Fengnian Xia1
Yale University1
In optical sensing, traditionally different types of optical sensors are utilized to measure different physical properties of light. The spectrometer, polarimeter and power meter are used to determine the incident light spectrum, polarization state and power, respectively. It appears that measuring of all the physical properties with a single device is not feasible. Here we introduce a new sensing scheme which challenges this conventional wisdom leveraging the reconfigurable quantum geometric properties in twisted double bilayer graphene (TDBG) and artificial intelligence.<br/>We show that the nonlinear infrared bulk photovoltaic response is largely determined by the quantum geometric properties of Bloch electrons (i.e., Berry connection and curvature). Moreover, due to the tunable quantum geometric properties and the bandstructures in TDBG, the bulk photovoltaic responses in dual-gated TDBG are dependent on the wavelength, the polarization states of the light, and the top and bottom gate biasing voltages of the TDBG device.<br/>Utilizing such unique properties of such dual-gated TDBG devices, we were able to generate a photoresponse map as a function of top and bottom gate biases for each incident light with specific power, wavelength and polarization state. Furthermore, we utilized the measured results to train a convolutional neural network (CNN), which was then leveraged to interpret the photoresponse map measured under the unknow excitation light. We showed that such a trained CNN was able to decipher the physical properties of the unknown incident infrared light measured with a single TDBG device.<br/>Our work represents a novel sensing scheme which is ultimately compact but is able to realize the functions of multiple table-top instruments. Moreover, the results also suggest a new pathway for nonlinear infrared photonics research.<br/><br/>*This abstract is prepared based on our recent publication in Nature 604, 266–272 (2022). Contributions from other authors will be discussed in the presentation.