The long wait is over as 3G technology goes global

When Japan s NTT DoCoMo launched its 3G service back in 2001, the rest of the world was expected to follow its lead before long. But the slumping telecom and semiconductor markets put paid to that idea and 3G went on the back burner.

2004 and 2005 now look like being pivotal years in the global 3G roll-out. Some critical developments have taken place among US carriers recently, with major consolidation the story of 2004. This has been followed swiftly by the award of billion-dollar contracts to accelerate the deployment of upgraded networks across the US.

ABI Research - a New York-based technology research think tank and consultancy - says that the $41 billion merger of US wireless telecom giants Cingular and AT&T Wireless (ATW) was a key development in 3G evolution. Prior to the merger, ATW had already launched 3G services across its Universal Mobile Telecommunications System (UMTS) networks in six major US cities: Dallas, Detroit, Phoenix, San Diego, San Francisco and Seattle.

Following the merger, Cingular (itself a joint venture between SBC Communications and BellSouth) said that it would use the spectrum acquired with ATW to roll out the UMTS services to carriers throughout the US. Some possible complications may arise from the frequency bands used. ATW s existing UMTS networks run at 1900 MHz, whereas the nationwide roll-out will operate at 850 MHz.

Nevertheless, Cingular says that it plans to be offering 3G services in "most" major markets by the end of 2006. "We expect to begin offering UMTS services in additional markets some time in the first half of 2005, with services available in the top 100 US markets by the end of 2006," the company told Compound Semiconductor. It expects to spend around $1 billion upgrading to both UMTS and higher-specification mobile standards in the next couple of years.

Ericsson, which will be building the Cingular network along with Lucent and Siemens, says that the existing GSM base-stations will require new RF hardware for UMTS operation. The company is currently evaluating its power amplifier (PA) technology and is considering silicon LDMOS and GaAs-based solutions.

In Europe, where the wireless networks are generally more advanced than in the US, companies such as Vodafone and Orange have put their considerable marketing muscle behind 3G services, and shops were full of the high-specification handsets over the recent holiday period. Handset maker Motorola has launched a raft of 3G products to meet the expected demand and the company is to add around 16 new wideband code division multiple-access (W-CDMA) models in 2005. Although the advanced handsets accounted for only around 5% of Motorola s phone sales in 2004, the company is expecting this percentage to increase dramatically in 2005. Meanwhile, Vodafone plans to buy a 3G network from Nokia for deployment across Australia and New Zealand. Nokia, which already has nearly 50 W-CDMA customers and 31 roll-outs in progress, also has big plans for its 3G handset portfolio, with 10 new terminals planned for launch in 2005.

III-V chipmakers stand to profit from the 3G roll-out in two ways: through continued network upgrades, and handset applications. And it won t be just RFIC manufacturers feeling the benefit. Phones that transmit video clips and allow video messaging need display screens with high-quality backlighting provided by high-brightness white LEDs.

According to Strategy Analytics, an international market research and consulting firm, the value of 3G base-station shipments will grow at a compound rate of 31% overall through 2009. But growth in the base-station PA market specifically, which is where GaAs (and potentially GaN) chips fit in, is expected to peak in 2007 as demand for GSM and EDGE infrastructure declines and average selling prices are eroded. The positive effects are already being felt by GaAs chip and RF module manufacturers, with Skyworks Solutions ramping up PA module production for W-CDMA phones, and TriQuint, Agilent and RF Micro Devices (RFMD) all launching either new PAs or RF transceivers.

Skyworks and RFMD - the market leaders in GaAs-based handset products - are cautiously optimistic about the opportunities presented by 3G technology in 2005. Steve Machuga, Skyworks vice-president of technology development, said, "In 2005, we will indeed see growth in 3G handset sales. As the 3G infrastructure continues to develop, most OEMs will offer multimode handsets to enable users to access both 3G and existing networks." According to Machuga, that means handsets will need two radios and more complex RF switches, both of which are likely to mean higher GaAs content per handset. He expects "substantially more" HBT amplifiers and PHEMT switches on the market in 2005.

And despite the range of device designs required for different wireless protocols and the need for backwards-compatible handsets with multimode radios, Machuga says that the manufacturing processes are similar, and that these more complex PAs will ultimately become solitary devices. As far as the networks are concerned, Machuga also sees increasing GaAs content as the build-out accelerates. "Some of the technical demands for 3G, such as improved linearity, will provide opportunities for higher-performance PAs, such as those based on GaAs," he said.

Over at RFMD, where the company is investing $75 million in an expansion of wafer-processing capacity, Jerry Neal and colleagues are seeing the benefits of advanced wireless communication at first hand. Neal himself has an RFMD-powered cell-phone handset that receives 32 television channels over the "2.5G" EDGE network in Greensboro, NC, while company employees use phones to transmit short video clips and high-resolution images in their everyday work.

Neal s feeling is that firms like RFMD will inevitably benefit as cell phones draw together more and more applications. "All of the [new] functions and the higher data rates are driving enhanced power in the phone," he said. This extra functionality must be handled while the power budget decreases, meaning that increasingly sophisticated PAs are required.

"We re going to a whole new level of sophistication when it comes to maintaining the efficiency at all power levels," Neal said. While he believes that demand will accelerate the use of high-performance 90 nm silicon circuitry, the RFMD co-founder says that GaAs-based PAs will remain the technology of choice for the next few years. That said, Neal thinks that the drive toward all-silicon PAs will gather momentum. "We have a lot of research and development going on in that area," he admitted. Cost is not the issue, however, as the widespread use of 6 inch wafers and the shorter processing times (compared with silicon) of GaAs devices means that the III-V technology is now very cost-competitive. "The move to integration is going to drive opportunities for silicon PAs."

On the network side, Neal says that RFMD will introduce GaN-based PAs for base-stations in 2005. "We have those amplifiers now, and the power-density performance we are getting is incredible," Neal enthused. "We re talking about a voltage breakdown in excess of 50 V, and a power density of 100 W/mm, which is just incredible."

With reliability tests in progress, Neal believes that the real technological battle in the base-station field will be between LDMOS and GaN in the final amplifier stage, with GaAs remaining in driver amplifier applications. "We think that GaN will give LDMOS a real run for its money."

Neal wouldn t confirm exactly when the GaN-based PAs would be launched, but he did indicate that RFMD had customers ready to implement the technology.

RFMD will face stiff competition from local rival Nitronex, plus the Japanese trio of Fujitsu, Matsushita and Oki Electric.

For Neal, however, one issue above all others will dominate 2005: "The thing we re addressing every day, what s really driving our business, is the issue of cost." With customers demanding a 15-20% price reduction annually, efficiencies must be driven up along with manufacturing volumes just to maintain current revenue, hence the $75 million Greensboro expansion.

China promises huge volumes, but it is one place where 3G networks have yet to take shape. Rumors abound that 2005 may see a breakthrough, with network trials in progress and local manufacturer Huawei launching its first 3G handsets. But the key development is the issue of 3G licenses, something that happened in Europe five years ago. For now, the jury is out on whether the Chinese government will issue the expected four licenses this year. If it does, 2005 ought to be remembered as the year 3G really did go global.


Box 1:GaN lasers September marks key launch date for next-gen DVD Just 10 years after Nichia researchers made the first, primitive GaN laser, high-volume manufacture of these devices is to begin this year as next-generation DVD players hit the big time. True, you can already buy a Blu-ray Disc player made by Sharp, Panasonic or Sony; but only if you re in Japan and willing to part with around $3000. That will all change in September, which is set to be a key month as it marks the launch date of NEC s HD DVD drive and (relatively) low-cost, high-volume application of the technology.

Despite Blu-ray having reached the market first with "early adopter" offerings, Toshiba s rival HD DVD group now appears to be making faster progress with lower-cost products. After NEC s September launch, look out for Hewlett-Packard s notebook computers featuring Toshiba HD DVD drives and HD DVD players from Sanyo. Blu-ray Disc Association (BDA) member Samsung will provide some competition for NEC with its $1000 player late this year.

For the compound semiconductor supply chain, this industry split is not a good thing, as globally agreed standards are always preferable for chipmakers and their suppliers seeking to hit the right manufacturing capacity. Still, the uptake of GaN lasers for whichever format that succeeds will be another shot in the arm for a sector that has propped up the optoelectronic chip industry in recent years thanks to the phenomenal impact of DVD players and recorders.

Upwards of 10 companies in the BDA could end up manufacturing GaN lasers in volume, although the bulk of the devices are likely to come from Sony and Nichia initially. The two companies have shared huge amounts of intellectual property relating to device manufacture and Sony tends to produce "cutting edge" laser diodes in-house before outsourcing "older" technology to its industry partners. Sony may also spring a surprise by employing GaN substrates for volume laser manufacture, although perhaps not immediately. While the BDA and the HD DVD Promotion Group have been falling over themselves recently to find favor at Hollywood studios, they have also been busy wooing Microsoft in a bid to get their technology implemented in the next Xbox gaming console.

Although not regarded as a key battleground, the gaming industry is an interesting market sector, as it pitches key BDA member Sony in direct conflict with Microsoft as they fight to sell their PlayStations and Xbox consoles to the world s gamers.

That ought to favor the HD DVD group, but according to some reports Toshiba s format is not the first choice among software developers, who cite the extra capacity of Blu-ray as a key advantage of the technology. Sony is certain to use Blu-ray in its PlayStation 3 console, while two major developers (Electronic Arts and Vivendi Universal Games) have now joined the BDA.

According to Nichia, its research team observed the first laser pulse generated by a GaN laser operating at room temperature in November 1995. Just 10 years on, the technology is the cornerstone of a global industry movement.

Box 2: Superfast transistors InP set to break the speed limit

High-speed transistor development targeting the terahertz regime will be a strong research theme in 2005. The technology is wanted by the US military for a variety of applications, including analog-to-digital converters for digital-radar applications. One of the leading researchers in the field is Milton Feng of the University of Illinois at Urbana-Campaign. His team recently made a 550 GHz transistor, and Feng predicts that we will see a 700 GHz transistor before the year is out.

The UIUC group is busy refining InP HBTs, structures that Feng considers best-suited to very-high-frequency operation due to a combination of their inherent speed coupled with a reasonable breakdown voltage. The advantage of such devices over competing structures can be clearly seen in the graph. "If you were going to make a terahertz transistor with SiGe you d run out of breakdown voltage," explained Feng. "With PHEMTs you would have the same problem. In our case [HBTs], we actually have [a breakdown voltage] of more than 2 V at over 550 GHz."

Feng suggests that InP HBTs will operate above 1 THz, but that SiGe HBTs and InP PHEMTs are limited to 500 and 700 GHz, respectively. He points out that bipolar transistors have better I-V characteristics, such as output conductance, than FETs: "When FETs shrink down to 20 nm, their output conductance starts leaking between the source and the drain."

The problem with current GaAs HBT technology is power consumption, which limits high-level integration. "If you make a 5000 to 10,000 transistor converter chip with GaAs, it would consume more than 10 W, and that burns the chip out," said Feng. "With InP you can increase the transistor level from MSI [medium-scale integration] to VLSI [very-large-scale integration], because each transistor consumes less power at a higher speed."

The research groups developing high-speed InP HBTs fall into two camps; those using a single-heterostructure design, such as Feng, and those fabricating double-heterostructure devices.

Box 3: Superfast transistors continued (DHBTs), such as HRL Laboratories. DHBTs fall into type I and type II categories. In type I structures the conduction band and valence band energies of the lower bandgap material both fall between the conduction and valence band energies of the higher bandgap material. However, in type II structures the conduction band of the narrower bandgap material may, for example, be positioned below the valence-band energy of the wider bandgap material.

Feng says that type I InP DHBT designs require a transition layer of 30-60 nm between the base and collector to address issues associated with bandgap discontinuity: "We choose SHBTs, because if you re going to shrink the collector to 500-600 Å, then the advantage of the DHBT disappears. Fundamentally, a type I DHBT will be slower than a SHBT." He concedes that while type II DHBTs have a great potential to provide the ultimate terahertz source, the development of such structures is still in its infancy.

Whichever type of HBT design that the researchers use, one universal goal is to reduce the emitter size, thereby increasing the base current. State-of-the-art high-speed transistors have emitter widths of around 0.25 μm, and are several microns long. Feng acknowledges that although reducing the dimensions only yields a small improvement in the current gain cut-off frequency (fT), it significantly improves fmax, the maximum oscillation frequency. fT is the more fundamentally important figure as it depends on the transport characteristics of the device, rather than its geometry.

Reducing the dimensions of the emitter also reduces power consumption. IBM is currently producing SiGe-based transistors with emitter dimensions of 0.1 x 2.5 μm: "If a three-five guy can shrink [an InP HBT] down to that scale by improving the doping concentration and doing some bandgap engineering, then I predict that the device will be close to terahertz operation," said Feng. He cites achieving sufficient doping in the base region while maintaining crystal quality, and refinement of device design to reduce emitter current density, as two key hurdles to overcome as the device dimensions are reduced.

MBE appears to be the preferred growth method used by researchers fabricating high-speed InP HBTs, which Feng puts down to the greater time taken to develop MOCVD processes providing high-quality interfaces. While type I InP structures favor MBE growth, type II structures appear better-suited to MOCVD.