The Analyst: Can Silicon Displace GaAs In Handsets?
Silicon has certainly driven GaAs out of mainstream markets before. In the 1990s, GaAs ICs for CDMA and TDMA handsets from companies such as M/A-COM, RF Micro Devices and TriQuint addressed the transmit/receive chain through on-chip integration of mixers, LNAs and gain blocks.
But, as the radio moved towards addressing multiple bands, GaAs processes were simply unable to match the more complex integration offered by silicon.
Another example was the rapid decline of GaAs in digital ICs. In 2000, Vitesse was the largest GaAs device manufacturer in the world. But when CMOS performance caught up, Vitesse s switch to silicon effectively killed off the market for digital GaAs ICs.
In handsets, GaAs became confined to the radio front-end, where it is now the incumbent technology for power amplifiers (PAs) and switches. Today, GaAs industry revenues are dictated by wireless applications, with the global market for GaAs devices – mostly MMICs – totalling $3.1 billion in 2006. But could history be repeated? Could silicon take over the handset PA and switch functions?
Integration has always been the unique selling point for silicon technologies and continues to be the primary argument used by companies looking to displace GaAs from handsets. Axiom Microdevices and Jazz Semiconductor are two silicon PA suppliers targeting cellular handsets, with Axiom claiming to be on track to ship 10 million quad-band GPRS PAs in 2007.
Certainly, its AX502 product is approved on a number of major semiconductor platforms and selling to phone manufacturers such as ZTE in China for low-cost handsets. Axiom is also looking at developing products for higher-end EDGE and wideband-CDMA platforms, and cites silicon s potential for a one-chip solution, with integrated PA, transceiver and baseband.
Jazz recently announced its "Silicon Radio" platform, again aiming for a future single-chip solution encompassing the transceiver, PA, antenna switch and power controller. In the shorter term, the company is introducing an integrated PA, power controller and antenna switch on a single die, effectively looking to displace GaAs-based discretes. Jazz also has EDGE and W-CDMA solutions in its sights.
Australia s Peregrine Semiconductor has targeted the handset switch market with its silicon-on-sapphire UltraCMOS process, which is based on superior integration and performance – if not necessarily lower cost. The key to the company s success was the initially better performance of its parts in meeting and exceeding the 3GPP IM3 specifications for W-CDMA/GSM operation. Coupled with the integration advantage, Peregrine s switches are rapidly displacing conventional silicon-based pin diodes in antenna-switch modules and have also been used in PA-switch modules, displacing GaAs-based PHEMT switches.
As new cellular standards emerge, handset manufacturers are seeking proven technology that can meet the requirements across their product range – from leading-edge to ultra-low-cost platforms. In this context, it is clear that GaAs HBT PA and PHEMT switch technologies currently meet handset OEM requirements for performance, time-to-market, packaging and even cost.
While GaAs technology performance and multimode and multiband architectures have served to increase the overall GaAs content being used in handsets over the years, chip manufacturers have also improved the integration capabilities of GaAs technology, increasing the value-add to their customers.
Anadigics BiFET process combines HBT and FET structures on one substrate and forms the basis for the company s cellular handset PA modules, which are designed to minimize current consumption while maximizing power-added efficiencies, especially in the 16 dBm regime where handsets operate for the majority of the time. Using BiFETs, Anadigics has also showcased a fully integrated 802.11 a/b/g WLAN front-end IC that combines high- and low-band PAs and LNAs, RF switches, filters and all other associated circuitry on a single die with the final product measuring 4.0 × 4.0 mm.
Skyworks BiFET process also combines a FET on a HBT substrate to offer biasing and power output on a single chip for both cellular handset and Wi-Fi front-ends. The power output stage of this process outperforms conventional GaAs HBT PAs, while the addition of the FET on the same chip displaces the requirement for separate silicon ICs used for biasing control and also offers the handset manufacturer a smaller footprint solution.
Early implementations by Anadigics and Skyworks have focused on using a HBT wafer with the FET grown on top or alongside using etch stops to separate the different devices. Other companies, such as WIN Semiconductor, are starting with a HEMT wafer, with a HBT and E/D PHEMT device optimized separately on the same wafer, again using etch stops. Adding more complex E/D PHEMT devices offers both switch and power amplification on the same die, plus the associated logic control that is typically performed by a separate silicon CMOS-based die.
Finally, E-mode PHEMTs have been demonstrated as a viable tool for achieving greater levels of integration on GaAs. Avago Technologies has developed a fully integrated front-end IC for 802.11 a/b/g that integrates high- and low-band PAs and LNAs, RF switches, filters and all other associated circuitry on a single die, with the final product occupying a 1.5 × 2.6 mm footprint.
Counting the real cost
One of the oft-repeated arguments from the silicon camp is that GaAs is an expensive, exotic technology that is merely tolerated by end-users as it currently offers the best performance.
In targeting the cellular front-end, Jazz and Axiom have focused largely on integration, but cost is also an integral part of their argument. Axiom cites $1000 for a typical 8 inch 0.13 µm CMOS foundry wafer, comparing it with a 6 inch wafer from a GaAs foundry that it claims could cost as much as $2500. Jazz cites a similar sliding scale for silicon wafers, from around $1000 for CMOS, rising to $2500 depending upon the layers of metallization and extra processes required.
SiGe-based PA foundry wafers on an 8 inch platform also cost around $2500, but the larger size offers almost double the potential device yield. So if we simply compare commercial wafer foundry costs, then clearly GaAs is at a disadvantage to both silicon and SiGe.
But that s not quite the whole story. It must be remembered that more than 95% of GaAs HBT PA and PHEMT switch manufacturing for cellular handsets is done by integrated device manufacturers (IDMs), such as RFMD, Skyworks, TriQuint or Anadigics, all of whom own in-house manufacturing facilities. We estimate that the real cost of an internally manufactured HBT PA wafer is closer to $800 – raising a serious question mark over silicon s supposed cost advantage.
Backing up this argument is evidence that GaAs technology is competing successfully against silicon CMOS in ultra-low-cost handsets. TriQuint is in volume production to China s ZTE for Vodafone s ultra-low-cost handsets, for example.
Closer analysis reveals that silicon technologies can actually be significantly more expensive than GaAs when considered for low-volume markets like infrastructure, automotive radar or satellite communications. This emerges when the costs of a mask set are taken into account. We estimate that a GaAs mask set typically costs between $25,000 and $50,000 – compared with between $50,000 and $300,000 (and, in extreme cases, more than $1 million) for silicon processes.
We therefore believe that silicon solutions can only be cost-effective for certain niche applications, typically where leading silicon IDMs can run specialized processes on their conventional high-volume manufacturing lines. These applications include millimeter-wave radios and automotive radar. The latter market is one example where this model is already being applied by M/A-COM and Infineon, and is an area where SiGe-based solutions will potentially displace GaAs in both long and short-range platforms.
Ultimately, however, GaAs technologies remain the most cost-effective solution for high-power, high-frequency applications, with demand from cellular handsets continuing to be the primary growth engine for the GaAs industry as 3G services expand.
For 2006, we estimated that 3G platforms accounted for only 13% of the handset market. However, we believe that this proportion will increase to 61% of global shipments by 2011. W-CDMA/EDGE-based handsets will account for 42% of phones shipped, with EDGE/GPRS accounting for more than 37% of handset sales in 2011.
Simply put, the bulk of unit growth will be in the higher end of the market, where complex multimode, multiband front ends will be required.
There will be a mix of approaches used in these front-ends, but we forecast that GaAs devices will continue to dominate this function and be used in a variety of products, including PA modules, PA-filter modules and PA-switch modules. Coupled with demand from other markets, such as cable-access TV, wireless base-stations, telecoms and datacoms, the overall market for GaAs devices will exceed $5 billion in 2011.
• Strategy Analytics latest reports on the wireless and GaAs sectors can be found by clicking here