In my previous entries, I have discussed the imminent limits to the energy efficiency of field-effect transistors, and argued that new digital devices will soon be required that are based on some physical principle other than the semiconductor field effect, and furthermore that are operated in an almost thermodynamically reversible fashion, in order for us to hope to make significant further progress in raw digital power-performance.
In 1996, in an article published in the journal Nanotechnology (vol. 7, p.325), the famous nanotech pioneers Ralph Merkle and K. Eric Drexler proposed and analyzed an ingenious new concept for implementing reversible computation at the nanoscale. The concept, called Helical Logic, involved individual excess electrons (and/or holes) being moved in a controlled fashion along helical wires of semiconducting material by an ambient electric field rotating at GHz frequencies in a high-quality microwave cavity. Electrostatic repulsion between electrons in neighboring paths would divert the electrons along one branch or another of a forked path, depending on whether an excess electron was present in a neighboring path. In the early 1980's, Fredkin and Toffoli at MIT had shown that such a "controlled-fork" or "switch gate" was a universal element for reversible (and irreversible) logic; that is, any digital system could be built out of nothing but such gates.
In their work, Merkle and Drexler carefully (if approximately) analyzed the various mechanisms of energy dissipation in their scheme, and concluded that with sufficiently precise engineering of the electron waveguides, the total dissipation could be made less than 10-27 Joules per reversible operation, at an operation frequency of 10 GHz. This amounts to a power dissipation per logic gate of only 10 attowatts! (An attowatt is a billionth of a billionth of a watt.) In other words, a digital system consisting of 1 billion of these gates running at 10 GHz would still only dissipate 10 billionths of a watt of power. This is roughly 12 orders of magnitude more power-efficient than today's MOSFET technology. Just think of the potentially enormous implications for low-power (and high-performance) digital applications!
Of course, at the time, the scheme was only analyzed on paper and by computer simulation, and much work remains to be done before a practical prototype implementation can be built. For one thing, we can't yet precisely fabricate the required smoothly-curving 3-dimensional interconnects, except perhaps in a macro-scale mockup. However, device physicists have already begun the more detailed theoretical and empirical study of the behavior of branching electron waveguides used as switching devices. These structures are called "Y-branch switches" or "Y-junctions. "The in-depth investigation of these structures actually began as early as 1992, in work by Thomas Palm and Lars Thyln at the Royal Institute of Technology in Stockholm, Sweden. More recently, many other researchers including Erik Forsberg from the same school (now moved to China) have continued the work; Erik is investigating its possible use in reversible computing. For example, see this article in Technology Review.
To date, most of the proposals for computing with Y branch electron waveguides have been relatively unsophisticated from a systems engineering standpoint; in fact, Merkle and Drexler's Helical logic proposal remains the only electron waveguide proposal to date that properly addresses important issues of signal timing and synchronization in an energy-efficient reversible logic framework that also provides an explicit mechanism for adiabatic propagation of signals between (as well as within) the devices. However, a promising direction for future research would be to combine sophisticated design concepts such as Merkle and Drexler's with the empirical experience of workers in the electron waveguide community, perhaps aided by emerging technologies for 3D integration, to produce simple experimental prototypes of Y-branch circuits that can reversibly shuttle around charge packets and do logic with them under the control of an oscillating EM field in a resonant cavity. Such prototypes would enable further empirical study and refinement of the idea, to see we can come close to realizing Merkle and Drexler's extreme energy efficiency predictions in practice. If we can, then the limits of MOSFET technology would be utterly shattered, and a dramatically superior new computing technology would emerge. The potential payoff is vast... "Y not" at least try?
What do you think? Go to TalkBack in the Forum section with your thoughts.
Dr. Michael P. Frank, Ph.D., Assistant Professor,
Florida A&M University and Florida State University,
FAMU-FSU College of Engineering,
Department of Electrical & Computer Engineering,
2525 Pottsdamer St. Room 341, Tallahassee, FL 32310,
firstname.lastname@example.org, (850) 410-6463, cell (850) 597-2046