"Ballistic Deflection Transistors" are a new design that coud replace regular transistors. (Thanks to reader Sigma for pointing it out). Instead of running electrons through a transistor as if they were a current of water, the ballistic design bounces individual electrons off deflectors as if playing a game of atomic billiards. Such a chip would use very little power, create very little heat, be highly resistant to "noise" inherent in electronic systems, and should be easy to manufacture with current technologies. All that would make it incredibly fast. The National Science Foundation granted the University of Rochester team $1.1 million to develop a prototype.
The team has already had some luck in fabricating a prototype. The ballistic transistor is a nano-scale structure, and so all but impossible to engineer just a few years ago. Its very design means that this "large" prototype is already nearly as small as the best conventional transistor designs coming out of Silicon Valley today. Feldman and Diduck are confident that the design will readily scale to much smaller dimensions.
The Ballistic Deflection Transistor (BDT) should produce far less heat and run far faster than standard transistors because it does not start and stop the flow of its electrons the way conventional designs do. It resembles a roadway intersection, except in the middle of the intersection sits a triangular block. From the "south" an electron is fired, as it approaches the crossroads, it passes through an electrical field that pushes the electron slightly east or west. When the electron reaches the middle of the intersection, it bounces off one side of the triangle block and is deflected straight along either the east or west roads. In this way, if the electron current travels along the east road, it may be counted as a zero, and as a one if it travels down the west road.
Added (pointer from Roland Piquepaille Technology trends): In today's silicon-based transistors, only 35 percent of the input current flows, via the channel, from a transistor's "source" to its "drain;" the remainder scatters as it collides with the rough edges of the insulating layer. The 1999 version of the BDT had 85 percent of the current being transmitted from the source to the drain, which yields the ballistic transport. So 5 times less heat should be generated. However, whenever a computing operation is performed heat is still generated. This heat can be reduced via reversible computing.
The reduction of heat from inefficient current flows seems like a necessary thing to do. There probably are several technical ways to do this. This may or may not be the best way. It looks like an important large incremental improvement.
A past article that I had discussed the physical limits of computing. Ballistic action was discussed as a cooling mechanism and the amount of computing it would allow based on heat from computation. It is 100 to 1 billion times more than fractal cooling. Fractal cooling is 100 times better than current passive cooling.
Slow atomic ballistic (theoretical 1 m/s coolant) flux 10 ** 26 bits/s cm**2
Fast atomic ballistic (theoretical relativistic speed coolant) flux 3 * 10 ** 33 bits/s cm**2
By using ballistic action to run cooler that would also increase the theoretical maximum and enable a transition to a superior architecture when they get it working.
When an architecture shift does happen, they should also try to get reversible computing worked in as well.
Press release from Rochester