Intel pulled back the curtain recently on some of its future research projects to continue making transistors smaller, faster and less power-hungry out as far as 2020.
In a briefing for reporters and analysts, Intel researchers discussed exotic materials such as carbon nanotubes and nanowires as well as novel techniques to take the transistor down to the atomic level.
The performance and cost benefits from ever-shrinking transistors have driven Intel and the rest of the IT industry for the past 30 years, but advanced researchers are starting to plan for the day in which transistor features simply cannot be made any smaller using conventional materials and techniques.
Once researchers get down to the atomic level, where transistor gates are no wider than an individual atom or two, current manufacturing techniques and materials simply won't work, said Paolo Gargini, an Intel fellow and director of technology strategy at the company.
Intel has just completed the transition to its 90-nanometre process technology. The company has plans in place for materials that will enable it to get down to transistors with gate lengths of just 10nm around 2011. But by 2013, Intel and the rest of the chip industry will need new materials to stay on a two-year cycle of shrinking transistors. A nanometre is a billionth of a meter.
Carbon nanotubes and nanowires are two promising materials that many companies are eyeing, said Ken David, director of components research at Intel. Carbon nanotubes are cylinders made from rings of carbon atoms that would be used as the channel between where the power enters and flows out of a transistor. Transistors work when electrical current is either blocked or allowed to flow through the channel, giving the transistor an "on" or "off" state that represents a bit of information.
Nanotubes are durable structures that allow electrons to move more quickly through the channel than current materials, David said. Intel recently built a carbon nanotube transistor that can theoretically run three times faster than a conventional transistor of the same size and power consumption, he said.
While many companies and universities have grown nanotubes, the challenge will be integrating them into a high-volume manufacturing process, David said. The same is true for silicon nanowires, which Intel hopes to one day use to build an entirely new transistor shape, he said.
The metal gate on a current transistor sits above the channel and silicon material. In about four years, Intel plans to integrate tri-gate transistors into its process technologies, where the gate material surrounds the transistor on three sides. This helps to control current leakage, especially when the transistor is turned off, David said.
But Intel believes the ultimate transistor shape would be a pure cylinder with a gate wrapped entirely around the channel, David said. This type of transistor would strike the best balance between electron mobility and leakage control, he said. A silicon nanowire could be the material used to build this transistor, but a great deal of research needs to be conducted in order to optimise the electrical performance of the transistor and ready the structure for manufacturing, he said.
Nanotubes and nanowires are materials that might come into use between 2013 and 2019, David said. After that, the company's plans are still very much up in the air.
Intel works very closely with leading colleges and universities around the world on research projects that will bear fruit in the 2020 time frame, Gargini said. These relationships allow Intel to offload the vast amounts of early research to the universities and then select the most promising techniques, he said.
One of those techniques is called spintronics, said George Bourianoff, manager of emerging research technologies at Intel. Gates have historically been used to generate opposite transistor states, but the company is looking into the feasibility of using other methods such as a magnetic field to accomplish the same thing.
Spintronics is the study of spinning electrons. When electrical charges revolve, they produce a magnetic field that points in a certain direction depending on the direction of the spin. A field pointing in one direction could represent the "on" state of a transistor, while its opposite could represent the "off" position.
One of the benefits of this technique is that it requires a very small amount of power to change the spin of the electron and have that electron represent the opposite state, compared to the power needed to switch a conventional transistor gate on or off, Bourianoff said.
IBM is also looking into using spintronics as an advanced transistor manufacturing technique. It has already used a similar method to create a giant magneto-resistive head on a hard drive, dramatically expanding capacity.