Adiabatic Switching, Low Energy Computing, and the Physics of Storing and Erasing Information

Introduction

Work on the physics of computation [1,2] has led to a better understanding of the physical limits of computer technology, but has not yet influenced the way computers are built. The Adiabatic Switching project at ISI is developing digital CMOS computer circuits that use much less energy than conventional circuits, by applying fundamental principles of physics. Conversely, by dealing with real VLSI chips and real computer circuits, we have gained added insights into the energetics of computational processes such as information storage and erasure and combinational logic.

Conventional CMOS digital circuits represent information as charges stored on capacitors[4]. They dissipate energy when actively computing, because changing the value of a bit of information requires converting the signal energy into heat. The idea of adiabatic switching is to instead recycle the signal energy, save it, and later reuse it to represent other information. It turns out that this can indeed be done, but that a small fraction of the signal energy is still dissipated during the recycling process. However, the slower we operate the circuit, the smaller this fraction becomes. In fact, the characteristics of adiabatic CMOS circuits confirm the theoretical arguments of Landauaer[7,8]:

What is interesting is that these properties follow directly from the properties of networks of real CMOS devices, and do not rely in any way on thermodynamics arguments. In particular, the size of the irreducible energy dissipation associated with erasure is determined by the sensitivity of the CMOS switches, and there is no mention of kT, the fundamental limit predicted by thermodynamics[3,7,8].

In this paper, we will sketch the principles of adiabatic switching, and explain what we think the connection is between kT and the sensitivity of the switches.