Kvantmekanisk Beskrivning av Fullständiga Halvledaranordning
- Diarienummer
- FFL21-0129
- Projektledare
- Wiktor, Julia
- Start- och slutdatum
- 220801-271231
- Beviljat belopp
- 15 000 000 kr
- Förvaltande organisation
- Chalmers University of Technology
- Forskningsområde
- Materialvetenskap och materialteknologier
Summary
The project concerns first-principles modeling of materials and interfaces that constitute semiconductor micro- and nanodevices. The work will build on a modeling methodology that I have developed, that considers materials in a realistic way, including imperfections, intrinsic effects, and temperature. The main objective is to decrease the computational cost of said approach, yielding it feasible to apply it in the studies of a device comprising multiple functional layers and interfaces, within the timeframe of a single project. To achieve this objective, I plan to train an Artificial Neural Network to replace the most computationally demanding step of my methodology - Ab-initio Molecular Dynamics simulations of temperature effects. I will use an approach that has been shown to work on simpler problems - the force-field method, but, capitalising on my expertise, apply it to a much more demanding systems containing defects and charge-lattice interactions (polarons). Once the ANN model is implemented in my methodology, it will allow for gaining insights into crucial local phenomena that partially govern the functioning of microdevices, within a fraction of time that is now necessary to carry out these calculations. I plan to apply this newly developed method to perovskite solar cells and tunnel field effect transistors, in collaboration with experimentalists. If successful, this methodology will constitute a great aid in microdevice optimisation.
Populärvetenskaplig beskrivning
Computational modeling is a way to predict properties of materials without ever forming them in the laboratory. We can also use it to understand the behaviour of known materials. I use first-principles, or ab-initio atomistic models, that consider matter at the scale of single atoms. To create such a model, we only need the positions of atoms, and their types. By using the Schrodinger equation, we can then simulate how the electrons around these atoms will behave, and so gain insights into different properties of the studied materials which the electrons govern. Depending on the material, this can be light absorption, electrical or heat conductivity etc. By learning about these properties, we can then decide whether a given material is suitable to be used in a solar cell or a transistor, for example. But the more precise results we aim to obtain, the more sophisticated the model should become. In my studies, I strive to recreate the materials as they are found in nature: imperfect. I introduce defects in the lattice, for example remove one atom, and consider what happens to electrons when they are in their vicinity. I also introduce temperature effects in my calculations. At room temperature, atoms in a material will vibrate, creating a level of disorder, which will also affect how the electrons behave. By taking into account all these effects, I make my models precise, but costly in terms of computing power, which in turn means the calculations are slow. In the frame of this project, I plan to significantly decrease this cost by using an artificial neural network. This network can be trained on a set of results generated from my current approach, to then be able to predict properties of unknown materials on its own. This Machine Learning approach has been used in the past to replace first-principles modeling, but never on such complex systems that I study. Once this aim is achieved, my calculations will be up to ten times faster than they are now. I will capitalise on this significant increase in speed, by applying this new approach to studying complete microdevices, which are composed of a number of very thin layers of different materials. As first examples, I plan to study a perovskite solar cell, and a tunnel field-effect transistor, which are emerging technologies that use advanced engineering techniques. Layers in these devices are as thin as a single atom, hence a fast atomistic modeling approach is necessary to speed up their development.