Master thesis:

Title: Design and optimization of dielectric particle accelerators for healthcare 

Aims: Project DOSE (Dielectric Optical acceleratorS for hEalth) funded by the Italian Ministry for Research (MUR) aims at demonstrating particle acceleration in a compact high-gradient Linear Accelerator (LINAC) based on a hollow-core waveguide operating in W-band, around 100 GHz, where prototypes can be fabricated and experimentally characterized with reasonable cost and complexity. LINACs are widely used in medicine for diagnostic imaging, cancer treatment, sterilization of equipment, irradiation of blood products, and other applications. In this framework, the increasing demand for the development of portable healthcare products pushes for the conception of compact LINACs capable of guaranteeing higher and higher particle energy. Nevertheless, in conventional metallic structures accelerating gradients are limited by material breakdown, power losses in metals, high cost and maximum power of radiofrequency generators. Instead, hollow-core dielectric accelerating structures (HODACs) are characterized by larger breakdown threshold and smaller absorption, thus potentially bringing to a considerable reduction of size and cost with respect to conventional devices operated in S-band.

During this internship a novel HODAC will be conceived. First, the most suitable geometry (e.g. photonic crystal, slot waveguide, metasurface) will be selected based on desired performance and fabrication constraints. Then, thorough numerical simulations based on a combination of commercial and in-house software will be performed in order to design the structure by solving coupled electromagnetic and beam dynamics problems. Last, but not least, optimization algorithms will be used in order to refine the design in view of fabrication and characterization of the final prototype.


Title:  Design of Huygens metalenses with an enhanced lookup table approach 

Aims: Metasurfaces are periodic 2D arrangements of nanostructures enabling unprecedented control of light propagation with subwavelength resolution. In particular, optical metalenses have attracted a lot of interest from the scientific community due to the potential application in integrated optical systems with advanced functionalities (e.g. VR/AR displays, wearable sensors, smart contact lenses). In this framework, precise wavefront shaping requires that phase, amplitude and polarization can be independently controlled by each “meta-atom”. Indeed, a conventional design procedure for a Huygens metalens starts from the creation of a lookup table which provides a systematic mapping between geometrical parameters of meta-atoms and the imprinted phase. Nevertheless, it is well known from the literature that this approach suffers from fundamental limitations, in particular when a rapidly varying phase profile is needed. As a matter of fact, nano-resonators never provide isolated resonances, therefore the standard technique based on a library of independent “pixels” is intrinsically affected by an approximation which hinders an accurate design without resorting to global optimization.

We envisage that metalenses composed of nanocylinders can be modeled as a two-dimensional waveguide array where coupling coefficients between adjacent waveguides can be evaluated by using techniques borrowed from classic waveguide theory. These coupling coefficients could be used to get an enhanced lookup table in order to improve the accuracy of the procedure. During this internship the novel technique for design of metalenses based on a novel lookup table approach will be conceived, and the effectiveness will be tested by comparing the estimated properties with full-wave simulations of a real metalens.


Title: Training of nonlinear metasurfaces with a backpropagation algorithm 

Aims: Deep artificial neural networks have been attracting interest from the scientific community for a plethora of applications in science and engineering. Nevertheless, the stringent energy requirements limiting the scalability of conventional approaches are pushing research toward alternative paradigms, which should be faster and more energy efficient. In this framework, optical neural computing is emerging due to its intrinsic high degree of parallelism and minimal energy consumption. In particular, it has already been demonstrated in the literature that passive neural computing can be performed by exploiting light traveling through a nanostructured device composed of linear and nonlinear scatterers. Nevertheless, up to now backpropagation algorithms have never been applied to train unconventional hardware beyond classic electronics. In this scenario, optical metasurfaces, which are periodic 2D arrangements of nanostructures enabling unprecedented control of light propagation with subwavelength resolution, seem to be the ideal platform to test the feasibility of this approach.

In this internship training of a second-order nonlinear metasurface through a backpropagation algorithm will be investigated. In particular, arrangement and geometry of the nonlinear nanoresonators composing the metasurface will be engineered in order to generate a beam at the second harmonic characterized by well defined features in terms of shape, amplitude and polarization in correspondence of a specified input beam at the pump frequency.  The activity will mainly consist of implementation of the training algorithm followed by testing of carefully selected examples.