Electromagnetic and multi-physics modeling and computation Lab

Nowadays, society depends on electronic devices for work, transportation, communication, entertainment, and more. To further improve these devices in terms of a reduced power consumption or increased memory capacity, the core of these devices in the form of integrated circuits (ICs) need to be produced with even smaller structures and finer details. A crucial part within the fabrication process of ICs is accurate monitoring such that production defects can be detected and calibration of the fabrication process can be performed to mitigate these production effects for the next batch of ICs. Failure to do so may result in the production of faulty ICs. We developed computational techniques that are capable of accurately estimating the shape and material properties of these structures, representing the particular details of an IC, during the fabrication process. 

Electromagnetics is the physics behind the interaction of electromagnetic waves and materials and is utilized in some of the greatest technological improvements of the last decade, like mobile communication, MRI, and radar. The behavior of the electromagnetic waves around technology is rather difficult to predict, hence engineers use simulations to visualize the electromagnetics and adjust their designs accordingly. Today, small features in geometry and complicated wave-material interaction result in simulation times in the order days, weeks, or even months. Too long for an iterative design process. We research simulation techniques based on time-domain integral equations that have the potential to reduce simulation times for a wide variety of materials.

Computing the electromagnetic field for a particular geometry and distribution of materials can be quite a task, but what if you want to go the other way: from a resulting electromagnetic field distribution trying to find out what the geometry or the distribution of materials is? The latter type of problem is known as an inverse problem. Many inverse problems are tough to solve and fundamental questions are associated with it, such as if the solution that you look for is unique or even if it exist. In detection problems, we want to figure out the material distribution based on measurements performed on the electromagnetic field. In a design problem we try to first formulate what we want to get as a response and subsequently figure out if that response can actually be obtained or reasonably approximated by a certain geometry or material distribution. 

Bodies of revolution (BoRs) exhibit rotational symmetry around an axis of rotation. The scattering by perfectly conducting or homogeneous bodies can be written in the form of an integral equation along the generating curve of the BoR and as such, a three-dimensional scattering problem reduces to a one-dimensional integral equation per Fourier mode with respect to the direction of rotation. Although long known, efficiently computing this Green function kernel for each mode was an open problem that we recently managed to solve by deriving stable recurrence relations between subsequent Fourier modes.

In electrical engineering, it is necessary to have accurate and reliable insight into the electromagnetic scattering from structures in terms of their material and geometrical properties for applications such as scatterometry for wafer manufacturing or for designing metamaterial lenses. To this end, the spatial spectral Maxwell solver was developed to accurately and rapidly analyze the scattering by complex geometrical dielectric objects in a dielectric layered medium. The core of the spatial spectral method is a Gabor expansion in the plane parallel to the layer interfaces. It allows us to form a domain integral equation that is solved partly in the spatial domain and partly in the spectral domain, thus avoiding the heavy workload of solving tedious Sommerfeld integrals. 

Our lab is thinking about thinking. As many of you know, thousands and thousands of tiny electric currents flow through your brain at any moment. It is how the brain processes information. That means that electromagnetic waves, which induce electric currents, have an influence on the functioning of the brain. This knowledge is currently used to treat patients with brain stimulation techniques, but the full extent of these clinical treatments is not clear. To better understand electromagnetic brain stimulation and to improve these treatments in the long run, our lab models the influence of electromagnetic stimuli on the functioning of neurons. 

Artificial intelligence, and deep learning in particular, have seen a tremendous surge in development over the last decade. This explosion in research has led to huge successes in fields like image processing and speech recognition. Generally, these deep learning approaches don't attempt to explicitly characterise the underlying problem. Instead, they rely on a data driven approach where a network is shown inputs and outputs and corrects itself until it can create the latter from the former. Within computational electromagetics we attempt to solve Maxwell's equations in different physical contexts using an array of different methods. These methods are often agnostic to the underlying specifics of a given EM problem, like for example the precise dimensions of a scatterer, the direction of the incoming wave etc.. So within deep learning, we solve problems using primarily a lot of data whereas in computational EM we find solutions using a direct mathematical approach. It is our goal to bridge these two fields. We can do this by incorporating neural networks directly into our solvers, training them using computational methods to solve the problem and then using these trained networks to improve the time and accuracy when solving similar problems in the future.

Wireless communication systems will have to improve in the future to meet ever increasing demand. The design of these systems requires the use of simulation methods. The performance of existing simulation methods is not good enough for design purposes. We are developing a simulation method for design applications for next generation communication systems (5G/6G). This simulation method is based on LEGO (Linear Embedding via Green’s Operators), which is an in-house developed integral equation domain decomposition technique. An example of a 5G/6G design application is an antenna array system.

In a multiphysics system, two or more physical processes occur and interact in the same configuration.  For instance, an electromagnetic field may lead to heat fluctuations in a structure, which may in turn lead to conductivity variations and shape deformations that affect the electromagnetic field.  Multiphysics computational modeling is very challenging, since it involves time-varying media and often nonlinear interactions over a wide range of time scales.  In addition to using commercial tools with (some) multiphysics functionality, we are developing time-domain computational techniques that are capable of handling specific and more generic multiphysics problems.
 

In collaboration with Draka-Comteq-Prysmian, in the 2000s our computational optics research was focused on nearly all the propagation properties of optical fibers.  We have computed the radiation losses of bent single-mode fibers, devised a mode counting and bracketing scheme for graded-index optical fibers, and developed code for local and global optimisation of single-mode optical-fibers.  In collaboration with the ECO group we studied the size limitations of quantum-dot microdisk lasers, and the selective excitation of silica-based graded-index multimode fibers by a single-mode fiber. From the 2010s, we have collaborated with TE Connectivity and later CommScope on a universal Wilson basis interface for modeling optical fiber excitation and connection attenuation, and the construction of ray-based encircled-flux compliant sources for multi-mode optical fiber launches.