An information machine with tunable correlations based on colloid particle diffusion

We realize experimentally an information machine converting information to work. Our experimental design is comprised of a colloidal particle diffusing in a microfluidic channel, with a repelling laser based barrier that is moved in feedback to the measured particle position. In a quasi-static mode of operation, the amount of used information is related to the Shannon entropy of uncorrelated steps. We develop a scheme to calculate this information at steady state at fast operation, which induces temporal correlations. We use this calculation to characterize the output power and efficiency of our information machine as a function of feedback cycle time.