In the early days of integrated components, we used to think kHz (~10[SUP]3[/SUP]Hz) processing was pretty wild. Those systems were quickly eclipsed by MHz (~10[SUP]6[/SUP]Hz) performers and now we’re blasé about GHz (~10[SUP]9[/SUP]Hz) speeds. Recently (2014), DARPA announced a THz (~10[SUP]12[/SUP]Hz) amplifier, built using indium phosphide transistors.
Why stop there? There has been recent experimental work on petahertz (PHz ~10[SUP]15[/SUP]Hz) switching. But maybe we should backup for a second and talk about power dissipation. In anything resembling conventional semiconductor processes, faster speeds mean packing devices closer together, which makes heat dissipation more difficult. And there are fundamental lower limits to switching power, particularly the Landauer limit at ~ 3×10[SUP]-21[/SUP]joules/bit. So just continuing to shrink and pack semiconductor devices closer together isn’t the path to petahertz performance.
We already know that when you want to get to very high frequencies (e.g. for communication) light is a good starting point. Researchers at ETH Zurich recently were able to observe differential optical absorption in a diamond crystal when stimulated by short infrared laser pulses (the frequency of this light is ~½ petahertz), where absorption was measured in pulses from an ultraviolet laser. This they attributed to a dynamic version of the Franz-Keldysh effect, in which optical absorption of a semiconductor changes when an electric field is applied, thanks to interband coupling induced by the field. This implies that electrons and holes in bands in the diamond must be responding to the laser light at petahertz frequencies.
Another related piece of research from the Max Planck Institute shows that that a short pulse (in a visible to infrared range) will, within a femtosecond, increase the AC conductivity of amorphous SiO[SUB]2[/SUB] by 18 orders of magnitude. Further, this process is mostly reversible (as a result, incidentally, the Landauer limit does not apply), so heat production is small, hence the close packing required for system performance at petahertz would be less of a problem. Also fascinating is inducing conductivity through coupling across a wide band-gap in what we normally think of as an insulator. That’s a kind of weirdness we may have to get used to in thinking about band-gap behavior at these frequencies.
Of course it’s a sizeable step from getting electrons to respond to femtosecond pulses, to building switching devices, and from there to having them communicate. But you’ve got to start somewhere. The DARPA chip is described HERE, the Landauer limit HERE, the ETH work HERE, the Max Planck Institute work HERE and HERE and a summary of some of this work HERE (a couple of these require a Nature subscription).