2014 BSJ Blog

Curing the Silicon Addiction

Every few years, we demand that the next iteration of phones, computers, and tablets be faster than the last. What we fail to think about is that each new iteration requires a technological innovation, someone in a research lab has to create a new and better way of making CPUs. At its most basic level, this is attempting to pack more transistors onto the same chip. By Moore’s Law, every 2 years the number of transistors that gets packed onto one chip doubles. Although this isn’t a law set into nature, it has become a benchmark for the processor industry, and everyone expects that it will hold. This has worked for the past 30 years, but we are fast approaching the limit of how small traditional transistors on silicon can get. We can buy MOSFETS (the transistor architecture used in all modern processors) that are just 32nm across. For perspective, that means that 2.6 trillion transistors could fit in the palm of your hand.  What happens now, can transistors get much smaller as they are now? What will processors look like in 20 years? In short, the future can’t be silicon based.

Moore’s Law, showing transistor counts doubling every 2 years, Wikipedia
Moore’s Law, showing transistor counts doubling every 2 years, Wikipedia

As transistors get smaller and smaller, many problems arise if we use traditional architectures on silicon. Firstly, there is a “leakage current” that occurs when the “off” state isn’t completely off. This is a bigger problem with small MOSFETS where the threshold voltage (the voltage where it switches from “on” to “off”) is very small and can barely be separated from random thermal noise. Even if the leakage current is a modest 100 nano-Amperes ( ) per transistor, this adds up quickly, and for a modern cell phone, this creates a current of 10 Amperes, which would drain a cellphone battery within a few minutes. As transistors get smaller, this leakage current becomes even more of a problem, as other effects, like quantum tunneling, come into play. The distance between the gate and oxide can be so small (up to ~ 2nm), that electrons just “tunnel” across the junction, increasing the leakage current. This current also produces heat, which processors have to dissipate. All in all, the more transistors you have, the smaller they are, and the harder they are to deal with.

The evolution of MOSFET architecture, Nature
The evolution of MOSFET architecture, Nature

To combat these problems, the next generation of processors have complex geometries to minimize these effects. Intel has produced 3d transistors which have crazy looking geometries that fix this problem, at least temporarily. But what happens when even these are not sufficient, what does the future hold?

In the future, we need to cure our addiction to Silicon transistors. The more complex geometries have their own problems, and can’t be considered reliable to arbitrarily small dimensions. If we want to continue Moore’s law for years to come, we need new exotic materials and topologies. There are some new technologies that hold promise.

One of the most promising is the Carbon Nanutube Field Effect Transistor (CNTFET). Carbon Nanotubes would be placed on a silicon substrate, and plated with metals to be used as a transistor. The nanotube is a much better conductor than copper, causing fewer heat dissipation problems. It also doesn’t have the same problems of threshold voltage and leakage current, so it can be scaled much easier than traditional silicon transistors. IBM has demonstrated a computer using 10,000 of these transistors, and researchers at Stanford, and other schools continue to work on this new topology. The basic technology is here, but it needs to be scaled and made reliable to meet consumer and commercial demand.

A depiction of a simple CNTFET, Infineon
A depiction of a simple CNTFET, Infineon

Another solution to the silicon addiction lies in using one of the problems themselves. The 1973 Nobel Prize in Physics was awarded to Leo Esaki for his invention of the tunneling diode and discovering the electron tunneling effect, but only in the last few years have we been able to make a transistor out of it. This is a very strange effect where, if an energy barrier is small enough, an electron can simply pass through it. It can be thought of like this: in the classical approach, to roll a ball over a hill, you have to roll it up and then roll it back down, the energy of the system is the same, but you needed the extra energy to go over the hill. With quantum tunneling, the ball would be able to pass through the hill if it was small enough. A few teams have been able to make this approach work with different materials like aluminum gallium and mixes of indium, gallium and arsenic. These “TFETs” (Tunneling field effect transistors) have few of the problems of traditional silicon transistors, they have little leakage current, their threshold voltage is stable, and they don’t heat up significantly.

A depiction of how quantum tunneling works. In the classical approach (top) the electron cannot pass through the energy, but with the quantum approach (bottom) the electron has the ability to pass through, IEEE
A depiction of how quantum tunneling works. In the classical approach (top) the electron cannot pass through the energy, but with the quantum approach (bottom) the electron has the ability to pass through, IEEE

These are only a few of the promising technologies in the field of transistor architecture. To solve the problems we have with our current transistors, we will need to kick the silicon addiction and adopt a novel technology. It is left to be seen which technology will be adopted, and what the next generations of computers will look like.