Neural Network Enhanced DEER Spectroscopy for Optoelectronic Materials

Jason Kestner, Department of Physics

We propose a physics-informed deep neural network (DNN) approach to design shaped pulses that make spin-spin distance measurements with double electron-electron resonance (DEER) spectroscopy more robust against ensemble dephasing. The bottom-up synthesis of new optoelectronic materials requires the ability to precisely characterize structure on the nanometer scale. DEER would be ideal for this except that it is limited by the ensemble spin dephasing time (T2), which may be on the order of tens of nanoseconds or less in many novel optoelectronic materials compared to a few microseconds in the frozen molecular proteins where DEER is currently most frequently used. However, this is not a fundamental barrier. In top-down, gate-defined lateral quantum dot spin qubits, the coherence can typically be prolonged by several orders of magnitude, close to the coherence time (T2), by the use of multi-pulse dynamical decoupling approaches such as the Carr-Purcell-Meiboom
-Gill (CPMG) sequence. While multi-pulse sequences exist for DEER, they generally do not improve performance due to the accumulation of imperfections in the spin rotations that comprise the sequence resulting from non-optimal pulse shapes that do not refocus all resonance heterogeneities. We will address this by using the physics-informed DNN approach to design shaped pulses that enhance sensitivity for DEER and allow access to shorter coherence times. This seed phase will focus on computational modeling, performance optimization, and experimental validation, leveraging the PI’s expertise in neural network pulse shaping and spin qubit control theory and the co-PI’s expertise in coherent spin spectroscopy and spin system modeling.