Shape speeds transistor

Shape, more than size, defines the latest Lucent Technologies’ Bell Labs innovation.

Dubbed the ballistic nanotransistor, Bell’s new unit is a quarter the size of today’s on-chip transistors, but it’s the smooth shape and unimpeded current flow that is most significant.

While the size of the latest units has been dramatically reduced, the revolutionary innovation is in the ballistic aspect of Bell’s work – this provides an unimpeded flow of current, similar in concept to a bullet travelling through the air.

The world’s semiconductor industry has been creating smaller and faster silicon chips with the help of shrinking transistors, but the industry had been approaching physical limits.

“It is becoming more and more difficult for a switch to provide a current as the switch gets smaller and smaller, and as you are trying to integrate more switches on to a chip,” said Greg Timp, a member of the technical staff at Bell Labs. “The accepted way to do this is to thin the gate oxide, the insulating layer between the transistor’s gate and channel, and it can’t get much thinner since a short circuit will occur.

“We think we can get down to sub 20 angstroms gate oxides, but the trouble is it doesn’t really help to go too much thinner…because if you thin the gate insulator too much, quantum mechanical tunneling starts to occur between the gate and the channel, and that is shorting the channel out,” Timp added.

Given that reality, Bell Labs have been concentrating on another major limiting factor: channel resistance. With the transistors available today only about 35 per cent of the input current actually makes it from source to drain, while the remainder scatters as it collides with the rough edges of the insulating layer.

“[There are] two things that decide how much current the transistor can produce: one is the thickness of the gate oxide, the other is how much resistance there is in the channel,” Timp said. “The idea is to make the channel very short and to remove as much scattering as we can, so that we can reduce the resistance to practically zero.” This would then allow more current to pass through the transistor. “The predominant scattering mechanism, in the field range that we are working with, has to do with how rough the silicon dioxide/silicon interface is,” he added.

And the field range they are working in is minuscule, to say the least. The length of the channel on the ballistic nanotransistor is 25 nanometres (nm). To put that into perspective, an average human hair is 10,000nm in diameter.

So how is a transistor with less resistance actually manufactured? Step one is silicon wafers must be properly cleaned to remove all metals and surface roughness. The wafers are cleaned in a combination of sulphuric acid and peroxide at 150 C. They are then bathed in pure water and finally hydrofluoric acid. Now, they are ready for the next process: rapid thermal oxidation.

“By introducing the right amount of oxygen at the right temperature (in a furnace at 1,000C) you grow an oxide on this very hot silicon wafer,” Timp said. The process takes anywhere from 15 seconds to one minute.

It is this overall technique which creates a smooth interface between the silicon wafer and the gate oxide, thus reducing resistance.

The researchers at Bell also discovered another interesting bit of information. When they tested transistors with gate oxides of 1.3nm they found the drive current efficiency to be 75 per cent, yet at 1.6nm the efficiency was 80 per cent. This is counterintuitive to the theory that thinner gate oxides allow for better current flow. Timp and his fellow researchers theorized that a thinner gate caused the electrons to be pulled up closer to the gate where there was still some surface roughness that caused scattering, thus reducing efficiency.

Down the road this may change, although there are no guarantees. Timp said it is difficult to imagine overcoming the physical limitations of silicon dioxide and even if a new compound is found it might not be manufacturable.

“Lots of compounds have better dielectric (insulating) properties but their tunneling characteristics aren’t as good and they are not generally manufacturable. There is no single dielectric that seem to meet all of the requirements as well as silicon dioxide and this is what is befuddling everyone, because we don’t know what is going to replace silicon dioxide,” he said.