Researchers from MIT and of Google brain joined forces to develop a revolutionary system. The latter not only allows to assess a design proposal at the same time, but also to provide improvements Track. This could significantly increase the number of upgraded configurations.
The new system, called differentiable solar cell simulator, is described in an article in the journal Computer Physics Communications. It was written by Sean mann, MIT student, Giuseppe Romano, MIT researcher, and 4 other individuals from MIT and Google Brain.
An innovative solar cell simulator …
Romano tells us more about the subject. According to him, conventional solar cell simulators predict the percentage of the energy of incoming sunlight, which is effectively converted into electric current. This is called efficiency.
However, this new simulator goes further. Above all else, he predicts the yield. Subsequently, it shows to what extent this yield is affected by one or the other of the input parameters.
“It tells you directly what happens to the efficiency if we make that layer a little thicker, or what happens to the efficiency if we change the property of the material, for example. “
… which is a real eye-catcher for classic simulators
Traditional approaches essentially use a random search for possible variations. But student Mann explains that with this new tool, it is now possible to follow a trajectory of change. Indeed, the simulator indicates the direction to change device. This makes the process much faster. Eh yes ! Instead of exploring all the possibilities, just follow one path. This will directly improve performance.
Advanced solar cells are often made up of multiple layers interwoven with conductive materials. They allow the electric charge to be transported from one to the other. Here, this calculation tool reveals how the modification of relative thicknesses of these different layers will affect the performance of the device.
This simulator is open source and can be immediately used to facilitate research in this area. To turn it on, it will have to be paired with a optimization algorithmor even a machine learning system.
For now, the simulator is based on a version one-dimensional of the solar cell. Even though two-dimensional and three-dimensional configurations will be included later, this is already a significant advance. This alliance between 2 entities of this scale seems to promise even better results.