学术文献库

Cryogenic characterization of transition-edge sensor (TES) bolometers is a time- and labor-intensive process. As new experiments deploy larger and larger arrays of TES bolometers, the testing process will become more of a bottleneck. Thus it is desirable to develop a method for evaluating detector performance at room temperature. One possibility is using machine learning to correlate detectors' visual appearance with their cryogenic properties. Here, we use three engineering-grade TES bolometer wafers from the production cycle for SPT-3G, the current receiver on the South Pole Telescope, to train and test such an algorithm. High-resolution images of these TES bolometers were captured and relevant features were calculated from the images. Cryogenic performance metrics, including a detector's ability to tune and superconducting parameters such as normal resistance, critical temperature, and transition width, were also measured. A random forest algorithm was trained to predict these performance metrics from the visual features. Analysis of the images proved highly successful. While the ability of the random forest algorithm to predict cryogenic features was limited with the chosen set of input image features, it is possible that an increase in data volume or the addition of more image features will solve this problem.