However, there is one thing common to most of the problems hardware researchers face: it's always about miniaturisation. Things are always getting smaller and more powerful (as if you didn't know that)-but the remarkable thing is that Moore's law or Kryder's law (a sort of Moore's law for hard disks) or whatever law keep on holding true.
We can't resist talking about Moore's "law" here. We say "law" in quotes because it isn't, of course, a law per se: it's an observation, one that says that the number of transistors on a chip will double every eighteen months. Or should that be "two years"? And "the density of components on a chip"? It's not important-it's only an academic exercise in history to find out what Gordon Moore originally said. What's important is that things will shrink exponentially, and that usually means that capacities and speeds and such will increase exponentially as well.
Now there's the 2018 thing to contend with. Industry analysts predict that Moore's law will hold good only until 2018, or generally, around the year 2020. Their point: that the shrinkage of silicon transistors cannot continue indefinitely, and that by around 2020, we'd have reached the stage where they cannot be shrunk anymore. But that's about silicon-what about other materials? Enter silicon's cousin in the periodic table, carbon.
The Stuff Soot Is Made Of
NEC's Sumio Iijima is reported to have been the first to observe, in 1991, carbon nanotubes. He saw them in electron microscope images of-of all things-soot. The nanotubes were essentially sheets of graphite rolled into concentric cylinders-one cylinder inside another; such structures are called multiwalled nanotubes. And in 1993, Iijima and IBM's Don Bethune found, in research independent of each other, that they could produce single-walled nanotubes, that is, no tubes within each other. These consist of a single atomic layer of the graphite structure, and look something like a hexagonal wire-mesh rolled up, as in the figure alongside (See figure A typical single-walled carbon nanotube). It is these structures we're interested in; we'll refer to them simply as "nanotubes"-and we'll come back to them later.
The (simple enough) structure of a typical MOSFET. The electrons flow from the source to the drain through silicon, and are controlled via a voltage applied to the Gate
This is something like a MOSFET that has a nanotube built into it. Electrons flow through the nanotube instead of through silicon
The All-Empowering Transistor
To tie up some of the things we've talked about, the idea is that the transistor is the basic building block of most ICs (Integrated Circuits), and it is transistors that do the actual, atomic-level switching of ones and zeroes. When you say a computer works in binary, we mean that it handles only ones and zeroes-simple enough; but this is true even at a much more fundamental, physical level. All of computing is, in a hardware sense, about switching between a high voltage and a low voltage. It's the transistors that do that.
Now, when we talk about the 65-nm manufacturing process, we mean that the conducting lines on the chip's silicon wafer that connect the transistor are 65 nanometres wide. As this figure goes down-that is, as we move to the 40nm or even 20nm process-the transistors are getting smaller. As they get smaller, the chips they comprise get faster, more powerful, and all that. You know the rest of the story-the world demands more computing power, there's a lot of money to be made, and so on-hence we keep shrinking the transistors, amongst other components.
Around 2020, it will not be possible to shrink current-gen-type silicon transistors any more, and that's where nanotubes come in. They will replace the silicon, co-existing with silicon in the interim, and Dr Moore will live on in our hearts and minds.
But to understand what's going to happen in 2020, we need to understand how a transistor works. A little digression…
…called the transistor revolutionised electronics, and therefore everything that depended on it, when it was invented in 1947 by Shockley, Brattain and Bardeen. It's a box with three legs-the source, drain, and gate. The source is where electrons come from; the drain is where they go out from-and the path through which they pass is called the channel. The gate is what controls the flow: an appropriate voltage applied to it reduces the conducting ability of the silicon channel inside, and there won't be any flow of electrons-gate locked. The gate can be opened by a different voltage applied to it. There needs to be an isolation between the channel and the gate, which in the case of MOSFETs (Metal-Oxide-Semiconductor Field-Effect Transistors-one type of a transistor that's used most frequently in ICs today) is a layer of insulating silicon dioxide.
Applying different voltages to the gate therefore has the effect of the transistor storing either a one or a zero-that is, a bit. The transistor is thus a switching device. In a nutshell, that's how you get your ones and zeroes-voltages applied to the gates of transistors.
The Shrinkage Barrier
For decades now, transistors have kept getting smaller. So what'll happen in 2020? Not to get into the details, MOSFETs are the most common type, like we said. Taking the MOSFET as an example, several things will happen when you try to shrink it beyond a certain point-to be precise, when you decrease the channel length and the thickness of the gate insulator.
A typical single-walled carbon nanotube. Of course, this is an artist's impression-they aren't really so colourful!
The quantum mechanical effect called tunnelling (see Jargon Buster), amongst other effects, begins to take over. The leakage currents that result are too large, and render the transistor possibly useless as a switch. On top of that, leakage currents add to the power consumption of devices scaled at this level, as if the current high power consumption weren't enough. Then there's the matter of the thickness of the metal wires that connect the transistor to the rest of the circuit: this needs to be reduced, too. And reducing this width leads, by the laws, to increased resistance-meaning a slowing down of performance, and the phenomenon called electromigration (see Jargon Buster). Electromigration decreases the reliability of the IC, and can even sometimes lead to the loss of a connection-and hence, failure of the entire circuit.
Finally, the resistivity (see Jargon Buster) of the interconnects, too, increases as the width of the interconnects gets close to the mean free path of the electrons, which is the average distance the electron travels before encountering a collision. This happens as a result of "electron scattering" in the metal-the electrons getting deflected from their path-and increased resistivity means more power required to pump the current through.
Enter The Carbon Nanotube
Nanotubes are extremely interesting little things. They can be good electrical conductors-meaning they can be used as nanowires (a regular wire at the nanoscale)-and they can also be good semiconductors. They have several other exciting characteristics important to different types of researchers, which we won't be going into.
Constructing a MOSFET using nanotubes is something like replacing the silicon channel with a nanotube. Now, nanotubes can be treated as one-dimensional: they're so thin that, for practical purposes, you can forget that they have a thickness! One result of their one-dimensionality is that there is there are only two directions for the electrons to flow through them-backward or forward, and electrons flowing in the reverse direction ("backscattering") is highly improbable. This means lower resistivity than a three-dimensional wire under similar conditions.
The transport of electrons in a semiconducting nanotube is "ballistic," meaning that the electrons shoot straight forward, without any deflections along their way. This property holds on the scale of a few hundred nanometres-long enough for them to be used in MOSFETs. This means low leakage currents, which means low power dissipation-which, in turn, means low heat. There is also no electromigration. Switching times are faster-which means higher performance. And finally, due to reasons we cannot explain here, there will now be no need to use silicon dioxide as the gate insulator when a nanotube is being used: materials with higher dielectric constants (see Jargon Buster) can be used, which again means higher performance.
The picture is more complex than we've presented-so let's take a real-world look at what's been going on in the direction of actually using nanotubes in transistors.
From Lab To Fab
May 2002 was a milestone month for nanotech. IBM announced it had created the highest-performing nanotube transistors ever, and proved that nanotube transistors could outperform state-of-the-art silicon prototypes. The proof centred around transconductance, which is a measure of how well something carries electrical current; the researchers involved reported they had achieved the highest transconductance of any carbon nanotube made thus far, and also that their nanotube transistors had twice the transconductance of the best silicon prototypes. Higher transconductance means transistors can operate faster, and that the resulting IC will be more powerful.
Dr Phaedon Avouris, manager of nanoscale science, IBM Research, is reported to have said, "Carbon nanotubes are already the top candidate to replace silicon when current chip features just can't be made any smaller, a physical barrier expected to occur in about 10 to 15 years."
To go into the details a little, the scientists at IBM made Carbon Nanotube Field-Effect Transistors (CNFETs) in a structure that resembled a regular MOSFET, and that's how they were able to compare the CNFETs with regular silicon transistors. The gate electrodes were above the channel, separated from it by a thin dielectric (see Jargon Buster). This was to measure the performance improvements that could be achieved by reducing the gate-to-channel separation, which is an essential aspect of shrinking the transistor.
The devices exhibited excellent switching characteristics-switching is the essential function of a transistor-and high transconductance at low voltages. The researchers concluded that future CNFETs would probably outdo silicon transistors by an even larger margin.
A quantum mechanical effect, tunnelling is the mechanism by which electrons can cross a barrier they are not "supposed to." Think of it like you facing a thin waterfall: it is a barrier, but you can go through it. More scientifically, tunnelling happens because, at the atomic level, things can be waves and particles at the same time: when an electron tunnels through a barrier, it is acting like a wave.
Electromigration is the result of momentum transfer from electrons moving in an electric field to the ions in the interconnect material. Electromigration can lead to the electrical failure of interconnects.
Resistivity is the measure of how strongly a material opposes the movement of electrical charge through it. If two objects of identical dimensions have different resistivities, the object with the higher resistivity will have a higher resistance.
A dielectric is simply an insulator-usually employed as an insulating medium between conductors.
This constant is the ratio of the charge-holding capacity of a condenser made with a certain dielectric material to the capacity of the same condenser with air as the dielectric. (see Dielectric)
Developments followed thick and fast. In early 2004, researchers at the University of California, Berkeley and Stanford University created "the first working, integrated silicon circuit that successfully incorporated carbon nanotubes in its design."
Carbon nanotubes are already the top candidate to replace silicon when current chip features just can't be made any smaller
Dr Phaedon Avouris,BM Fellow and manager of Nanometer Scale Science and Technology at the IBM, J Watson Research Center
According to Jeffrey Bokor, principal investigator of the project, this was a first step in building "the most advanced nano-electronic products, in which (carbon nanotubes are placed) on top of a powerful silicon integrated circuit so that they can interface with an underlying information processing system." In other words, nanotubes had definitively entered the Integrated Circuit.
Nanotubes are naturally difficult to manipulate because of their size, and another groundbreaking announcement came in late 2004: NEC developed a way of positioning nanotubes, while also controlling their diameter. (It was an NEC researcher, if you recall, who discovered carbon nanotubes in soot.) NEC's goal was to develop chips running at 15 to 20 GHz (wow!) while consuming the same power as a Pentium 4-and this discovery was reported as an important step in that direction.
A photo of a Y-shaped nanotube that can act as a transistors, developed by researchers at UC San Diego and Clemson University. The voltage is applied to the stem
Not to be left behind, Intel announced in mid-2005 that it was stepping up its own research on nanotubes. Intel and other chipmakers will continue to use silicon for a long time, but their foray into nanotechnology-and nanotubes in particular-is indicative of how important this field of research is right now.
The Plot Thickens
August 2005 saw the emergence of something entirely new: it turned out that nanotubes could be transistors! At the University of California in San Diego, Prab Bandaru and colleagues, and Apparao Rao (hoorah for Indians abroad!) of Clemson University, constructed Y-shaped nanotubes that could act as transistors. Applying a voltage to the stem of the Y precisely controlled the flow of electrons through the two branches. This is revolutionary when you think about the sizes: silicon transistors measure about a hundred nanometres today at the state of the art, and these Y-shaped nanotubes are just tens of nanometres large. It's almost unbelievable, actually: the transistors are fully self-contained, according to Bandaru.
In March of this year came an even more startling announcement: IBM researchers, working with researchers from the University of Florida and Columbia University, created an integrated circuit out of a single carbon nanotube. To be specific, they used an 18-micron long nanotube to build a 10-transistor ring oscillator, a type of circuit composed of an odd number of NOT gates. The IC is a million times faster than earlier ICs built using multiple nanotubes, but it clocks in at only 52 MHz-no GHz just yet!
But here's a dampener: in regard to commercialising nanotube chips, Fred Zieber, an analyst at Pathfinder Research, said: "It's a way off… It could be a few years or an eternity." But remember that this comment is about all-carbon chips, not about nanotubes being used in transistors: if you recall, transistors that use nanotubes can already outperform their silicon-only counterparts.
Nanotube research is on at various levels. Silicon and nanotube transistors can co-exist on the same substrate. Then there are efforts on to make nanotube transistors faster. The Y-shaped nanotubes that act as transistors are especially interesting, because that means no more silicon: and even more interesting is the recent development of an entire IC being constructed from a nanotube.
The major obstacles in all this are the manufacture of the nanotubes themselves-they come out as clumps of semi-conducting plus conducting types, and need to be filtered-and, of course, assembly, which is difficult because of the nanotubes' extremely small dimensions.
Things are moving fast: it was only proved in 2002 that nanotube transistors could outperform silicon ones, and in 2006, we're already seeing an entire IC made of a single nanotube! There are many years to go-Intel's roadmap, for example, contains several more generations of silicon. And it's only in 2018 that Moore's law-which seems to have envisaged only silicon-is supposed to die, as it were. Who can say what kind of research will happen in the next 12 years? We're prepared for year 2018, and carbon might take over even before that.
Now that is a happy ending!