MSc. Project – Solar AI

We all know about the dire need to switch from carbon based fuels to renewable energy sources. Here at EPFL a new material is being developed that will help realized this transition – Perovskite. Since it’s discovery as a solar cell material in 2009, perovskites have developed rapidly. In just nine short years they have moved from 4% to 23% efficiency, making them competitive with silicon solar cells. So what makes perovskite solar cells special? They can be fabricated just by mixing some chemicals together, making them ten times cheaper than silicon cells. However, this material still faces some challenges until we can get it to commercialization. One of these challenges is quality assurance. Currently we need to test each solar cell by connecting it to a solar simulator and measuring its output power. This process is slow, and may act as a bottleneck for production. We think that the quality of a cell can be tested optically. By using machine learning techniques on images of these cells, is it possible to make predictions about their behavior? This is what we want you to help us find out!

The Project

Part I
To start learning, we first need to collect labeled samples. Luckily, perovskites have been a very hotly researched topic, so lots of data exists. However, there is not a central database for this data. Thus, the first part of the project will be to create a data mining program that can obtain images and parameters of already reported perovskite solar cells. We will also be setting up an imaging station to make high throughput images of our perovskite solar cells. You will then get to create your training data using samples created daily in our lab, and get to explore the data.

Part II
Once a sufficient amount of data has been obtained, it’s time to release the algorithms! What sort of properties can you infer about the cells given the data you’ve collected. Are there indications of
efficiency or lifetime?


Supervisor: Brian Carlsen ([email protected])
Group: Prof. Anders Hagfeldt / LSPM
Time Frame: 3 – 9 months
Background: Data science, and some hacking. The data we need already exists, we just have to get our hands on it and coax it into a usable format. To do this you should have a basic idea of how to interact with Database APIs and be ready to hack around some of their limitations (all white hat, of course). You should also have an interest and basic knowledge in ML/AI.

MSc. Project – Degradation of Perovskite Solar Cells

A new material that can act as a solar cell, called Perovskite, started to be developed at EPFL only a few years ago. These cells offer a cheap alternative to silicon solar cells, and work just as well, except for one problem: They don’t last very long. While silicon solar cells can last for 25 years, perovskite solar cells are lucky to make it one year. Even though we are still quite a ways from the 25 year goal, there is much promise for these cells! The cells are made up of a combination of organic an inorganic materials, and it is thought that the organic molecules are the main cause for their quick demise. Some think that the organic material evaporates from the cell, while others think environmental oxygen and water are absorbed in to the cell.

The Project

To test these controversial hypotheses we will track the weight of different types of perovskite solar cells over time. We will also attempt to setup a vacuum chamber to perform spectroscopy of the
atmosphere and the cells to track their chemical composition. Using this data we will draw conclusions on the internal processes occurring within the cells leading to their degradation.


1. Set up the Microbalance: Your first task will be to set up the microbalance we will use to measure the cells’ weights.
2. Set up the Vacuum Chamber: Sounds easy, right? Well, things always get a bit more interesting when you start thinking about the details.
3. Collect and Analyze Your Data: After performing all your measurements it’s time to draw a conclusion. Can we infer anything from the data about what is causing the degradation within the cell?


Supervisor: Brian Carlsen ([email protected])
Group: Prof. Anders Hagfeldt / LSPM
Time Frame: 3 months
Background: Basic chemistry or physics, and very creative thinking. Because nobody really knows what processes are occurring, it will be up to you to analyze the data and propose the most likely scenario.