For Intel, moving away from silicon is a matter of when, not if.
Chip manufacturers will continue to follow the evolutionary path described by Moore's Law for several more years, but engineers will have to substantially change the underlying design and ingredients of their products, said several Intel researchers on Friday at a presentation at the company's headquarters in California.
And one of the big changes will involve enhancing, and then ultimately moving away from, chips made with silicon transistors. By 2014, chips may include transistors constructed with carbon nanotubes or silicon nanowires, rather than silicon. By 2020, more radical changes will likely be required.
"By 2010, we should have a pretty good idea of what the device that will take us beyond CMOS [is]," said Paolo Gargini, director of technology strategy at Intel. CMOS, which stands for complementary metal oxide semiconductor, is the technology base for silicon transistors.
Intel's outline of the future captures the challenges facing the industry overall. From the 1960s until the turn of the century, chip designers largely increased performance by shrinking the transistors that go on their chips and then increasing the number of those transistors.
Unfortunately, the benefits have begun to wane. Starting in 2000, designers moved into what Gargini calls the "equivalent scaling" era, in which chip designers improve performance in part by shrinkage, but also by using additional technologies.
In 2007, for instance, Intel expects to start producing chips using the 45-nanometer process. These chips will differ from existing chips in part because the transistor gate will be made of metal rather than silicon, while the "gate oxide" -- an insulating layer that controls the flow of electrons inside the transistor -- will be made out of something other than silicon dioxide, which has been the material of choice for decades.
The new oxide gate will be about twice as thick as a gate oxide made out of silicon dioxide, but it will behave as if it were a thinner, higher-performance layer. Besides increasing performance, the thicker layer made of yet-to-be-disclosed material will leak less electricity.
"From an electrical point of view, the designer cannot tell the difference," Gargini said.
Enhancing silicon in this manner will allow engineers to develop chips made using the 22-nanometer process (a nanometer is a billionth of a meter), which should begin in 2011 or 2012. However, performance will need to be boosted again by 2015.
Then, designers will move into the "integrated solutions" era, in which chipmakers will replace the transistor gate with different materials, such as carbon nanotubes or carbon nanowires.
"Silicon is still the substrate for building these devices," said Ken David, director of components research in the Technology and Manufacturing Group at Intel. Nonetheless, many of the underlying components inside these CMOS-like transistors will be substantially different, as will many manufacturing processes.
