Evidence of the exploding market for wireless communications equipment is readily available in almost any urban area - just look at the number of people talking on cellular phones, or wearing pagers. There is also a plethora of other types of wireless equipment: cordless phones, direct broadcast satellite (DBS) TV, wireless local area networks (WLANs), global positioning systems (GPS), mobile computing, etc., to name just a few. How big is the market? One research firm, Electronic Trend Publications [San Jose, CA], estimates that almost 91 million units of wireless communications & computing devices will be shipped this year, and that this figure will rise to more than 175 million units by the year 2000.

The wireless world is, at present, a jumble of competing technologies and standards, making generalizations risky. But for the most part there is agreement that some of the potentially very large wireless applications, such as WLANs and other two-way data transfer systems, will operate at high frequencies (2.4 GHz and above) that are well suited to GaAs's capabilities. For now, however, the most interesting areas are cellular and cordless telephones, which operate at frequencies <2 GHz, where Si is more competitive. The huge demand for cellular services is quickly filling up the 800-900 MHz frequency range - so the companies that provide cellular services in the US recently spent over $7 billion at the FCC Broadband Auctions for the right to use the PCS frequency band (1800-2000 MHz) for future cellular/data communications services.

Every one of the millions and millions of consumer-oriented handsets requires RF or microwave components. While there are many different types of wireless equipment, each with their own specifications for operating frequency, output power, etc., all share certain fundamental building blocks in the RF "front end" portion. It can generally be divided into four sections: 1) the receiver, which amplifies the incoming signal through a low noise amplifier (LNA); 2) the transmitter, which amplifies the outgoing signal through the power amplifier (PA); 3) the downconverter, which takes the incoming signal from the LNA, reducing it to the IF, and then passing it along to the non-RF portion of the phone, as well as the upconverter, which receives the IF from the non-RF portion of the phone, increases it back to the RF frequency, and then passes it into the PA; and 4) a transmit/receive switch to control incoming and outgoing signals. (See the Tutorial on page 42 for more information about the basics of RF front-end design.)

The RF/IF components used in today's wireless phones range from GaAs and Si MMICs to GaAs and Si discretes. Device types used include Si-BJTs and Si-CMOS, while GaAs-based devices include MESFETs, HBTs and HEMTs. At the moment, Si-based discrete devices dominate the RF/IF market, but that fact does not help explain where the market is going. Depending on the type of wireless device, its capabilities, its niche in the market place (high-tier or low-tier), and the quantity of the product being produced, each device type offers unique advantages and disadvantages. One of the more interesting contests currently being waged is not between GaAs and Si, as one might expect, but between discretes and MMICs - and some of the more sophisticated manufacturers are trending toward discrete solutions.

"GaAs has its natural place and silicon has its natural place. GaAs must compete head-to-head with Si only where their capabilities overlap," says Joe Skovron, marketing manager of TriQuint Semiconductor (Beaverton, OR), a GaAs-based MMIC fabrication house. At the moment, there is not one best set of RF components for a cell phone, and there will not be one as long as the requirements and capabilities of the phones themselves are as varied as they are in the current marketplace. But by using the guideline of finding the "natural" place for a component - matching the phone's needs with a device's capabilities - a design approach can be established. For example, performance requirements will vary depending on the type of phone (high-tier or low-tier, analog or digital). The capabilities of the phone manufacturer are also important. Smaller companies may not be able to afford large numbers of designers to optimize a discrete solution, and will therefore often opt for the MMIC solution, whereas the larger and more sophisticated manufacturers with more design resources may prefer to wring every last drop of performance out of their phones by assembling more complex solutions based on discrete devices. Of course, the availability or lack thereof of in-house discrete or MMIC capability is also important.

Each electronic device also has its own unique set of capabilities, which may or may not make it compatible with a specific phone. For example: there is widespread agreement that GaAs is a natural choice for PAs. But what kind of GaAs? Depletion mode MESFETs and HEMTs provide the best power added efficiencies (PAEs) (approaching 80% at 3V operation), making them suitable in the PAs of many high-tier phones where long talk time is needed. However, these devices require a negative power supply which adds to the cost of the phone, and in the case of HEMTs, require sophisticated epitaxial growth capability. Enhancement mode MESFETs and HEMTs do not require a negative supply, and they provide excellent PAEs for PA applications, but there are limits to their maximum output power. HBTs require no negative supply, have very competitive PAEs, and exhibit excellent linearity, but they also require sophisticated epi. And there are silicon alternatives as well. Silicon bipolars require no negative power supply and are relatively low cost, but they have inferior RF integration capabilities due to the lossy silicon substrate and have poorly-competitive PAEs (40-50%), making them ill-suited for high-tier phone applications. RF MOSFETs require no negative supply and they suffer from the same inferior RF integration as silicon bipolars, but exhibit competitive PAEs (60-70%) and are very low cost. These different characteristics make all these devices applicable to some phone applications and inappropriate for others.

Major GaAs producers such as TriQuint and Anadigics (Warren, NJ), supply a full range of GaAs-based RF MMICs for cellular phone applications. According to Jerry Miller, sales manager of Anadigics "we are currently shipping over 100,000 power amplifiers for 900 MHz applications (AWT0900-series - PAEs of 55% ) a month. What we find is that our real competition is not from Si MMICs, but from both Si and GaAs discretes. Customers add up the price of the discretes and ask if can you meet the price - if not, then they'll go for the discrete solution."

While MESFETs are the mainstay of Anadigics and TriQuint, HEMTs are being commercially produced by other companies. Hewlett Packard Communication Components Division (Newark, CA) offers an integrated LNA/mixer, as well as the LNA and mixer alone, all based on MBE-grown PHEMTs. This part requires 8 mask layers, E-beam written gates, as well as a gate recess etch. Henrick Morkner of Hewlett Packard says, "because we have been running this part in high volume for the last three years, we have established a very stable, manufacturable process - in which we are running small enough die so as to get 20,000 LNA per 3 inch wafer. Despite the complexity of the process, we can sell a PHEMT LNA in volume quantities for under $1." Morkner has found that some customers actually prefer building blocks in which the LNA and mixer are sold separately, thereby giving them more design latitude.

Jack Di Bartolo of Pacific Monolithics (Sunnyvale, CA) also finds that the best marketing strategy is not to overly integrate their GaAs-based MMICs. "We don't want to be too applications specific," he says. "If we over-integrate a device so it meets one phone manufacturer's need, then we will not be able to sell that chip to another manufacturer who has slightly different needs."

GaAs houses are not the only MMIC producers to experience stiff competition from discretes. Harris Semiconductor offers a Si-BJT-based LNA/Mixer - the HFA-3600. Using a wafer bonded process so that the entire bipolar sits on SiO_², this configuration cuts down on the high frequency parasitics which are normally present in Si-based MMICs. "Discretes are the chief competition and will probably dominate for years," says Wes Kilgore of Harris. "For the LNA/Mixer, we are competing against discretes which a good designer can put together for $0.60-0.80. The advantage that we have is that we feel the MMIC approach is more manufacturable, since you don't have to worry about matching discretes which vary from lot to lot."

Once a part meets operational specifications, all that matters is cost. When it comes to cost, it is very difficult to beat a MOSFET. Hitachi (Brisband, CA) offers a wide spectrum of two and three stage MOSFET-based power modules. These devices operate at 4.8 V, and can deliver up to 1.2 W at 50% PAE. Because of these characteristics, Paul Patterson, Senior Product Marketing Engineer, says that "we have obtained a dominant position is GSM systems." However, Patterson says that he is the first to recognize that MOSFETs do not represent "the best solution to all the customers needs." But even though MOSFETs might not meet the needs of every customer, Patterson feels that Hitachi's product line as a whole can do the job.

What companies like Hitachi, Siemens, M/A-COM, Motorola, Hewlett Packard, and Texas Instruments offer is not a single technology, i.e. GaAs MMICs, Si-bipolar, or MOSFET modules, but a full range of technologies and device types. Andreas Nitschke of Siemens (Iselin, NJ) says that "we offer several alternatives to design. Depending on the customer's approach, we can offer discretes in both GaAs and Si as well as GaAs and Si MMICs. We see a lot of 3 V requirements going to GaAs-based parts, while 5 V applications seem to go to Si. But even this is not always true, especially since we introduced our SIEGET bipolar process which can operate down to 1 V. MMICs have the advantage of requiring less design and are already input and output matched at 50 ohm, but some customers prefer to do their own designs with discretes."

For these manufacturers, the device type or material used does not really matter. All that matters is that business is growing. According to Nitschke, 'business is occurring at an unforeseen growth rate - it is booming!" And Siemens is not the only company turning in good results. RF MicroDevices (Greensboro, NC) has seen phenomenal growth over the last three years. The fabless facility, which sells only RF MMIC devices, has seen bookings climb from $1.9 million in 1993 to $9 million in 1994 to projected bookings of $18-20 million in 1995. Jerry Neal, marketing manager of RF MicroDevices says, "growth is absolutely explosive. We're experiencing sales that are more than doubling every year and we will need to double our head count from the present 50 to 100 by the end of the year in order to keep up with demand." RF MicroDevices offers a wide range of MMIC technologies, including HBT-based circuits from TRW, MESFETs from TriQuint Semiconductor and ITT, and Si-based parts from IBM. Of the 5 million ICs they project selling in 1995, 60% of their sales will be due to GaAs-based MMICs with the remaining 40% due to Si-based MMICs.

Things are also looking good in the GaAs-only camp. Like Anadigics, TriQuint is having a very good year. They are currently running more than 10,000 4" wafers per year, and are predicting sales in 1995 of $45 million, up by $15 million from 1994. TriQuint's most popular RF part, the TQ9203, an LNA/Mixer for cellular phone operation in the 800-1000 MHz range, has recently surpassed the 3 million production level point. TriQuint believes that this growth is tightly coupled to their 1.0 mm, 13 mask layer E/D MESFET process, which they think is competitive in performance with the most aggressive bipolar process (which, with trench isolation, is typically a 21 mask layer process), and just as competitive in price.

Even though Si-discretes currently dominate the RF/IF market, this trend is beginning to give way to GaAs-based MMICs. The current predominance of both discretes and low levels of integration are a direct consequence of the literally hundreds of different models of cellular phones, and the more than a dozen different operating standards which currently exist. MMIC manufacturers find that they can not over-integrate their chips, or they will become too application-specific to a given phone, which may not have the sales volumes to warrant spending the resources to develop such a chip. When a given phone model can be produced in the hundreds of thousands of unit quantities, then it can make economic sense to spend the resources on designing and fabricating a specific chip set for that phone, where the volume of production will drive down costs, allowing the MMIC approach to not only compete, but to beat the discrete approach. Evidence of this trend is seen in two cellular phones which have just entered the market place - the QCP-800 from QualComm and the PocketTM by Kyocera. Both of these phones are entering into large production runs, and both phones exclusively use GaAs-based MMICs for their entire RF section. See our next story for more information.

The GaAs industry has learned from hard experience that it should always mix some caution in with its optimism. Nevertheless, when one looks around the cellular, cordless, and general wireless fields, it is easy to believe that the days of reliance on low-volume applications are over for good.