Will graphene replace silicon in computer chips? An interview with Bhagawan Sahu by Sander Olson

Bhagawan Sahu

Yes, the center is mainly focused on semiconductor research. That research includes graphene, silicon, germanium, and to some extent 3-5 compounds such as gallium arsenide also. This research should take us to about 2020, but at that point we will have taken CMOS transistors as far as they can go and will need a replacement. My research consists of a possible replacement for CMOS, based on graphene and other novel non-silicon materials

Question: Intel has predicted that they will be able to scale CMOS transistors to 7 nanometers. Do you think they will succeed?

Yes, I do. But at 7 nanometers, the advantage to using silicon CMOS will have ended. So by that point we will need to have new solutions. At 7 nanometers, Intel may not be able to use silicon and will need to employ either germanium or some 3-5 compound s or graphene or something else

Question: Will 7 nanometer graphene transistors be feasible?

Graphene itself is only atom thick – .about a half nanometer. A graphene transistor will perform much better than any silicon transistor, because it has much higher electron mobility than silicon. If that is the only metric industry is looking for in a replacement transistor than graphene is ideal but I am afraid there are other metrics that have not been understood completely related to graphene transistors. If our comprehensive research programs are successful, a graphene chip with low-power, half nanometer thick transistors operating in GHz regime may be attainable by 2022 or thereabouts.

Question: What are the technical challenges to using graphene?

First and foremost, graphene does not switch off. Although graphene can switch at speeds up to 100 GHz, this is only for analog applications which are not required to completely switch off. But we are working hard to try to open a reasonable gap, by various means, for graphene to be suitable for field effect based transistors, on the same operating principles as silicon transistors and be suitable for digital applications.

Question: Is it possible to open up a bandgap with a single layer of graphene?

Graphene is metallic, and therefore does not have an intrinsic bandgap. However in bilayer graphene, an intrinsic band gap can be opened by external voltage and can be tuned. So bilayer graphene (two coupled sheets of graphene) is considered to be a more serious contender to Silicon transistors than the monolayer graphene.

Question: What is a BiSFET? How does its operation compare with that of CMOS?

The bilayer pseudospintronic field-effect transistor (BiSFET) is a novel concept, currently pursued at the center, with the help of semiconductor industry funding. It utilizes two independent sheets of graphene separated by an insulator or vacuum. The charge resides either in the top or the bottom layer and very low voltage operation can be envisioned as this device operates on collective motion of charge carriers or electrons flowing on these two layers. We have used the Ranger supercomputer, at the Texas Advanced Computing Center, to do calculations that might be relevant for understanding the external and internal environment variables that can affect BiSFET operation. The current version BiSFETs do not need an intrinsic gap in graphene for its operation.

Question: How would the performance of a BiSFET compare with that of a gallium-arsenide transistor?

The mobility of a 7nm gallium-arsenide CMOS transistor may be quite high, comparable to that of graphene. But doping a 7nm transistor and its control is quite difficult, and so that route has its own set of technical challenges. A gallium-arsenide transistor will also consume more power than a BiSFET device due to its band structure so I believe that the BiSFET graphene devices, at this moment seem to be the option.

Question: When is the earliest that a microprocessor using graphene could emerge?

The industry would like to see a prototype by 2015. This prototype would then be taken to corporate R&D labs, and hopefully by 2022 the first commercially available graphene microprocessors could emerge. This microprocessor could be low power but still have many billion transistors and operate in an ultrafast regime.

Question: What is the Southwest Academy of Nanoelectronics (SWAN)?

SWAN is located in the Microelectronics Research Center at the University of Austin, and it is one of the four centers of excellence in nanoelectronics in the USA funded by semiconductor research corporation (or SRC). The SRC member companies (Texas Instrument, Intel, IBM, Global Foundries and Micron) has set a research and development agenda for the University PIs at these four centers to examine beyond Silicon CMOS technologies.

Question: Is graphene appropriate for spin-based devices?

Yes. The spin (other attribute of an electron which carries current in a transistor) life time and the response of the spins to external voltages are the critical factors that decide whether charge or spin based device will win at the end. These factors, in graphene, are found to be, better than any other non-carbon based material so far. Therefore, graphene is emerging as a Spintronics material in addition to novel charge based material. BiSFET is a novel charge based device.

Although the electron mobility is similar in nanotubes and graphene, nanotubes are considered not suitable for switching devices. This is partly because of the chirality of the devices, and partly due to the difficulty in connecting nanotubes to other electrical leads. The majority of graphene researchers are nanotube researchers who realized that nanotubes were never going to be used for logic devices.

Question: How confident are you that graphene will displace Silicon CMOS?

The problem with graphene isn’t performance; it is finding a way to fabricate these devices with conventional fabs. With Intel’s and Global Foundries high-k dielectric solution for current and next generation technology nodes, they found a way to integrate high-k material such as hafnium dioxide into their chip-making processes at reasonable cost. But integrating graphene into chip-making fabs may considerably more difficult. It is important to remember that keeping Moore’s law going will rely as much on chip architecture as component technology. Nevertheless, I see a path for Moore’s law continuing for the next 20 years with novel architecture and materials. I am confident that with new surprises related to graphene research coming regularly, replacing silicon CMOS with graphene may not be difficult.