Yoshio Nishi keeps adding new facets to a sparkling career. Nishi spent 20 years at Toshiba Corp., where he did pioneering work on applying electron-spin resonance techniques to study the silicon-oxide interface. He then jumped the Pacific, eventually heading up silicon technology development at Texas Instruments Inc. Today he's director of two centers at Stanford University: the Stanford Nanofabrication Facility and the Center for Integrated Systems. Nishi sat down recently with EE Times to talk about nanoelectronics research.
EE Times: There seems to be a certain tension between the traditional silicon people, who claim they are doing nanoscale devices, and those developing alternative types of logic and memories.
Nishi: We can divide this into evolutionary and revolutionary nano.
Evolutionary nano includes silicon-based CMOS with new materials, such as metal gates and high-k, and nonsilicon materials for the active semiconducting part, such as germanium. But the basic principles are the same.
Revolutionary nano can be divided into zero-dimensional, one-dimensional and even two-dimensional devices. Zero-dimensional includes such things as quantum dots or nanodots. There are some people who are working on replacement of chemical dyes, so that red, blue or yellow can be achieved just by adjusting the diameter of the dot.
Nanowires or nanotubes are almost one-dimensional conduction devices, with reasonably well-understood physics and chemistry. Today we can use carbon nanotubes to make NMOS transistors with very high current densities. That is the good news. What is missing in one-dimensional nano is the capability to grow those wires or tubes in very specific spatial locations, and with the needed characteristics. That will require more of an engineering breakthrough for CNT [carbon nanotube] circuits to become a reality.
EET: What about two-dimensional nanotechnology?
Nishi: Two-dimensional nano is a little bit premature. In that category I include molecular electronic devices. Organic devices have nonvolatility and are used mostly for memory functions, but unfortunately they cannot switch fast enough. Then there are spintronic devices, which have been experimentally verified at low temperatures. Theory shows that they could be operable at room temperatures. I [also] put the single-electron devices into the two-dimensional category.
EET: Will any of these alternative devices work out?
Nishi: In a broad sense, nanoelectronics and nanotech have created a lot of excitement within the research community. There is some hype, but there is a lot of real meat inside.
Some people tend to focus on the very futuristic stuff, such as nanorobots that move through body to provide drugs at certain areas, perhaps enab-
ling a new form of chemotherapy.
For the most part, nano research activities are not that futuristic. NanoMEMS nanoscale electromechanical systems are a promising field. There are nanoactuators and sensors. There are converters to go from electrons to photons and from photons to electrons. Within the medical and biological fields, a lot of really good stuff already has been productized.
EET: What is going on in nanoMEMS that is interesting?
Nishi: NanoMEMS research is mostly in biological applications. One field of research is to measure extremely small forces, such as from the muscle of an ant, where we need to have a very high-sensitivity sensor.
People are working on ways to separate a protein, depending on the molecular weight or size. There are structures with many posts standing on a silicon surface, where a molecule will be accelerated either by the density gradient or by the electric field. And then, depending on the molecular size or weight, you can separate one from another.
Similar techniques are being used to do DNA sequencing, because once DNA is separated it resembles a large molecule. Those techniques are being developed for use in drug design and research, but they still have some ways to go.
EET: What is the difference between these kinds of research and, say, chemistry?
Nishi: Nano research is interdisciplinary. It is difficult to separate what is chemical, biological and medical. We can manipulate macromolecules by different means, including micromachines. Microvalves and chemical sensors require and utilize those nanostructures.
Cross-disciplinary collaboration is a characteristic feature of Stanford and other institutions of a similar caliber. Stanford's initiative for nanoscale materials and processes involves two professors from the EE department, namely Krishna Saraswat and myself; three professors of materials science and engineering; two chemical engineers; one mechanical-engineering professor; one professor of applied physics; and others.
EET: Give us a progress report.
Nishi: We have made excellent progress in terms of finding the metal work function control mechanism for the metal bilayer structure for the MOS gate. We have developed germanium MOSFETs in both the n- and p-channel, developed a quantum chemical understanding of the ALD [atomic-layer deposition] precursor mechanism and figured out how to do area-selective ALD. And we are working on a hafnium dioxide-based high-k gate stack and the physical and electrical characterization leading to better design principles for such structures.
EET: The big companies also employ chemists. IBM comes to mind as a company that has some world-class scientists and does interdisciplinary research.
Nishi: It used to be that new devices always were started by the EE profs. Now, chemistry professors may find these new devices. So we need to have strong capabilities in many different disciplines, and that is very inefficient for most companies. . . . That is why many companies want to send researchers to Stanford, and my friends at MIT would say the same.
EET: Do you think carbon nanotube-based devices will work out practically?
Nishi: CNTs certainly come under the category of revolutionary nanoelectronics. The good thing is that CNT transistors have much higher hole mobility, especially good for p-type, depletion-mode devices. CNTs can carry five to 10 times more current, per unit of cross-section, compared with silicon. If used for interconnect, they can carry an order-of-magnitude more current than copper, measured on the basis of amperes per square centimeter.
But if you want to carry the same amount of current as copper or aluminum, you must bundle thousands of nanotubes. The difficulty at this moment is that CNTs are like a mesh winding together, made up of carbon and hydrogen. If the mesh is twisted or not twisted, that determines the chiralty and whether or not the CNT is semiconducting or metallic. Depending on how you make it, a certain percentage will be single-walled nanotubes. Or if you created multilayered, multiwalled CNTs, they will have different electrical characteristics.
The first challenge is that we do not yet know how to control this process. At Stanford, we mostly are making semiconducting-type CNTs. Professor Hongjie Dai's group has demonstrated the world's first CNT MOSFET with high-k and metal gates, and more recently they have built both n- and p-channel CNT MOSFETs and showed excellent current-voltage characteristics, indicating ballistic transport.
EET: Does that mean we're close to practical applications?
Nishi: I would say that the basic characteristics of CNTs are great. The real challenge before practical use for integrated electronics is: How can we grow CNTs on the X/Y plane exactly where we want to have them? And we want to control semiconducting or metallic-type CNTs with 100 percent accuracy. Some people may see that and say, "Oh, it's impossible." Others will say it is an opportunity for a breakthrough.
EET: Some people believe that too much nano research money goes to CNTs and not enough to other areas.
Nishi: At Stanford, only two or three faculty members are growing CNTs. It is probably the same at Harvard. The equipment needed is very inexpensive, so I don't believe we are spending too much on CNT research. The same argument would apply to germanium and all forms of nanowires. Nanowires are guaranteed to be semiconducting. The bad news is that the performance improvement is not as great as for CNTs, but potentially we would improve the integration density significantly.
EET: You led the team for Toshiba's 1-Mbit CMOS DRAM. Some people say CMOS may be the most efficient way to do logic. Do you have a gut feeling about that?
Nishi: To be very frank, CMOS does have tremendous noise immunity, and that can be translated into a large on- and off-current ratio with basically zero standby power it does not waste idle power as PMOS or NMOS used to do. CMOS has a well-established capability to integrate billions of devices, which makes it the most economic way to create circuits.
That raises the question: Can we make something else? Are any of these forms of revolutionary nano going to replace CMOS? If we can have controlled growth of CNTs or silicon wires, my answer is yes. But if I take a safer look at this, I may say that CMOS may remain as the basic platform, with some functions replaced by revolutionary forms of nanoelectronics, especially in the memory side. Instead of flash, we could have new forms of memory on top of a silicon platform that would significantly increase the memory density. That is the most likely scenario, compared with everything being based on CNTs or nanowires.
EET: Many people seem optimistic about memory but pessimistic about logic.
Nishi: I would agree. Already, we have phase-change memories, ferroelectric memories. The revolutionary, molecular memories still have a long way to go, but they can be made very nicely once combined with self-assembling mechanisms to create highly dense arrays on a two-dimensional surface. The problem is that molecular memories as we know them today are not fast enough to be a universal high-speed switching memory. But they are dense. And we are always looking for some kind of breakthrough to get some kind of conducting molecules.
My question is: What is the limit of the switching speed of such molecules? Even if the molecule is fast, unless we have enough current coming out they cannot be high-speed. Even with theoretical calculations, there appears to be very high impedance.
Also, there are magnetic devices, such as MRAMs. And we are looking at spin transistors. The spin remains unless it is scattered by lattice vibration. Temperature comes into play, and the on- and off-current ratio of such spin transistors may not be big enough. These areas present a lot of interesting opportunities, so we need to do more research. Oftentimes in the history of the semiconductor industry, something that is viewed to be impossible becomes possible.
EET: Do you have any suggestions about how the nano research money could be well spent? Are there bottlenecks?
Nishi: We need some way to update the infrastructure at universities. Stanford is fortunate; there are many companies supporting us. But if I look at the national level, not every university has the infrastructure needed. If we could update that ability to support mostly experimental research, then society could get a greater return and students would be even better trained.