Submitted Abstract
The computational power of modern information processing technologies is severely limited by the generation and removal of waste heat. Modern supercomputers devote about 50% of the energy budget to refrigeration. The share of computing and digital information processing in the global energy consumption has been estimated to be as high as 10%, and is only expected to increase. On the other hand, except for a few case studies such as the erasure of a bit, the thermodynamic costs of information processing remain poorly understood. There is thus a pressing need to establish rigorous and concrete thermodynamic considerations which can be used for the development of new information processing technologies. By treating energy and information on equal footing, recent progress in stochastic and quantum thermodynamics offers a rigorous framework to address these questions. Our focus here will be on electronic circuits. Apart from their obvious role in modern information processing technologies, they constitute a platform in which experiments can be conducted with a high degree of control. We plan to construct a consistent thermodynamics of RLC networks, in which resistors may be at different temperatures and circuit parameters may be externally changed in time. This will allow us to design circuits which can operate as heat engines or refrigerators and may be useful for heat management in computing devices. We will then extend our formalism to the quantum regime at low temperatures, enabling the design of experiments involving superconducting circuits. Non-linear circuits will also be considered with the aim to address the thermodynamic costs associated with different switching devices and logic gates. In particular, we want to study the tradeoffs between accuracy (e.g. signal-to-noise ratio), energetic cost and speed. We will also analyze how the thermodynamic cost of networks scales with their size and complexity. The knowledge gained will be of value in the search for new energy-aware and performant information processing technologies. Collaborations with experimentalists are envisaged to build and study some of these circuits.