Angular spectrum-encoded single-shot ultrafast photography.

Capturing transient events on ultrafast time scales demands imaging at up to trillions of frames per second (Tfps). Yet leading methods, including compressed sensing-based photography, time-resolved shadowgraphy, and other active approaches, are limited by bulky optics, high cost, and repeated measurements. Here, we propose angular spectrum-encoded single-shot ultrafast photography (ASUP). ASUP combines the time-wavelength mapping of a chirped-pulse probe with dispersion-encoded angular spectrum information. The latter is realized by a multilayer dielectric thin-film photonic chip, inversely designed using a deep Q-network (DQN) reinforcement-learning framework. This chip performs pixel-level encoding without the need for bulky dispersive optics. An enhanced residual convolutional neural network with Transformer blocks then decodes the measurements to reconstruct high-fidelity ultrafast dynamics. In experiments, ASUP achieves 0.83 Tfps with six frames in a single exposure and captures picosecond laser-induced damage and plasma dynamics in metal films. ASUP delivers performance comparable to state-of-the-art ultrafast photography while remaining compact, highly integrated, and cost-effective. It overcomes key limitations of existing ultrafast imaging and offers a scalable solution for high-speed optical diagnostics, laser-matter interactions, and transient phenomena studies.