Mantech heralds better times ahead for the GaAs industry
This year s conference theme was "New Challenges, Great Opportunities", chosen no doubt to put the events of the last year in some sort of positive light. Indeed, the talk in the technical sessions, and more informally around the booths of the 90 companies in the exhibit, was of the myriad ways that businesses are adapting to the new tough operating environment they now find themselves in. These not only include the adoption of new technologies and methods of manufacture that are designed to lower costs and improve yields, but also consolidation within the industry. For example, Conexant s wireless business has recently merged with Alpha to form a separate company called Skyworks Solutions, and Infineon is getting out of GaAs manufacturing altogether by selling its GaAs business to TriQuint (see Compound Semiconductor June 2002, p3).
Modules are essential for success
Earl Lum of CIBC World Markets gave the opening talk of the conference with a summary of CIBC s recent global market handset forecast (see Compound SemiconductorApril 2002, p63). CIBC has identified trends that will determine the leaders in the GaAs IC market as growth returns this year. These trends are the transitions from MMICs to multi-chip modules and from FETs to HBTs, and supply-chain consolidation. Some suppliers have been offering PA modules for a number of years, but the majority of PAs are still ICs. The move to modules should accelerate this year, with those suppliers who can manufacture most of the components that comprise the modules running out as the winners. These companies will also be able to offer a mix of substrate and IC technologies for the modules.
HBT technology is now firmly entrenched, and is likely to become the majority of the PA market despite the continued acceptance of enhancement-mode PHEMTs as a viable alternative. Lum predicted that further industry consolidation would be likely as GaAs IC manufacturers compete for business from a decreasing number of handset OEMs.
MESFETs still going strong
With all the talk of HBTs, PHEMTs and MHEMTs that takes place at an event like Mantech, it would be easy to dismiss the implanted GaAs MESFET as an old and inferior technology that is being phased out. To some extent this is true, and the MESFET has given way to CMOS and the GaAs HBT in many applications. But Marty Brophy of TriQuint was keen to explain that there was plenty of life yet in implanted MESFETs. Brophy discussed several areas where MESFET performance is adequate, gives a distinct performance advantage over CMOS and SiGe, and is also price competitive with these technologies. GaAs MESFETs give excellent low noise performance in the lower microwave bands, low RF losses through the semi-insulating GaAs substrate and enable the monolithic integration of high-Q capacitors and inductors.
Sub-circuits consisting of a few active devices and many passives are common in microwave receivers and transmitters. Using GaAs MESFET technology, the critical RF passives can be integrated on-chip with the active circuitry. In silicon bipolar or SiGe, the passives components require precision assembly onto a PCB, a relatively time consuming and costly process.
The MESFET s place in digital electronics has been taken by CMOS and SiGe, but significant analog niches remain in OC-48 and OC-192 circuits. Here, the low noise and high ftx breakdown-voltage product of MESFETs can be exploited by integrating a photodetector and TIA, or laser and driver, on one chip.
Following on from Brophy was E Mitani of Fujitsu Compound Semiconductors, who presented the work that the company has been doing on high-power GaAs MESFET devices for W-CDMA base stations. Fujitsu has a 240 W (53.8 dBm) output device in mass production, with a 300 W device in development. The two critical parameters for these high-power, high-linearity devices are high transconductance (gm) and low thermal resistance. A thin highly doped channel provides a high gm, improving both linearity and PAE. The gate dimensions are optimized for low thermal resistance, and the GaAs substrate is thinned to 28 µm and gold plated to a thickness of 30 µm to aid heat transfer. Operating at 2.14 GHz and a saturation power of 240 W, the device s linear gain is 11.5 and PAE is 54%.
GaAs-based HBTs are now a mainstream technology for many wireless and optical communication circuits, and HBTs with InGaP emitters are now superceding those with AlGaAs emitters. At Mantech, WIN Semiconductors and Motorola discussed the development of high-yield 6 inch InGaP HBT processes. Mariam Sadaka from Motorola s CS-1 fab in Arizona described some of the critical process steps and the challenges that the company overcame to develop its InGaP HBT process.
Sadaka described the InGaP emitter etches that were developed to overcome the problems caused by the phosphorus tail from the InGaP into the top layers. This tail gives interfaces of non-uniform thickness and composition, which are difficult to etch. Though she was giving little away, Sadaka said that a phosphoric/hydrochloric acid mixture is used to etch InGaP with different degrees of ordering and excellent selectivity to GaAs. A wet etch and single layer lift-off are used to define the base TiPtAu contact. The collector is defined by a reactive ion etch using the base metal and photoresist as an etch mask.
One of the unique features of the Motorola process is the use of airbridged interconnects designed to reduce thermal resistance and parasitic losses (figure 1). The gold airbridges are ruggedized by depositing silicon nitride over them. This means they can survive subsequent backside processing and assembly steps.
Michael Hattendorf from the University of Illinois presented the results of a collaborative project with EpiWorks, aimed at investigating the effects that lateral scaling of an InP/InGaAs SHBT s dimensions has on its performance. Theory suggests that ft should be independent of device size, while fmax should be inversely proportional to the square root of the emitter width.
The HBT structures were grown by EpiWorks using low-pressure MOCVD. The double-mesa HBTs were fabricated at the University of Illinois with emitter widths ranging from 0.35 to 3 µm and emitter lengths of 4, 5, 12 and 16 µm (figure 2). Values for ft and fmax were measured for devices at each emitter length, and at emitter widths down to 0.35 µm.
The relationship between fmax and emitter area was in accordance with theoretical predictions for small-area transistors. However, ft increased considerably with decreasing emitter width, probably as a result of reduced self-heating for smaller devices. For a device with an emitter area of 0.35 x 12 µm2, values for ft and fmax of 180 and 340 GHz were recorded. The group believes these to be the highest yet reported for an InP-based SHBT.