Is There a Connection between Adiabatic Switchingand Reversible Computing ?

AND REVERSIBLE COMPUTING? A. Schla er, J. A. Nossek Institute for Network Theory and Circuit Design, Munich University of Technology Arcisstrasse 21, D-80333 Munich, Germany Tel: +4989 289-28 513, Fax: +4989 289-28 504 E-Mail: Andreas.Schla [email protected] Abstract Since in 1973 C. Benett published his paradigm for avoiding energy losses during computation by using reversible computers [3], many attempts have been made to propose electronic circuitry to put this paradigm into practice, most of them using the framework proposed by Fredkin and To oli [5]. Just recently, with the advent of adiabatic switching [1], new attempts have been made to realize such reversible gates by CMOS transistors. However, it is still questionable wether ammending lossy circuitry in order to embed it into a logically reversible system will help to get rid of losses. Note that the aim of reversible computing was never to avoid losses inside the logic gates, but to avoid losses resulting from the destruction of stored information. Since adiabatic switching on the other hand tries to reduce dissipation inside the gates the question arises: \Is there a connection between adiabatic switching and reversible computing?" In this paper it will be shown that the predictions about the power dissipation of adiabatic switches for in nitely large charging times will result in a circuit element, further denoted as the \controlled ideal switch", that turns out to interfere with the second law of thermodynamics and is therefore an overidealization. The interference results from the fact that such a switch allows for the construction of Maxwell's demon. A smaller example hints that it is even possible to destruct information losslessly if an ideal switch is available.