Abstract
Deciphering the composite information within a light field through a single photodetector, without optical and mechanical structures, is challenging. The difficulty lies in extracting multi-dimensional optical information from a single dimension of photocurrent. Emerging photodetectors based on information reconstruction have potential, yet they only extract information contained in the photoresponse current amplitude (responsivity matrix), neglecting the hidden information in response edges driven by carrier dynamics. Herein, by adjusting the thickness of the absorption layer and the interface electric field strength in the perovskite photodiode, we extend the transport and relaxation time of carriers excited by photons of different wavelengths, maximizing the spectrum richness of the edge waveform in the light-dark transition process. For the first time, without the need for extra optical and electrical components, the reconstruction of two-dimensional information of light intensity and wavelength has been achieved. With the integration of machine learning algorithms into waveform data analysis, a wide operation spectrum range of 350-750 nm is available with a 100% accuracy rate. The restoration error has been lowered to less than 0.1% for light intensity. This work offers valuable insights for advancing perovskite applications in areas such as wavelength identification and spectrum imaging.