OnScale is happy to announce a newly released Electromagnetic (EM) solver capability, now available on the OnScale Cloud platform. In this blog post we describe the new Electromagnetic (EM) solver in OnScale, designed specifically for next generation antenna array optimization, photonic waveguides, radar for ADAS systems, and many other use cases.
Watch full webinar here!
Optical waveguide examples and results
Let’s dive right in to an Electromagnetic (EM) example and show you how OnScale’s new solvers can be used. We’ll start with optical waveguides. One of the reasons we've targeted this application is because there is a huge amount of renewed interest in optical waveguides both in communications and sensing, for a diverse range of applications. Simulating these devices is extremely challenging, since we're talking about wave propagation over many tens or hundreds of wavelengths. Which typically involves problems somewhere in the tens of millions, potentially billions of degrees of freedom.
Image above shows a waveguide with a silicon guide on a silicon dioxide substrate, with a wavelength of around 1550 nanometers. We are interested in finding out is how this transmission varies with wavelength. We also want to know how it is affected if we put a bend radius on our wave. If we go through a 90 degree corner - how will that affect the curve off transmission core fashion versus wavelength?
We are using a straight waveguide as an example. The graph above shows a transmission coefficient on the y axis, plotted against the wavelength on the x axis in nanometers.
Watch our full webinar on-demand to see all of the results, which are generated from our new time domain electromagnetics solver in OnScale!
The table above gives a summary of the run time statistics for a 3D EM Simulation of Optical Waveguide. The size of this model is around 8 million degrees of freedom. We executed this in time domain for 3,700 times steps. This model requires a significant amount of calculation, but with OnScale we can run these simulations in only four minutes using a very insignificant amount of ram (150 megabytes). However if you were to use a legacy package and wanted to run these simulations sequentially, 500 simulations for a Monte Carlo Study would take around 42 hours for the straight waveguide or even 180 hours for the bent waveguide. Being able to run simulations simultaneously with OnScale can turn around these studies in 5 and 13 minutes respectively.
Electromagnetic Antenna Examples
We live in a very connected world and the design of RF antennas is a very challenging one. There is a lot of time form factor restrictions that in some cases determine the type of antenna has to be used, in addition to very aggressive performance demands. Let’s take a look at the design of a 6 gigahertz patch antenna array which can be used for 5G back call applications in RF communication applications.
The image above shows a copper patch above a substrate with a ground plane beneath it, that is excited by a lumped port. From these simulations we can gain some very fundamental insights of this antenna which can be seen below.
Above we can see the electrical impedance plotted at real and imaginary vs frequency. We can see from the results that the antenna has a strong resonance of around 6 gigahertz. We can also see the beam patterns of the device.
OnScale allows you to run these basic patch simulations in under 5 minutes! Similar to the previous results, we are able to turn around a 400 simulation design sweep in 5 minutes, as opposed to the 33 hours that it would take to go through these simulations sequentially.
Above is an example of a 4x4 patch antenna based on the 6 gigahertz design that we previously described. As we can see in the time domain video showing Electric Field in Z which is out of plate, we can see that we are able to apply arbitrary drive conditions to each patch which allows us to simulate beam forming.
Click here to watch our full webinar on demand to see antenna array vs single patch results.
We can use the arbitrary phasing to look at beamforming performance to evaluate the resonance of the array to custom beamforming algorithms. Here we have applied some basic phased delays in order to steer the beam from 0 degrees, to the 30 degree angle shown at the bottom right in the image above. As you will be able to see from these images, the beam is deflected and has steered off.
Here we can see the 3D data behind the cross sectional graphs, visualized with MATLAB. This is a much larger simulation than we have shown previously, it is around 30 Million Degrees of Freedom (DoF). In OnScale the simulation runs in 10 minutes and uses only 1.35 GB of ram, and would be much slower, if not impossible to solve with other tools available. This OnScale simulation only required 4 core hours, allowing for a lot more potential to accelerate these simulations.