PhD - Leuven | More than two weeks ago
Flat optic devices, often metasurfaces with unprecedented functionalities, made with mass manufacturable technologies have the potential to replace bulky optical systems. This was already demonstrated in combination with many applications such as for functionalized VCSELs, improved CMOS imagers, and high-quality MR/AR displays. However, the design of these flat optic devices often relies on many simulations for device optimization. As each of these simulations relies on solving full-vectorial Maxwell’s equations to include near-field coupling in the structures, the overall optimization effort could ultimately require unrealistic computation resources to finish. Recently, a new method was proposed to drastically accelerate the device optimization via physics-augmented deep-learning, in layman terms: teach an algorithm to gain intuition at solving Maxwell’s equations based on topology. So far, this approach has only been implemented to very simple demonstrator models. It is our goal to adapt this method to achieve to address two application domains. On one hand to optimize towards a target functionality the light emission from integrated light sources with a to optimize flat optic structure. On the other hand, to implement smart optical sensors that achieve on-chip optical neural networks to avoid the energy consumption otherwise required for off-chip data-processing.
The goal of this project is two-fold. In a first phase, the target is to co-design metasurfaces with integrated light sources such as VCSELs or uLED. Hereto, it will be needed to use physics-augmented deep-learning to drastically accelerate the optimization speed, while including near-field effects between the light source and the flat optics. Depending on the available technology, a functionalized VCSEL or uLED will be designed for fabrication and characterization.
In a second phase the same simulation method will be used to design multilevel metasurfaces which represent neural networks with the purpose of e.g. pattern recognition. Hereto, a device stack will have to be proposed to allow for flexible neural network realization. An actual optical neural network will then be designed and fabricated. Finally, the fabricated devices will be characterized in a way to demonstrate compatibility with CMOS image sensors.
This PhD project combines optical design, machine learning and device characterization.
You are a highly motivated student, with background in nano-engineering, physics, material science, electrical engineering, or related. You have an interest in nanofabrication and flat optics applications, both from design and characterization side. It is expected that you will present results regularly. You are a team player and have good communication skills as you will work in a multidisciplinary and multicultural team spanning several imec departments. Given the international character of imec, an excellent knowledge of English is a must.
Required background: Engineering Science, Computer Science or equivalent.
Type of work: 60% simulation, 20% literature, 20% experimental
Supervisor: Jan Genoe
Co-supervisor: Xavier Rottenberg
Daily advisor: Bruno Figeys
The reference code for this position is 2023-128. Mention this reference code on your application form.