Transforming Metastable Memories: The Nonequilibrium Thermodynamics of Computation

Abstract: Framing computation as the transformation of metastable memories, we explore its fundamental thermodynamic limits. The true power of information follows from a novel decomposition of nonequilibrium free energy derived here, which provides a rigorous thermodynamic description of coarse-grained memory systems. In the nearly-quasistatic limit, logically irreversible operations can be performed with thermodynamic reversibility. Yet, here we show that beyond the reversible work Landauer's bound requires of computation, dissipation must be incurred both for modular computation and for neglected statistical structure among memory elements used in a computation. The general results are then applied to evaluate the thermodynamic costs of all two-input--one-output logic gates, including the universal NAND gate. Interwoven discussion clarifies the prospects for Maxwellian demons and information engines as well as opportunities for hyper-efficient computers of the future.