3G power amps to use revised CMOS

Wiberg: MASMOS to hit cells in 2009.

Is it possible to make a low-cost CMOS power amplifier (PA) for 3G cellphones that can operate with the power efficiency and robustness of a GaAs PA? Acco Semiconductor believes it has found a way.

However, observers questioned whether the company's MASMOS technology can deliver as promised. And one analyst said the issue of CMOS PAs for 3G could be moot as more front-end modules appear that house the PA along with filtering and a CMOS controller.

CMOS has been used for GSM cellphone PAs and for the PA requirements of Bluetooth and WLANs. But for the more complex modulation schemes and higher frequencies of evolved 3G and WiMAX, GaAs or similar technology has been necessary.

"The main difference between the GaAs-based technologies and the CMOS technology available today for PAs is the breakdown voltage of the devices," said Denis Masliah, founder and chief technology officer of Acco. "In saturated amplifiers for very high-efficiency regimes, the voltage across the device can reach 12V or more for a 3V bias. And in linear amplifiers, the back-off from the saturation makes it necessary to have an equivalent voltage excursion to limit distortion at the highest power levels.

"Furthermore, it is necessary to accept impedance mismatch from the antenna on the devices. That can produce a large voltage across the device, and can destroy it if its voltage capability is too low," said Masliah. Acco's MASMOS technology, he said, can "solve this breakdown issue for all CMOS technologies available for RF today, from 180nm to 65nm. MASMOS has a gain capability equivalent to GaAs-based devices and a breakdown of over 14V routinely."

Masliah said the company is "filing a cluster of patents" on its CMOS PA concept but wouldn't say exactly what MASMOS is, or what Acco has done to improve the breakdown immunity of CMOS. He would only acknowledge that MASMOS is more than just a design or architectural tweak.

But if the new technology amounts to something physical in the process, does that make it non-standard CMOS?

"This is something we don't want to discuss," Masliah said. "It is scalable with technology from 90nm to 65nm, 45nm and so on. It is applicable to any CMOS foundry. It can be done on SOI wafers."

"Our goal is to have products in handsets in 2009," he added.

Going fabless
Acco hopes to tap a multibillion-dollar market, transforming itself into a fabless design house in the process. The company has been performing RF IC design services since 1994. "We've worked with 20 different companies--either system companies that wanted their own chip or IC companies that have outsourced overflow work" to Acco, said Masliah.

Then came what Masliah calls the "breakthrough" in the use of CMOS for the implementation of PAs. He went to Pond Venture Partners Ltd, believing the technology could transform his company from a design services operation into a fabless semiconductor house.

Pond was sufficiently impressed with Acco and its CMOS PA to join with Partech International and put up $10 million in venture capital. Half of that money was given to Masliah early in 2007 to pay for the manufacture of prototypes to prove his concept. Acco was also able to attract Jamie Urquhart, former chief operating officer at ARM and now a partner with Pond, to sign on as interim CEO.

"The technology could have application beyond the PA," Urquhart said. "But if Acco only focuses on the cellphone PA application, the market is big enough to make it worthwhile."

Stan Bruederle, research VP with Gartner Inc., doesn't dispute the size of the market but said that the case for CMOS in W-CDMA and Edge has not been made to date because of the need for a broad range of power with linear operation. "Amplification can be backed off from 10W to run at 2W, but you can still have an instantaneous peak of 10W. The result is that you are running at 10 percent or 5 percent efficiency most of the time to support the peak signals."

Meanwhile, said Bruederle, "front-end modules are becoming popular that include filtering in the package with the PA and a CMOS controller. A move to CMOS, he said, could be unnecessary "once the PA is contained in a module."

The total annual market for PAs, including WLANs and handsets, was $1.85 billion in 2006, up from $1.43 billion in 2005, according to Bruederle. It will almost certainly top $2 billion this year. But the primary source of the market expansion has been high unit shipment growth as manufacturers have battled declining average selling prices. Handsets now routinely have two or three PAs on board as Bluetooth and Wi-Fi capabilities have been added, bringing the market's unit shipments to about 2 billion devices per year.

Going to GaAs
"It's hard to displace GaAs. There's no evidence that the PA market is going to CMOS--if anything, it is going to GaAs," he said. "I haven't seen a compelling argument for CMOS amps yet, and the trend toward modules works against integration."

Communications base stations, with their much higher power requirements, tend to use PAs fabricated in LDMOS silicon or gallium nitride, observed Gerard Winpenny, chief technology officer of Nujira Ltd, which focuses on offering improved efficiency for base-station power amplification through architectural techniques. "The nature of the modulation is such that in the downlink from the base station, you end up having very poor efficiency unless you do something. The uplink has been designed to be less critical."

Winpenny argued that moving CMOS PAs beyond GSM handsets and Bluetooth into "evolved 3G" and WiMAX applications, where modulation schemes are more complex and carrier frequencies higher, would be difficult. He acknowledged that a CMOS technology that performed more like GaAs, as Acco is promising, could reduce cost in handsets.

But "claims are easy to make," Winpenny said. "The proof of the pudding will be in the eating."

- Peter Clarke
EE Times