Alexander Pil
25 August 2021

When it received the first prototypes of its optical communication chips, Effect Photonics suffered from many unusable modules due to surface cracks. To avoid failures and improve the yield, the company from Eindhoven used simulation tools from Comsol Multiphysics, quickly resulting in an impeccable design.

We live in a data-hungry day and age. Everyone wants to stream more videos and download more data. More and more bandwidth is needed to facilitate that high information demand. At some point in time – and we’re getting closer and closer – electronics will reach its limits. New technology has to be developed to avoid a deadlock. Enter photonics. According to Eindhoven-based Effect Photonics, migrating to optical communication might not be easy, but photonics certainly has the future. The company believes it can deliver solutions to satisfy the data hunger and fill the void.

Effect Photonics combines its optical system-on-chip and cost-effective packaging technology to develop and manufacture modules for industry-standard and customized form factors, addressing a 3-billion-dollar market. The spinout from Eindhoven University of Technology takes a platform approach to photonics integration using high-yield building blocks within the wafer. Growing different quaternary alloys of indium phosphide on a single wafer allows for the integration of all the system’s active and passive optical functions within a single chip. The processed chip is then combined with simple packaging designed for high-volume, low-cost manufacturing from the very start. All of Effect’s optical components are tunable, giving users flexibility in provisioning and system deployment. The company aims to deliver these modules for optical fiber networks and 5G infrastructure, amongst other applications.

Effect Photonics cracks
Initial designs of Effect’s optical chips showed many cracks in the surface. Credit: Effect Photonics

Unwanted cracks

Although Effect was able to start its development process with heaps of knowledge from its alma mater, getting the products to the desired performance level was no walk in the park. “After the first prototypes were produced, we saw several cracks in different parts of the surface of our photonic integrated circuit, or PIC,” remembers Aly Abdou, multiphysics engineer at Effect. “These cracks would introduce a high risk of failure in the heart of our module, so we had to find a solution.”

Finding the root cause proved to be challenging, as Abdou relates: “Our PIC is a very dense chip, with many layers and different materials. Features are complex in topology and extremely close together. So it was hard to tackle the problem.” To get to the bottom of the issue, Abdou and his colleagues at Effect used simulation tooling from Comsol Multiphysics.

The first step was to simplify the problem. “Why? Because we wanted to run as many simulations as possible,” explains Abdou. “With 3D models, a lot of topological issues have to be taken into account. You need an expert to sit down, set up the model and define the mesh, before you can run a simulation. That can take hours, if not days. We needed a faster turnaround to be able to test many iterations and to find the best solution quickly.”

Preliminary simulations showed higher stress levels at the same spots where cracks occurred in real life. Effect engineers were on the right track but needed more information, so they extruded the 2D model in the direction perpendicular to the surface. “We wanted to see what the effect of the thickness was,” clarifies Abdou. In the beginning, the top layer was about 0.4 microns thick. “We proposed to double that. Simulations showed us that we then would have reduced the stress by 75 percent. From physical tests, we already knew that the stress levels weren’t very high. Therefore, we opted to increase the thickness to 0.6 microns, reducing the stress by 50 percent. That was more than enough. Less thickness is preferable since it means less production cost. Maybe we could have achieved the same result through a lot of samples and many physical tests, but with simulation in Comsol, it took us hardly any time.”

Effect Photonics first design
Simulations in Comsol unveiled where high levels of stress occurred, matching the locations of the real-life cracks. Credit: Effect Photonics

There’s an app for that

It was a nice result, but it wouldn’t make sense to completely ignore all the underlying layers, as they most certainly influence the stress levels at the surface. “We identified the six main fabrication steps that could cause stress,” explains Abdou. The first step was depositing the passivation layer, a wafer coating layer on which the rest of the structure is built. “We deposited such a layer on a silicon wafer and measured the stress experimentally. In parallel, we simulated the process in Comsol and found the thermal stress at the deposition temperature. Subtracting the two results gave us the intrinsic stress, which we used as the starting value for the layers on top.”

The second layer, built from polymers, is added for planarization, resulting in a more robust base. “Using hygroscopic absorption simulations, we were able to quantify the swelling due to moisture absorption to be 6 percent,” says Abdou. To prevent resulting stresses, Effect covered this layer with a silica coating. “However, cooling down from the silica deposition temperature of about 300 degrees, the polymer shrinks by almost 2 percent, while the silica only shrinks by about 0.02 percent. This mismatch is another source of stress that we need to take into account when we build new layers on top.”

Next came the fabrication steps of electroplating, bonding and another coating. “In Comsol, you can link the stresses from each layer to the next,” Abdou points out. “Of course, this takes a lot of time. And when you introduce changes, you have to start all over.” For Effect, that wasn’t a workable solution. “We wanted a simplified 2.5D model of the top layer.” After exact modeling of the stresses in 3D and matching them experimentally, the topological features were omitted. Effect created a 2.5D model based on the boundary conditions induced by all the layers below.

“It’s not ideal,” admits Abdou, “but it was definitely close enough to see where potential cracks would occur. This simplification allowed us to create an app in Comsol. That way, designers don’t have to come to me to run an intricate simulation to check whether or not their design would survive all mechanical and thermomechanical stresses. They can simply import their design into the app, fill in the required thickness and hit ‘run.’ Within a couple of minutes, they have an answer. With the app, designers can try out a hundred geometries a day, which is of course much better and way more efficient than waiting a few days for a full 3D simulation for every idea they want to investigate.”

Effect Photonics new design
With quick iterations, Effect engineers uncovered the source of the stress and came up with a new and improved design, seriously increasing the production yield. Credit: Effect Photonics

Yield to 99.5 percent

The initial simulations showed many yellow areas where a crack could potentially start. “Sometimes even more than the number of cracks we saw in practice. That’s because once a crack occurs, it releases all the stress in the system,” Abdou explains.

In the simulation of the optimized design, no yellow areas are visible. “It’s beautiful,” states Abdou proudly. “No high stress, no risks. And that showed in real life: no cracks. We went from a yield of less than 60 percent to 99.5 percent of crack-free modules.” With over thirty process changes and iterations, it wasn’t easy getting to that stage. “It would have been very difficult to achieve this experimentally. Simulation has saved us a tremendous amount of time. Years of experimenting were brought back to just a few months of simulating.”

This article was written in close collaboration with Effect Photonics and Comsol Multiphysics.