. The wide-bandgap material featured heavily in Austin, with three entire sessions and several other scattered talks reflecting widespread efforts to further develop and commercialize it for both RFIC applications and power microelectronics.
Reliability has proved to be the critical theme in recent years and judging by the number of delegates interrogating Nitronex s Samir Singhal after his talk, it still is. Singhal discussed the qualification of the company s NRF1 process, which is gathering steam for volume production of amplifiers used in WiFi and WiMAX infrastructure.
Toshihide Kikkawa, voted author of the best paper at last year s Mantech, updated delegates on Fujitsu s GaN HEMTs, focusing on the high efficiency that will be required to achieve the 100 Mb/s transfer rates expected for 4G wireless technologies at reasonable power consumption. At high power levels, gate leakage currents rapidly reduce this efficiency. Fujitsu has addressed this by placing a SiN layer over the top of the GaN layers in its device, but this led to an unwanted drop in threshold voltage. Kikkawa suggested that this was because the SiN layer traps electrons and showed that the effect on the threshold voltage was minimized by optimizing the quality of the SiN cap material.
Fumikazu Yamaki of Eudyna Devices also talked about cutting GaN HEMT leakage currents, this time by eliminating imperfections in the substrate material. He showed a strong correlation between devices with the highest leakage and areas of high defect density in the SiC substrate.
Miro Micovic of HRL rounded off the reliability session with a look at typical failure mechanisms of GaN HEMTs. Two accelerated experiments and one real-time test identified oxidation at the AlGaN layer when the tests were performed in air, and under nitrogen an increased number of surface states between the gate and drain. Micovic suggested that hermetic packaging would significantly improve device reliability.
The GaN theme continued through a session focused on the development of electrical power devices. Wataru Saito of Toshiba began by estimating the size of the market for power electronics at $8.4 billion, of which power MOSFETs constitute about half. The trend towards higher power density is now pushing the performance limits of silicon. GaN HEMTs have great potential as an alternative, but current collapse at high voltages is a big drawback. Saito identified the cause of the collapse in Toshiba s HEMT as trapping of electrons by imperfections in its GaN layer.
Eudyna s Eizo Mitani reported a record-breaking AlGaN/GaN HEMT grown on a SiC substrate. Presenting impressive figures such as 50% efficiency and 800 W output, Mitani claimed the highest power ever reported for an S-band transistor.
Switching to SiC-based power devices, SiCED s Peter Friedrichs said that the applicability of some manufacturing processes already established for silicon gave SiC technology a head start in attaining the power equivalent to "Moore s Law" – doubling power performance every two years.
The session on novel substrates was almost entirely focused on finding better material for GaN devices. Aixtron s Bernd Schineller demonstrated the company s early-stage development of vertical HVPE equipment for GaN boule growth. He suggested that a practical off-the-shelf solution for high-quality GaN substrate growth wasn t too far away.
Diamond protagonists arrived in force, and Patrick Doering of Apollo Diamond anticipated that Apollo s 2 inch wafers would be available at the end of 2007. He predicted, perhaps boldly, that 4 inch material would be available by 2009 at a similar cost to SiC.
Silicon-on-diamond substrates offer an unparalleled level of thermal control, added Jerry Zimmer of sp3 Diamond Technologies, while Dan Francis of Group 4 Labs showed a flip-chip method of making GaN HEMTs on diamond after first growing the transistor on sapphire or SiC.
The substrate debate continued into a "rump" session. Perhaps the composition of the audience distorted the outcome of that discussion, but the overriding impression was that SiC still had a strong lead as the GaN substrate of choice – much to the chagrin of Doering and the diamond followers.
