In Search of the Next Switch
Pete Singer Editor-in-Chief
Jeff Welser, director of the Nanotechnology Research Initiative, is on a mission to find a replacement for the transistor, which is rapidly approaching fundamental limitations. It just cannot be scaled much further. He said the goal of the NRI, which falls under the umbrella of the Semiconductor Research Corp., is to discover the next switch, a new mechanism for computing that goes beyond simply improving today’s transistor.
“We haven’t found it yet,” Welser said. “That’s a very safe statement for me to make. There’s no question that CMOS is a very difficult technology to supplant in that it’s not perfect on any of the metrics but it’s extremely good at most of those that you care about: power density, power efficiency, performance, speed; obviously you can make it quite small. What you’re seeing in terms of the limiting factor is the power density. We continue to make it smaller but power density remains constant. The scaling rules always allowed you to half the power for each transistor, so you had double the amount of transistors at the same power as they were before, but half the power for each transistor. That’s all broken down since about the 90nm node since we’ve reached the voltage limit. Somewhere around 1V, it’s hard to scale the voltage much lower and that breaks down all the rest of the scaling rules. Leakage power gets a lot of press, but also active power, when you’re switching the device off and on. That’s what used to scale for us every generation and it doesn’t anymore.
That’s the key issue you have to find in a new device. You don’t just have to find something smaller — CMOS is going to get really small and I have no doubt it’s going to be difficult to find something that goes significantly smaller — but you need to find something that actually uses less power in that footprint but somehow has the same or greater computational capability.
The newest NRI center is called Midwest Institute for Nanoelectronics Discovery (MIND) and is headquartered at Notre Dame University in South Bend, Indiana. The MIND research team includes Purdue University, Penn State, Illinois, Michigan, National Magnet Lab at Florida State, and the University of Texas at Dallas, with research collaborations linking the National Institute of Standards and Technology, Argonne National Laboratory, and the National High Magnetic Field Laboratory. MIND research is organized around two concepts: energy-efficient devices and energy-efficient systems. The energy-efficient devices themed include development of (1) tunnel transistors with low voltage and low subthreshold-swing, (2) graphene transistors based on spin, tunneling, and thermal rectification, (3) quantum transport modeling tools for MIND devices, and engineering of energy dissipation in nonequilibrium devices. Magnetic quantum dot cellular automata (QCA) is a core focus area.
Wolfgang Porod, director of Notre Dame’s Center for Nano Science and Technology, said that the semiconductor industry has so far relied on the CMOS transistor, which is based on charge based state variables. “Moving charges around is costly, and there’s dissipation that contributes to the heat problem that the industry is facing. Heating is probably the number one issue that might limit future scaling.”
Porod said there’s quite a bit of work going on across the NRI to look at spin-based state variables. “We’re doing some work here as part of the NRI on exploring magnetism to do logic. Mostly magnetism is used for data storage. For information processing we for the most part use charge but there’s some prospects of also basing information processing on magnetic phenomenon. At Notre Dame, a more general scheme that we’ve worked on for some time is quantum dot cellular automata. It’s not really quantum dots that we’re working on but it’s a different way of representing information. So the idea of the QCA scheme is to represent information by the arrangement of charges rather their flow. A scheme like that is inherent in lower power and it also works for magnetic implementation. Instead of having arrangements of charges, we can also work with arrangements of magnetization — essentially flipping magnetizations to manipulate information,” Porod said.
Although early research on some “next switch” alternatives, such as the spin transistor, appear promising, it has been done at very low temperatures, which would be impractical in real-world applications. “If there’s an interesting science effect at 4°K and it will always be at 4°K, then it’s probably not something we want to spend a lot of time with at NRI. On the other hand, it might be a really interesting science effect at 4°K and you might continue to do a lot of science work at it at low temperature but at the same time work out what are the limits of that temperature. Are there ways to get it higher? We’ve definitely seen examples of that already where, by asking that question early on, you find out pretty rapidly that this is something that can operate at room temperature,” Welser explained.