Dataset used for Binary Classification of Light and Dark Time Traces of a Transition Edge Sensor Using Convolutional Neural Networks

This dataset is the basis of our analysis presented in https://arxiv.org/abs/2509.16243 The data can be accessed through CERN's ROOT software framework, https://root.cern/. Data was measured using a tungsten-based transition edge sensor fabricated at National Institute of Standards and Technology (NIST). The working point of the voltage-biased TES was set to 30% of its normal state resistance and its current was monitored by an inductively coupled SQUID (manufactured by Physikalisch Technische Bundesanstalt; PTB) with 5 GHz gain bandwidth product at a sampling rate of 50 MHz. The TES+SQUID module was operated within a Bluefors SD dilution refrigerator at 25 mK base temperature. We first gathered data by illuminating the TES with a highly attenuated 1064 nm laser source for a total of 5 s. The laser light was coupled to the TES via a HI1060 single mode optical fiber. During this time interval, a total of 4722 pulses above the 10 mV trigger threshold were recorded, where each time trace corresponds to a 200 μs time window with 104 samples (50 MHz sampling frequency). Right after measuring the light pulses, we proceeded to measure the extrinsic background over a period of two days using the same system configuration except for disconnecting the optical fiber from the laser source and sealing its open end with a metallic cover. A total of 8872 background events exceeding the 10 mV trigger threshold were observed.