Semiconductor processes have their differences. Some provide high power densities; some excel in integration of different functions. Understanding the differences is helpful not just to those choosing foundry services, but for anyone trying to understand the capabilities of different integrated circuits (ICs).

More than three decades ago, a point of debate in RF/microwave semiconductors had to do with whether not only if a fledgling technology such as gallium arsenide (GaAs) could improve upon vacuum-tube devices, but whether GaAs was even an improvement upon silicon bipolar technology. As with many new high-frequency semiconductor technologies, GaAs owed its early development to military funding, but gained a boost from commercial applications, such as television-receive-only (TVRO) satellite systems in the 1980s and cellular/wireless technology in the 1990s.

Similarly, military investments fostered the early growth of newer highfrequency process technologies, such as gallium nitride (GaN) and silicon carbide (SiC). Not to be forgotten in those years of development were similar defense-related investments in indium phosphide (InP) semiconductor processes, particularly for millimeter-wave applications. Along the way, commercial silicon foundries, encouraged by the pressing needs of computers and telecommunications for faster processing speeds and data rates, have quietly eased silicon CMOS technologies well into the millimeter-wave range. With so many process choices, it is easy to wonder exactly how they differ and why one would choose a particular process.

GaAs metal-epitaxial-semiconductor field-effect transistors (MESFETs) have been the mainstay of both low-noise and large signal solid-state RF and microwave designs for well over 30 years. The semi-insulating material offers a dielectric constant (12.9) that is compatible with microstrip structures in microwave-integrated-circuit (MIC) designs and supports operating frequencies into the millimeter-wave region. The initial process technology has been extended over the years with heterojunction-bipolar-transistor (HBT) and high-electron-mobilitytransistor (HEMT) configurations that take advantage of the material's basic electron mobility characteristics for improved gain at higher frequencies. But, in spite of impressive investments in time and capital, GaAs process technologies still fall short in terms of output power per device compared, for example, to vacuum-tube electronics. As a result, much recent commercial and military funding has supported substrate materials such as GaN and SiC with potentially higher power densities. In recent cases, the two materials have been combined (GaN deposited on SiC) to achieve high-voltage devices capable of high output-power densities with good linearity.

Before low-cost silicon semiconductor processing is written off, it should be noted that much recent development work has been in scaling down the features of silicon CMOS and BiCMOS processes to achieve cutoff frequencies beyond 100 GHz. Also, the use of laterally diffused metal-oxide-semiconductor (LDMOS) silicon processes has resulted in robust transistors capable of outputpower levels rivaling the best reported results for GaN and SiC devices. In addition, processes such as silicon Bi- CMOS readily support the integration of RF, analog, digital, and mixed-signal functions on a single die.

Unfortunately, silicon has poor insulating properties compared to GaAs, so that transmission lines tend to suffer high losses at high frequencies. Even in silicon germanium (SiGe) processes, which have been used to fabricate HBTs for many wireless applications, the performance of these devices is limited in terms of power and noise figure compared to GaAs devices (albeit at lower cost). And the cost of extending the high-frequency performance of SiGe processing has been a reduction in breakdown voltage and, consequently, output power (as demonstrated by IBM's SiGe process enhancements over the last few years). The number of silicon foundries is large worldwide, and includes such firms as austriamicrosystems, IHP Microelectronics (SiGe, ), JAZZ Semiconductor (RF CMOS, SiGe), and Taiwan Semiconductor Manufacturing Company Ltd. (TSMC).

On silicon carbide, however, some impressive high-power devices have been fabricated, notably by companies such as Microsemi and Cree for pulsed radar applications. Unfortunately, the wafer sizes are relatively small (to about 3 in. diameter) compared to traditional silicon wafers, resulting in higher costs per device than traditional silicon CMOS or bipolar processing.

GaN processes and their variants, such as GaN deposited on SiC wafers, have great appeal for high-power device fabrication because of the high breakdown voltages of GaN processes (100 V and more). Numerous studies on GaN devices have shown greater highfrequency potential than SiC HBTs and GaAs MESFETs because of the material's excellent electron transport properties and short transit times, implying higher gain at higher frequencies than GaAs or SiC. In addition, GaN-based materials have shown great potential for high-performance small-signal devices, such as low-noise amplifiers (LNAs), indicating that a single GaN foundry could support both power and low-noise device fabrication. A further benefit of a GaN-based LNA is its potential to withstand higher input power levels than GaAs-based LNAs. A number of foundries currently offer GaN and GaN/ SiC processes, including Cree, Nitronex, TriQuint Semiconductor , and RF Micro Devices, like TriQuint, long a supplier of GaAs foundry services. Additional GaAs foundries include WIN Semiconductors, United Monolithic Semiconductors, OMMIC, and Northrop Grumman.

In fact, military interest is so great in GaN on SiC that the US Department of Defense (DoD) has issued an online grant offer, number BAA-04-08-PKM-CALL-17. Sponsored by the Air Force Research Lab (AFRL), the grant aims to establish a domestic open-foundry merchant supplier of GaN on 100-mm SiC substrates.