The warning bells that went off back in 1984, tolling the eventual end of bipolar transistor scaling, are ringing again, this time for bulk CMOS. That's the view of Bernie Meyerson, the vice president of IBM's Communications R&D Center, who is best known for his work fathering silicon germanium-first as a risky technology worth developing within IBM, and now as a process foundation for an increasing variety of communications products by a growing list of companies.
Speaking at the International Reliability Physics Symposium in Dallas, Meyerson warned that "endless extrapolation" is always dangerous, and that applies to Moore's Law as well.
As bipolar devices shrunk to 1-micron dimensions, an increased level of p-type dopants was required in the base to keep the device turned off. When the dopant levels in the base "butted up" against the dopant levels in the emitter, band-to-band tunneling occurred that resulted in catastrophic failures. That tunneling got so bad that "we might as well have have used a paper clip" between emitter and base.
With this in mind, IBM researchers realized in 1984 that bipolar would run out of gas by 1990, and the switch-a "brilliant changeover" to CMOS FETs-was under way.
Today, the industry faces a similar set of challenges, Meyerson warned. "Bandgap engineering is coming to CMOS, and life is going to get very interesting."
"Isn't it interesting," he noted, that bipolar ran into trouble 15 years ago, when the width of the base reached 1,000 angstroms, and CMOS also is running into problems at the 1,000-angstrom (100-nm) generation. These range from controlling power consumption and ensuring reliability to improving materials. Citing IBM's experience, large investments were made to improve silicon-on-insulator substrates enough so they could be used in manufacturing.
Similar materials challenges confront the industry as companies deposit extremely thin films across 300-mm wafers. Even more difficult will be the nearly inevitable transition from silicon dioxide to a new class of high-k gate oxides.
The design community faces different challenges. As circuit performance pushes past 1 GHz, "digital CMOS isn't so digital anymore." RF-like transmission effects come into play that will require major adjustments to the device models and parasitics used in the subgigahertz domain, he said. Some of these problems will show up not as catastrophic device failures, but as soft errors, causing subtle timing problems that may lead to system failures.
Send all feedback to dlammers@cmp.com.
RSS
RSS| 
Digital| 
Mobile
