Conexant Systems Inc. will take on IBM Corp. in silicon germanium (SiGe) process technology, in competing papers to be presented during the coming months. Conexant's paper will present current and future processes that company officials claim are better than IBM's efforts, including IBM's groundbreaking transistor announced in June.
IBM's transistor, a working device with transit frequency (Ft) of 210 GHz - enabling clock speeds of 100 GHz - could be ready for products as early as 2003, according to officials from the communications research center of IBM Microelectronics. in Yorktown Heights, N.Y. But Conexant (Newport Beach, Calif.) officials told EE Times that those devices will be handicapped by the more critical Fmax parameter, and that their own SiGe process, while producing more modest numbers, has a more realistic shot at the high end.
Both companies are scheduled to present details of their current-generation SiGe processes at the IEEE's Bipolar/BiCMOS Circuits and Technology Meeting beginning Oct. 1 in Minneapolis. Later, both plan to present their future SiGe plans at the International Electron Devices Meeting in Washington during the first week of December.
Neither session is likely to determine which process is faster, however. IBM, Conexant and other high-end SiGe players such as Hitachi Ltd. have a variety of specifications they can quote, according to Bernard Meyerson, vice president of IBM's communications research.
"There are so many subtleties in the business. That's why I refuse to get into the numbers," Meyerson said.
IBM's latest SiGe advancement came in June when the company claimed researchers had created transistors with Ft of 210 GHz in a 0.18-micron bipolar SiGe process. The devices were the result of a SiGe process that is the likely follow-up to the company's 7HP process.
But Ft tells an incomplete story, said Theodore Zhu, executive director of Conexant's communications foundry services. Ft reflects the highest frequency at which a transistor can show a gain in current; at Ft exactly, the current gain is 1, meaning the transistor can't be switched any faster. But according to Zhu, a more accurate measure of a process' worth is Fmax, the highest frequency for a power gain - power equaling current multiplied by voltage.
Zhu's contention is that IBM's process has run out of steam on the Fmax side. "Our understanding is that Fmax is only 70 [GHz]," Zhu said. Conexant, by contrast, is preparing a 0.18-micron BiCMOS SiGe process that will have Ft of 160 GHz and Fmax of 150 GHz, he said.
IBM fellow Bernard Meyerson disputed those numbers, noting that IBM's current SiGe processes have Fmax exceeding 100 GHz. "They have to be kidding. Why would I go back?" he said. But Zhu insists that he knows IBM's SiGe process is stuck in pushing Fmax higher.
It's an inconclusive fight, because sources agreed that like most parameters, Ft and Fmax can tell incomplete stories.
"Ft is used because Ft is used. It's just a standard number people quote," said Bill Krenik, advanced architecture manager at Texas Instruments Inc. "Many in the RF world complain it's not such a great number," but all parameters have their pros and cons when used as benchmarks, he said.
Zhu said it's easy to build one device with extremely high Ft. "It's basically transient time for electrons to move from a transmitter to a remote collector," making low Ft possible through the use of thin epitaxy layers, Zhu said. But Fmax reflects more system-level conditions such as base resistance, base capacitance and system noise, making it a better measure of a process' speed, he said.
But Meyerson noted that Fmax is a wildly variable number that isn't measured in any standard way. Fmax is heavily influenced by parasitics surrounding the device, and testing conditions often can be manipulated to change Fmax by a factor of 2, he said. "It's not like Ft where you can measure and cite a number," he said.
It's for that reason that IBM Micro rarely quotes Fmax and does so in a "phenomenally conservative" way, Meyerson said. "I've been hosed so many times by so many people I thought were reputable, I just gave up."
Meyerson recalled a case years ago where a device was demonstrated to have Fmax of 160 GHz. Most attendees at that conference believed it - including Meyerson himself, who was chairman - but it turned out the results had been pumped up by the use of an 8-volt bias across the transistor, as opposed to the 1 V others had been using.
Fred Zieber, semiconductor analyst and president of Pathfinder Research Inc. (San Jose, Calif.), said both sides are correct about the shortcomings of Ft and Fmax as benchmarks. But it's important to keep in mind that a 40-Gbit/second device doesn't need Ft or Fmax in the 200 range, he said. "It really doesn't matter if you're doing 40-GHz stuff. You've got plenty of bandwidth."
Zieber also noted that SiGe's benefits go beyond raw speed. "You can do stuff at higher speeds with far less power, and that also means far less noise," he said.
Zieber said he believes that IBM's 210-GHz SiGe is stronger than Zhu infers. "IBM doesn't announce something until it's there or almost there. They've made a fair number of ICs with this process already," he said.
"One of the reasons IBM did this announcement is to show its customers it had a road map. You've got people talking about indium phosphide, gallium arsenide, other stuff - and a lot of this stuff at these frequencies is just smoke at the time," Zieber said. "Customers want to know if there's a next step."