In a presentation here, Hu showed how alternative technologies are more likely to rejuvenate CMOS rather than replace it. Specifically, he said a thin film of just 3 nanometers of germanium could replace silicon dioxide to provide the high electron mobility needed for next-generation semiconductors. Beyond that, research has shown promise in using carbon nanotubes as a mobility layer that could take 10 times the current of today's materials.
"High-k materials can't go much further, but if we shift to high-mobility materials, we can go a long way," Hu said. "It's likely [that] alternative technologies will just fuel CMOS."
Similarly, new structures like FinFETs--which Hu helped pioneer--"will allow us to extend the life of transistors far beyond what anyone had thought," he added. Companies including IBM, Intel, Freescale and Texas Instruments are already working with the new dual-gated structures as a way to contain current leakage, Hu said.
CMOS is also likely to adopt new structures, such as tunneling transistors, and new techniques like the self-assembly of molecules using lithography-based anchor struc- tures, he said.
In all these areas, "CMOS will be the platform for new devices," Hu said. "I think of CMOS as the mutant monster. You throw things at it, but it just absorbs the energy and keeps on coming."
However, the transition from silicon dioxide to new materials such as germanium may take 10 years. "You don't fix something until it looks like it's really broken--but now it looks like it's broken," Hu said.
Babel at home
In wireless infrastructure, the big opportunity is in undoing the Tower of Babel in home networking, said Jan Rabaey, associate chairman of UC Berkeley's electrical engineering and computer science department. While cell phones approach annual unit volumes of a billion, a combination of wireless cameras, game machines, MP3 players and other devices could add up to tens of billions of units a year, he added.
"We see a huge number of devices emerging at a very rapid rate, and all of these will someday have wireless interfaces," Rabaey said. "So, the opportunity in wireless infrastructure is in ad hoc connections of gadgets to create new operational modes."
Rabaey envisions a new class of wireless router that has intelligence about, and connections to, all the devices on a home network. When a user requests simple functions like playing a song or a video, the router finds the content on the network and the systems that can deliver it.
Such routers will support transcoding among a variety of wireless interfaces and media formats. They will also use sensors to determine the location of a user and the context of his or her requests.
Rabaey said that the routers will have to be very low in cost and power-efficient, yet powerful. Multiple routers in a home--linked on mesh networks--may need cognitive radios to efficiently use available spectrum, he said.
The routers will require a new class of processors, he suggested--perhaps something that emerges from the RAMP project. The Berkeley Wireless Research Center that Rabaey directs is prototyping high-level software called Ambient OS as part of the project. The group expects to publish a white paper on its efforts in a couple of months.
"The Achilles' heel of this effort is configuration, management and control" of devices, Rabaey said.
Separately, other Berkeley researchers said there is no end in sight for the need to create software that is more secure and less buggy.
Toward that end, associate professor George Necula discussed a new class of sketching tools that let programmers outline the fundamental code for a program. Automated compilers fill in the thorny details of the program, which can often be the source of coding errors.
Tests showed that programmers using the sketching tools can finish a program twice as fast as peers writing C programs. In addition, the sketched programs run up to 50 percent faster than the C programs, said Necula.
Berkeley researchers are also working on ways to more thoroughly test and monitor software programs. "Software development today is too cryptic. You have to bring in more information and make it easy for programmers to do that," Necula said.
Necula's team is defining ways to create executable files that include theorems used to check the semantic accuracy of the programs. The theorems require no more additional code than the digital signatures used to verify the identity of the programmer who created it. However, they do require a new class of certifying compilers.
Separately, assistant professor David Wagner discussed new tools to find bugs in software. One tool his team created found as many as 108 bugs in a Red Hat distribution of Linux.
Both Wagner and Necula agreed that despite any new tools on the horizon, programmers will have to put up with buggy and insecure software. "There is no silver bullet here," Wagner said.