Rapid optimal work extraction from a quantum-dot information engine

The conversion of thermal energy into work is usually more efficient in the slow-driving regime, where the power output is vanishingly small. Efficient work extraction for fast-driving protocols remains an outstanding challenge at the nanoscale, where fluctuations play a significant role. In this Letter, we use a quantum-dot Szilard engine to extract work from thermal fluctuations with maximum efficiency over two decades of driving speed. We design and implement a family of optimized protocols ranging from the slow- to the fast-driving regime, and we measure the engine's efficiency as well as the mean and variance of its power output in each case. These optimized protocols exhibit significant improvements in power and efficiency compared to the naive approach. Our results also show that, when optimizing for efficiency, boosting the power output of a Szilard engine inevitably comes at the cost of increased power fluctuations.