Minimum energy for computation, theory vs. experiment

How much energy is required to compute a bit of information? As the end of the CMOS scaling era approaches, this question becomes ever more important. A group at the Semiconductor Research Corporation (SRC) has postulated a limit at k_BT ln(2) for charge-based computing, and suggested that other state variables are necessary. Our work shows that the SRC analysis is flawed, and investigates devices and circuits that will dissipate energy only when information is destroyed.