Computer Simulations for Better Solar Cells

An illustration of the 2D perovskite material that was studied by the researchers. The yellow parts illustrate the linker molecules while the purple and pink parts show the perovskite layer. (Image: Chalmers University of Technology | Julia Wiktor)

More stable and efficient materials for solar cells are needed in the green transition. So-called halide perovskites are highlighted as a promising alternative to today’s silicon materials. Researchers at Sweden’s Chalmers University of Technology have gained new insights into how perovskite materials function, which is an important step forward.

Halide perovskites is the collective name for a group of materials that are considered very promising and cost-effective for flexible and lightweight solar cells and various optical applications, such as LED lighting. This is because many of these materials absorb and emit light in an extremely efficient way. However, perovskite materials may degrade quickly, and in order to know how best to apply these materials, a deeper understanding is required of why this happens and how the material functions.

Within the perovskite group, there are both 3D and 2D materials, the latter often being more stable. Using advanced computer simulations and machine learning, a research team at the Department of Physics at Chalmers University of Technology studied a series of 2D perovskite materials and gained crucial insights into what influences properties. The research results are presented in an article in ACS Energy Letters.

“By mapping out the material in computer simulations and subjecting it to different scenarios, we can draw conclusions about how the atoms in the material react when exposed to heat, light, and so on. In other words, we now have a microscopic description of the material that is independent of what experiments on the material have shown, but which we can show to lead to the same behavior as the experiments. The difference between simulations and experiments is that we can observe, at a detailed level, exactly what led to the final measurement points in the experiments. This gives us much greater insight into how 2D perovskites work,” said Professor Paul Erhart, research team member at Chalmers University of Technology.

Using machine learning has been an important approach for the researchers. They have been able to study larger systems, over a longer period, than was previously possible with the standard methods used just a few years ago.

“This has given us both a much broader overview than before, but also the ability to study materials in much more detail. We can see that in these very thin layers of material, each layer behaves differently, and that’s something that is very, very difficult to detect experimentally,” said Associate Professor Julia Wiktor, member of the research team.

Julia Wiktor, Associate Professor, Department of Physics, Chalmers University of Technology, Sweden. (Image: Chalmers University of Technology | Anna-Lena Lundqvist)

2D perovskite materials consist of inorganic layers stacked on top of each other, separated by organic molecules. Understanding the precise mechanisms that influence the interaction between the layers and these molecules is crucial for designing efficient and stable optoelectronic devices based on perovskite materials.

“In 2D perovskites you have perovskite layers that are linked with organic molecules. What we have discovered is that you can directly control how atoms in the surface layers move through the choice of the organic linkers and how this affects the atomic movements deep inside the perovskite layers. Since that movement is so crucial to the optical properties, it’s like a domino effect,” said Paul Erhart.

The research results provide greater insight into how 2D perovskite materials can be used to design devices for different applications and temperature variations.

“This really gives us an opportunity to understand where stability can come from in 2D perovskite materials, and thus possibly allows us to predict which linkers and dimensions can make the material both more stable and more efficient at the same time. Our next step is to move to even more complex systems and in particular interfaces that are fundamental for the function of devices,” said Julia Wiktor.

Here is an exclusive Tech Briefs interview, edited for length and clarity, with the research team.

Tech Briefs: How did this project come about? What was the catalyst for your work?

Research Team: We were already working on a closely related family of so-called bulk perovskites with very intriguing atomic scale dynamics. In these 2D variants these effects are even more pronounced and can be exploited to even bigger effect to modify the electronic and optical properties of these materials. This opens up countless possibilities to optimize and tune these materials for applications not only in solar cells but also for example in lighting.

Paul Erhart, Professor, Department of Physics, Chalmers University of Technology, Sweden. (Image: Chalmers University of Technology | Anna-Lena Lundqvist)

Tech Briefs: What’s the biggest technical challenge you will face when using computer simulations/machine learning in this manner?

Research Team: These materials have a complex atomic structure, and the atoms interact with each in very subtle ways. This means that traditional approaches for modeling their dynamics are computationally extremely and impractically expensive. Here, we therefore used machine learning techniques to avoid these bottlenecks. In fact, our models are so efficient that they can be run on a graphics card intended for playing computer games as opposed to a data center.

Tech Briefs: Can you explain in simple terms the process/how it works?

Research Team: We start with an approximate picture of how the atoms in the material are distributed and the chemical bonds that are present, i.e., the connections between atoms. Then, we use quantum mechanical calculations, which are very accurate but also computationally very demanding, to compute the energy and the forces between the atoms. We repeat this process for a number of other configurations.

We use these data to train a machine learning model, specifically a so-called artificial neural network. When presented a new arrangement of atoms, this network can then predict its energy and forces using a miniscule fraction of the computational effort needed for the quantum mechanical calculations. Once we have a machine learning model that describes the interactions, we can use it to follow the trajectory of each atom in the system over time, which in turn allows us to predict various properties of the material.

Tech Briefs: Any further research/work/etc. on the horizon? If not, what are your next steps?

Research Team: We are currently expanding our approach to even more complex systems and other properties such that we can expand our predictions, guide experimental studies into these materials, and ultimately design materials with not only optimal properties but even new functionalities.

Tech Briefs: Do you have any advice for engineers/researchers aiming to bring their ideas to fruition, broadly speaking?

Research Team: Be curious and have fun at what you are doing.