Alyssa Apsel looks to photonics as the small solution for a big problem.

By Jerry Gabriel

When mirrored objects seem to shrink into infinity, it's merely an optical illusion, but optical signals may be the real deal for shrinking the integrated circuit.

FOR those of us who use computers to do uncomplicated things like check the weather or write research papers, the how of these machines remains elusive. Most people’s understanding probably doesn’t run much deeper than a rudimentary awareness that computers operate by means of binary code—0s and 1s operating as messages to the system to do this and not that. We might thank the ’90s tech boom for even this scant knowledge; before the internet came along and rescued home computing, there may as well have been little sprites doing all of the work in there as far as we were concerned.

But understanding the binary is really just part of understanding computing. It’s the interface with the real world that seems so difficult to get one’s mind around: What is it that goes on inside these things that allows you to hit the delete key and have that action translated somehow into 0s and 1s, ultimately prompting the machine to carry out your wish? The answer to the question is circuits; they are the place where the analog “real” world (in the form of voltage) meets—and gets translated back and forth with—the digital cyber world.

And if you think of the effort to make computers—and devices that rely on computing, like calculators and iPods—smaller over the last half century as a race, then at the center of that race has been the integrated circuit. Making the body of a computer is after all not a difficult task; what was difficult was making a smaller computer that performed at least as well as the bigger one.

In 1965, Intel co-founder Gordon Moore predicted in a paper that the number of transistors per square inch on integrated circuits would double each year for the foreseeable future (as they had since the integrated circuit’s invention in the late 1950s). The time frame of the prediction was a little off—it was more like 18 months—but the gist of the idea, known since as Moore’s Law, has held steady, a bit of wisdom more valuable in the computer industry than any of Twain’s best one-liners.

The industry’s ability to scale, or shrink the number of transistors on an integrated circuit, has accounted for improvements in power, density, and performance like clockwork ever since. The problem is that the future that Moore was able to imagine in 1965 has nearly run its course. Scaling, in other words, is beginning to meet with material limits—and just about everyone agrees that this is true. In its 1999 “International Technology Roadmap for Semiconductors,” the Semiconductor Industry Association itself predicted that by 2007, electrical fixes for chip-to-chip interconnect problems would be growing scarce.

Rajit Manohar, associate professor of electrical and computer engineering at Cornell, explains the nature of the dilemma thus: “There are a number of different physical mechanisms that together govern the behavior of devices used in a processor. Some are desirable, others not, because, for example, they cause a device to stop behaving like an ideal transistor. When devices are made very small, these less-than-ideal effects begin to dominate, making it difficult to design fast devices that have low power consumption.”

If we agree that our yearning for faster and smaller computers is going to continue over the next decade and beyond—and there seems no reason to think otherwise—then the computer industry is faced with a simple though thorny question: what next?

Alyssa Apsel, Clare Boothe Luce Assistant Professor of Electrical and Computer Engineering, believes she knows the answer to that question: a new paradigm.

Whether the actual limit is 35 nanometer scale or 15 or 10, Apsel argues, the limit for shrinking the integrated circuit is real and it calls for a different way of thinking. “We’re going to have to look at a lot of approaches of doing processing in order to improve system performance,” she says. A few such approaches already under examination include innovations in pipelining (the way a processor orders tasks) and optimizing a processor for a single application (as opposed to creating a machine, like our current computers, that can do a little of everything, though nothing spectacularly well).

At the chip level, says Apsel, “we’re going to have to build integrated systems that combine approaches, that use things like nanotechnology, MEMS (micro-electro-mechanical systems), and other sensor devices.

“And,” she adds, obviously leaving the best for last, “photonics.”

Photonics is something that has been used most notably in telecommunications: fiber-optics is a word familiar to everyone’s ears. Optics, because of the speed of light, are an excellent means of passing data over long distances very quickly (say, between continents). But over short distances like those on a chip, the thinking has gone, optics are no better than wire—in addition to being more expensive.

But Apsel and her lab group, which includes graduate students, undergrads, and one postdoc, have been investigating the truth of this wisdom. What they’ve found is that it might be possible to combine photonics with more conventional electronic approaches in order to design circuits capable of achieving a lower cost, higher performance machine.

“We began to ask what an optical signal was good for, rather than just speed,” Apsel explains. “There’s a lot of benefit to using an optical signal in a fiber when you want to transmit data from one point to another. We’re looking at ways to make that system useful for a computer.”

The integrated photodetector shown here is used to convert light signals into electrical signals in a silicon-based optical receiver.

One of several possibilities that Apsel has been investigating is electrical isolation. Electrical isolation means that part of a system does not interact electrically with another part. Because of this, you can transmit data from one source to many nodes at higher speeds than you could achieve electrically. This is because each node no longer drains current directly from the source, as it would if they were electrically connected. With electrically isolated nodes, connecting more circuits or nodes to the “receive” side of the interconnect will not slow down the signal because in this method, the signal has a smaller effective load to drive.

“There are plenty of examples in circuit architectures where driving large loads is a problem that slows performance and costs energy,” says Apsel. “With electrical isolation, a signal can be broadcasted to more destinations without degradation of the signal speed.”

Some notable areas where electrical isolation might come in handy, says Apsel, are clock distribution and memories. Memories, for instance, have multiple cells connected to a single address line, all receiving the same signal. As memories increase in size, it takes longer and longer to access them because of the larger load that needs to be driven. Electrical isolation can speed up that access by connecting to all the cells at once with the ample drive.

Apsel speculates that in the next decade or two, we will be seeing these types of technology—if not this specific one—put in place in industry. “We’ll see basic systems with higher performance pushed not by scaling, but by the integration of other technologies,” she says. “You’re going to see optics within a computer. I don’t want to bet my life on it, but I’m pretty sure.”

Graduate school trains you for much of what you end up doing as a professor, but it does not train you to run a lab. This aspect of the job, says Apsel—and particularly working with graduate students—has been a very pleasant surprise.

“It’s a real collaboration,” she says. “I find it the most interesting thing to see how others think, what kinds of ideas they generate, how they work creatively. A graduate student who is excited about a problem can very easily infuse me with their excitement.”

In her own graduate school experience, at Johns Hopkins, she felt that she had a lot of freedom to take things in the directions she was interested in, which allowed her to get very excited about the work she was doing. This excitement, she says, goes a long way toward understanding.

“Graduate students don’t need you very much actually, maybe just ten percent of the time,” she says. “The value you add is by having a global context, being able to look at the world and see how things fit together and being able to come up with interesting new problems.”

In describing the atmosphere of the Apsel lab, second year graduate student Zhongtao Fu stresses the importance of the collaborative, interdisciplinary atmosphere. At weekly meetings, the group members informally present their work to one another. The other members then respond, sometimes asking probing questions or seeking clarification. These discussions help to drive the discovery process for everyone. Each member, in a sense, is learning about the diverse areas that each of the other members wander into in the course of research. “When we don’t know how to make a circuit meet a particular requirement,” Fu says of working with Apsel, “we learn how to do it together.”

When Apsel first embarked on her current research, she says she didn’t know much about optics; her background was in more conventional circuit design. But she saw this deficiency as an opportunity to learn, an exciting challenge. The hallmark of her career so far has been her ability to see the system in its entirety and understand not just how the parts work together, but based on how those parts work, what could be improved upon.

This receiver circuit, constructed on a sapphire substrate, is bonded to an array of photodetectors and, because sapphire is transparent, can be seen through the back side of the chip. Optical signals can be beamed through the substrate to speed movement of data across a single chip and from one chip to another.

Tania Khanna, an undergraduate in the lab, sums up the interconnectedness, the necessary interdisciplinary nature of this work: “There is no way you can design a useful product or idea if you know nothing about where it will fit into the grand scheme of things, though the way you view each part in this scheme is strongly affected by your area of study.”

Apsel says that as a researcher, she’s learned that there is more than one way to proceed—sometimes you might address an existing and well known problem, which is the way we tend to imagine engineers working. But sometimes you might go looking for a problem that no one else is trying to tackle. More often, in fact, she says it is the latter of the two that drives her research agenda.

“Frequently,” she says, “my job is not really looking at the problem and trying to come up with the solution; it’s looking at the solution and coming up with problems. I’ve done that a lot.”

The optics work is a good example of this. Apsel knew there were certain advantages to optics; the long-distance telecom applications made this abundantly clear. But she began asking what other kinds of applications weren’t being used, if they were already being explored and, if not, if they were worth exploring.

“Being an academic is almost like being a comedian,” she says with a grin. “You can’t just sit there and grind out jokes, because they won’t be funny. It has to be organic.”

For Apsel, keeping the process “organic” seems to be connected to the combination of collaboration, imagination, and interdisciplinarity; together they provide perspective, and perspective is something Apsel values a great deal. She has a history of finding ways to get it. Her roommate at Swarthmore College, Sylvia Chong, herself an assistant professor of literature at the University of Virginia now, says that though Apsel knew she wanted to be in engineering early on, she was always straying outside her field—sometimes far outside—in her undergraduate coursework. She would take upper-level courses in subjects like history, and during her last two years of college, says Chong, Apsel acquired a strong interest in oil painting and studio art. Chong and many of their friends modeled for her.

“I always thought it was unusual to have a scientist who was interested in the visual arts,” says Chong. “Most scientists I knew who did anything artistic were musicians, maybe because music was more mathematical or precise than visual art.”

But for Apsel, it was the precision she was getting away from. Painting, she says, was a way of balancing her mind, giving her a break from logical thinking, allowing her to look at the world in a different way.

“It’s very hard to paint and stabilize a circuit at the same time,” she says of the period. “It was almost like a way of forcing myself to do meditation. After several hours, my perspective would be shifted and I could leave the studio and either attack problems that had been stymieing me with a fresh approach or I could at least sleep well without obsessing about an equation.

“I sometimes feel,” she adds, “that the way we study engineering encourages myopia, the way we have students be hyper-focused, only looking at one type of problem with one set of applications. I would urge young engineers not to fall into the trap of specializing too soon, of becoming pigeon-holed. Focus on fundamentals, on basic science. Don’t spend a lot of time learning detailed protocols. The most interesting ideas I’ve seen have come out of breadth of ideas.”

Jerry Gabriel lives in Ithaca and is a lecturer in
Cornell’s Engineering Communications Program.