Behind all the glitzy advertisements and stories surrounding the latest in electronics, the basic laws of physics govern and limit the circuitry in the machine; most circuit designers have come to accept this as set in stone. Tinkering with this delicate world of particles could, however, be the key to unleashing the true potential of tomorrow's technnology.
by Eugene S.Yoo
How fast can microchips operate? Intel, Motorola, and IBM-the major players in the microprocessor industry-are on a relentless race to outdo each other on speed. The catchwords include clock-doubling, clock-tripling, and so on, but it comes down to the fact that each of these companies wants to convince Joe Public that their chip is the fastest. From Motorola's initial offering of a 1-MHz (megahertz) chip over a decade ago to Intel's latest sizzler, a 100-MHz Pentium, the speed jump seems impressive. One hundred megahertz... that's one hundred million cycles per second-how long would it take you to process one hundred million instructions? Yet some of Cornell's researchers are working on devices that operate in the gigahertz range-a thousand times faster than a megahertz-and they are the ones truly pushing the technology envelope.
The Underlying Technology
CMOS (complementary metal oxide switch) technology
is the backbone of the majority of modern electronic systems. On the most
basic level, even more fundamental than the world of AND and OR gates,
two different types of transistors drive the technology of electrical switches.
The first, an n-channel transistor, works when electrons form a charged
path over a channel from a source to a drain-this, in effect, makes the
switch a conducting one. The other type of transistor is known as a p-channel;
p-channels complement n-channels in that holes (particles denoting the
absence of an electron) traverse the channel. Both transistors go into
the formation of CMOS switches.
Holes, however, travel slower than electrons because of the simple fact that they are heavier‹the more massive the particle, the more energy needed to move it. In practical terms, this translates into a trade-off between size and performance for p-channel devices. P-channel devices are inherently slower than n-channel ones operating at the same voltage. In cases where timing is critical to the application, though, p-channels must be made wider to provide the same current as an n-channel device. In the world of area-hungry integrated circuit design, a p-channel device that is three times larger than its n-channel counterpart is not a welcome factor.
A Solution?
Professor Yosef Shacham-Diamand, assistant professor
of electrical engineering, and Kaushik Bhaumik, a doctoral student at Cornell,
worked to develop a new pchannel device to overcome these size and performance
constraints. Their work was made possible because of a joint university/industry/government
research effort involving the National Research Council's Institute for
Microstructure Science, the NASA Lewis Research Center, AT&T Bell Laboratories,
and Cornell University. Traditional p-channel devices yield maximum switching
times of 10 gigahertz (GHz); the p-channel device developed as a result
of these collaborations, however, has a maximum switching speed of 35 GHz.
In terms of transistor speed, this is a speed improvement of over 40%.
Shacham and Bhaumik fabricated their transistor by creating a quantum well for the holes to travel through. Simply put, they accomplished this by layering on a silicon-germanium layer over a traditional p-channel device. Ordinarily, particles (such as electrons and holes) will interact with their surroundings and other particles on the journey from source to drain. With each collision, the particle loses a quanta of energy. There is a minimum threshold, however, on the magnitude of the collision for the particle to lose this quanta of energy-if the collision is below this threshold, the particle will lose no energy. A quantum well serves to confine these particles to a limited region of space. In so doing, this limits the number and magnitude of collisions. As a result, the particle conserves more of its energy and travels faster across the source-drain channel.
As part of the collaboration efforts of the researchers, parts of the layer fabrication were done outside of the confines of Cornell. For example, the apparatus needed to grow the silicon-germanium layer is a very advanced and expensive one, one that Cornell does not possess. AT&T's Bell Labs, however, are specialists in growing the SiGe layers, and were able to produce the layers for Shacham and Bhaumik. The actual fabrication of the device was done in-house at the National Nanofabrication Facility (NNF) at Cornell. In Shacham's opinion, "Cornell is the leading lab in nanofabrication. There is no comparison-everything else is second-league." Other labs, he admits, can be more advanced for specific technologies. The University of Wisconsin is leading in x-ray technology, which should prove to be increasingly important in the future; MIT leads in sub-0.1-micron systems; and Berkeley and Santa Barbara are making names for themselves in the realm of quantum technology. Bhaumik stresses that Cornell's breadth of research is what sets it apart. "[Cornell] caters to so many different projects at the nanotechnology level. We have researchers coming from all those other schools because it's so much more versatile, so much more flexible...they can get all their work done here. Their universities are specialized in a particular area so they cannot provide the range of services that the NNF facility can give."
Previous Research
Research on p-channels in itself is nothing new;
the industry has been trying to solve this particular problem for over
thirty years. However, the bulk of this research concentrated solely on
silicon or gallium-arsenide materials. The most recent example of p-channel
research, and the most related to Shacham and Bhaumik's work, took place
at IBM Corp.'s research labs from the late 1980's to the early 1990's.
A prime difference between IBM's research and Cornell's, however, was the
gate length. The gate length-the distance between the source and the sink
for the particle-was one to two microns for the IBM devices. Cornell's
group focused their energies on fabricating a p-channel device with a gate
length of only 0.2 microns. It was this reduction in gate length that led
to the large performance boost the Cornell group found.
Professor Shacham acknowledges that IBM's research, being first, was very important. He and Bhaumik were able to build on the foundations laid down by IBM's work. The real novelty of Cornell's research, as Shacham puts it, is that it was the first time "all the building blocks have been put together." Just as IBM was able to create a long-channel (>1 micron) p-channel device, other universities have been able to create quantum wells in silicon for several years prior. However, "the real contribution to the industry is in short-channel devices", so Cornell's creation (0.2 micron) is more indicative of the technology that is being used in the marketplace. A gate length of 0.2 microns is shorter than currently used-Intel's Pentium has 0.6-micron construction-so by that fact alone the Shacham-Bhaumik switch is ahead of its time by ten years.
Time to Market
Perhaps one of the most impressive things about
this technology is that it is relatively easy to implement. Shacham drives
home this point with an example: "Let's have a scenario where Motorola
decides to use [this technology]-it will take them between a year and eighteen
months to modify existing designs and install the necessary equipment."
He sees the main deterrent to market acceptance as cost, not a technological
barrier. "You can buy commercial systems for the silicon-germanium [processing]-today."
It is a marketing question whether the possible performance benefits would
outweigh this investment in additional equipment. IBM's own scientists
realized a 15% enhancement in circuit speed when they were conducting their
own research, but this was not deemed to be enough of an incentive to warrant
introduction of the technology to the marketplace. "Though [our] device
shows a 40% enhancement [at the transistor level]," explains Bhaumik, "seeing
how that translates into overall circuit performance is much more complicated."
After corporate labs test for integrated-circuit performance, it becomes
a question of the additional revenue generated versus the capital invested
for many companies. "However," adds Shacham, "it bothers us very little-we
are more interested in the problems of physics."
Future Advances?
As with everything else in nature, there is an upper
limit to the speed of electrical systems-it has to do with the physics
of how fast an electron can travel through a crystal. Presently, transistors
based on gallium-arsenide (GaAs) can operate as fast as 213 GHz, and some
specialized devices such as diodes can operate at 2.5 terahertz (THz).
Professor Shacham muses that for silicon devices, "500 GHz will probably
be a barrier that [will not be] reached in my lifetime." Bhaumik is quick
to point out that 20-35 GHz is plenty; most chips today are running at
100MHz at the most, "roughly an order of magnitude less." Keep in mind,
however, that consumer technology tends to operate at a level five to ten
years behind research technology. The applications looming on the horizon
such as cellular telephones and wireless communications are the ones that
will operate in the 10-20 GHz range. That window of frequencies leaves
a huge bandwidth for potential products and applications. In Bhaumik's
view, his project will be able to satisfy technologies for the next 20
years. Of course, there are applications operating in the 200-GHz range,
and even some in the terahertz range, but they are few in number and very
specialized. "You could count the number [of them] on your hands," says
Bhaumik. "The real good stuff-the money-making stuff-is down where our
transistor operates," he adds with a grin. "There is a physical limit,
of course, but the practical limit has not yet been reached."
Author's Note:
Eugene Yoo is a sophomore in the College of Arts and Science
majoring in Mathematics/Economics, wondering if Asian Studies could possibly
fit in as well. He sends out a greeting of "An-nyong" to his brothers and
sisters in the world.
A Brief Look At Allocation of Research Dollars
The U.S. is now fighting a battle with Japan over the semiconductor field. Professor Yosef Shacham sees the U.S. leading in its funding for university research-"[Japanese] universities are poor and understaffed." However, he hastens to add that "they are not like third-world countries by any means-they are [on some level] between the U.S. and Europe." It is at the level of corporate labs, instead, that the Japanese are at the forefront of semiconductor technology. Without a doubt, claims Shacham, the first batch of 256-megabit memory chips, a 'status quo' many companies are racing towards, will come from the shores of Japan. [Note: At press time, though, the first production 256-megabit samples hailed from the shores of South Korea.]
Kaushik Bhaumik notes that "a lot of the initial development is done [in the U.S.]." For example, the Japanese are investing heavily in electron-beam technology as a future means of commercial lithography; the field measurements for developing that technology, however, were originally done in the U.S. at the Naval Research Laboratory (NRL). In response, Shacham quips that "VCRs were also invented here!" It is simply a reflection of Japanese willingness to invest in technologies that do not have a pay-off within a year.
U.S. industry is also working to address its past shortcomings. For instance, American companies are now much more willing to collaborate and develop a relationship with another company; for example, Advanced Micro Devices (AMD) is jointly pursuing research with Sony for future static RAM (SRAM) technologies. In the past, though, American companies had a reputation, not totally without merit, for trying to make as much money as they could. "The sales pitch of many companies used to be hit and run‹sell and forget," notes Shacham. And now, quality is not necessarily the only reason that consumers flock to Japanese companies for their offerings. The tools that many companies use, American or otherwise, are basically the same; "[consumers] buy equipment today from the Japanese mainly because of the name, you trust them, they service you, and you know they can deliver for the long range. This is changing; slowly now, and perhaps with a lot of pain for many shareholders," laughs Shacham, "But it is getting better."
Are the lines between pure science and engineering being broken down? On the contrary, engineering seems to have a much greater push towards applications. "There was a lot more basic research in the 50s through the 70s-you had scientists that pursued the most arcane subjects...but they were heavily funded," notes Bhaumik. "[Now] the push from both industry and government on universities has been on more practical applications," he adds. Much of what Professor Shacham does is geared directly for the integrated-circuit industry. For example, his group is funded partially by the Semiconductor Research Corporation (SRC), a group dedicated to channelling money from industry to research. "They basically tell you that industry thinks this is important, and you do it. The days when a professor could come in, say, ah, I will pursue this for the next five years, writes a proposal, gets a million dollars...those days are gone. They are really gone," Bhaumik says wistfully. And what of this? According to Bhaumik, even government labs such as Los Alamos and Lawrence Livermore are now told explicitly that they have to concentrate on U.S. business to increase U.S. competitiveness. And as a consequence, some basic research is being swept aside in the rush for profitability. "In my personal view," says Bhaumik, "that is very dangerous. From basic research comes more applied stuff."
Shacham explains that "we are sacrificing our long-term for the short-term because of the assumption that there is no long term without short term." Indeed, it is true that much of the long-term research is funded by the revenue generated by short-term projects. Currently, the balance is on the short-term. And if the U.S. makes enough money, focus will again shift toward the long-term. This is a necessity of the economy because the bulk of American companies are losing money, he notes. So as a researcher who would prefer to do long-term studies, "we are now in a crunch time because we are trying to give the economy a jump start." -ESY