Conference Agenda

Our work centers on the question how you can use far field diffraction of light to read out or reconstruct sub-nanometer spatial information as relevant for, e.g., metrology in nanofabrication processes, given that you are at the same time free to design a nanophotonic scattering target, free to program the incident wavefront, and able to angle-resolve diffraction patterns over a high NA. We use resonant multiple scattering motifs borrowed from the field of dielectric and Fano-resonant metasurfaces, which provide strong scattering resonances with both strong near fields and characteristic far fields. I will discuss both linear scattering experiments, and experiments in which we measure third harmonic generation diffraction pattterns. The experiments elucidate how metasurface resonances and their interferences optimally transduce tiny near-field perturbations into far field information.

In optical metrology, Fisher information is a central metric that

quantifies the precision that can be achieved in a measurement. For coherent

light, it has been shown that the Fisher information can be written as a Hermi-

tian operator using the scattering matrix of the system. The maximum eigen-

states of this operator are the incident light fields that give the largest possible

Fisher information and therefore give the most precise measurements. Here,

we extend the operator to multiple parameters, representing Fisher information

as a matrix. The measurement precision is related to the inverse of this matrix

by the Cramér-Rao bound, however optimizing the inverse matrix is not trivial.

We consider several scalar functions of this matrix in order to optimize for all

parameters simultaneously, and then corroborate our findings using a scattering

system comprised of coupled dipoles in 2D.

Many researchers consider AI too unreliable for scientific use, but with Simulation Based Inference (SBI) we present a class of ML models that produce comprehensible results that adhere to the high statistical standards of conventional publications. In SBI, the model is trained on simulated samples, which allows the scientist to exert full control over the learned features and only requires to usually much simpler forward measuring process. After the training, the model is used to solve the inverse problem of finding the best parameters given an experimental data point with its variance purely based on the relations defined in the simulator. We present a novel model architecture for the application in nanoimaging and investigate the performance of the method for practical metrology applications.

Artificial neural networks (ANNs) have emerged as powerful tools for imaging through complex scattering media, where conventional approaches fail due to dynamic and unpredictable light propagation. However, the fundamental limits of such ANN-based imaging systems remain largely unexplored. We present a model-free approach to estimate the Cramér-Rao bound (CRB), which sets the ultimate precision limit for parameter estimation, and apply it to evaluate the accuracy of the ANNs trained for imaging through complex scattering media. We compare how well various ANN architectures can localize a reflective target obscured by dynamic scattering. Our approach addresses high-dimensional, non-Gaussian, and correlated data using principal and independent component analysis combined with non-parametric density estimation. Comparing several ANN architectures, we find that convolutional networks with coordinate-aware layers can approach the CRB, achieving near-optimal localization performance. This method provides a general benchmarking tool to assess and guide the design of deep-learning-based imaging systems and opens an opportunity for precision metrology in complex and disordered environments.

In this work, we experimentally demonstrate new types of hybrid supercell metasurfaces that exploit different types of supercells and unit cells in the same design, creating a smooth transition between them. We use this new method to control the phase and amplitude of light at the same time while designing speckle-free holograms.