SAN FRANCISCO — The International Solid State Circuits Conference (ISSCC) renewed a running debate over the suitability of CMOS for radio frequency circuits in a heavily attended evening panel session Monday (Feb. 5) called "Ten Years of RF-CMOS, but How Many Products Today?"
While the conclusion was that there are a few CMOS RF "products" today, usually where RF specifications are suitably relaxed to allow CMOS to operate, the main debate was between those who foresee CMOS eventually taking over RF applications for economic reasons and the bipolar advocates who see reasons why that cannot happen in the near future.
Previous years' panel sessions on the topic involved panelists arguing fervently before bemused audiences, but this year the panelists, some of the same as in previous years, were more restrained and audience members — when they got their chance on the microphones — were more animated.
Academics and younger protagonists largely comprised the RF-CMOS camp, while the older, industry speakers tended to argue pragmatically that CMOS versus bipolar was not important compared with efficient system partition and design.
Michiel Steyaert, professor at the Catholic University of Leuven (Leuven, Belgium) pointed to the increasing number of ISSCC paper submissions touting integration in CMOS. But RF designers such as Josef Fenk, manager of the wireless product group at Infineon Technologies AG (Munich, Germany) and Frank Op't Eynde of Alcatel (Zavantem, Belgium) saw CMOS as co-existing with other technologies such as bipolar, GaAs, BiCMOS, SiGe and GaAs/InP (indium phosphide) long into the future.
Digital CMOS is regarded as something of a "savior" because of its low cost and manufacturing ease, said Josef Fenk. But that cost is based on a vanilla CMOS — "CMOS with not too many components" — and does not include the cost associated with a stumble to market that might accompany a departure from the bipolar norm. A technology like BiCMOS offers significant performance improvements in terms of noise and is not much more expensive than CMOS, adding only four or five mask steps, even as the base technology scaled from 0.35 to 0.12 micron, Alcatel's Op't Eynde pointed out.
Barry Gilbert, manager of Analog Devices' Northwest Labs (Beaverton, Ore.), bemoaned the rise of CMOS for high-frequency and RF circuits but said it seemed to be happening despite the superior capabilities of bipolar process technologies for controlling many analog circuit parameters.
Call for multi-die packaging
But at a second visit to the podium, Gilbert argued for the idea of multi-die packaging, with the implementation of circuits in the most appropriate process technologies. "Why risk millions of CMOS transistors for a couple of dinky analog components?" he asked.
Gilbert referred to designers' infatuation with CMOS as the "technology du jour." It was a "tacky process," forcing front-end designers to cope with its non-linearities, he said. "I just don't get it: we're trying to be good engineers. [Manufacturing] process should be beside the point," he lamented.
Gilbert confessed to sadness at the passing of bipolar. "The technology du jour is some kind of CMOS. Can we shrug off all other technologies? Well I suppose if you are a start-up or a university professor and must use TSMC for everything — then you have to go CMOS."
But Steyaert, the university professor on the panel, argued that the development of RF-CMOS is exactly on course and is gaining acceptance in industry.
In previous years, papers had appeared from academic authors, Steyaert said, but at this year's ISSCC there are papers on the topic from authors at commercial organizations. He jokingly argued that a future panel could debate the topic, "Ten Years of RF-CMOS, Where Are the Bipolars?"
Infineon's Fenk hit back by saying, "A CMOS [low-noise amplifier] or [voltage-controlled oscillator] is not a product."
"The reason for the hype around RF-CMOS is the vision of the single-chip software radio. All of that additional effort and risk in the integration of analog, digital, RF and RAM has to pay off in cost and market share," Fenk said, implying that it would rarely do so.
He went further, saying multimode, multiband, multifunction communications terminals would have to cover 800-MHz to 6-GHz carrier frequencies with multiple antennas, switches, low-noise amplifiers and power amplifiers. He argued that for most systems it would continue to make sense to partition between RF in bipolar and digital baseband in CMOS.
Sven Mattison of Ericsson Mobile Communications AB (Lund, Sweden), introduced as the father of the Bluetooth short-range wireless network specification, spoke of the cultural pressure toward CMOS but also of its limitations. "RF-CMOS: Industry won't touch it and says it can't be done, so researchers say, 'let's show them,' "
Audience response
The response of the large audience, often aggressively partisan in favor of CMOS, marked a major difference between this debate and an ISSCC panel of two years ago on the prospects for a single-chip cell phone.
Designer Tyson Tuttle of Silicon Laboratories (Austin, Texas) claimed his company had engineered a precision voltage-controlled oscillator (VCO) and frequency synthesizer in CMOS, and attacked the panelists for their timidity with this process. He even criticized Ericsson for the excessive number of components in its cell phones. Alcatel's Op't Eynde countered that a VCO/synthesizer was not an LSI part (like a single-chip radio) but in the category of "building blocks" — equivalent to ISSCC paper submissions.
"I regard Bluetooth as a victory for CMOS," another audience member commented. "It's no good arguing that the specs have been relaxed to allow CMOS implementation. That's what we'll do elsewhere, we'll change the specifications to allow CMOS."
The final word fell to an audience member, Asid Abidi, a professor at University of California, Los Angeles, and a frequent presenter at ISSCC and wireless conferences. He argued that the RF chip market has been driven by the cellular phone with its classical partition between analog in bipolar and digital in CMOS.
He said that now, "short links" such as Bluetooth, 802.11 and Home networking are becoming consumers of RF ICs, and this is fostering an interest in alternative processes such as RF CMOS, he concluded.
Abidi said, "It's only when we get a shift in the portfolio that you may get a competitive reason for a change. That's why we haven't seen many (RF-CMOS) products and it's why we soon will."
