Nanoscale Systems, Electromagnetic Simulations, and Machine Learning
About Jackson Meeks
Aside from my research at Penn State, I also like to play the guitar and drums, as well as draw and paint. I'm fascinated, as I've mentioned, by artificial intelligence, and I'm also very interested in nanoscale materials built for the purpose of energy demands. I like to program in my spare time, as well as spend time with my wife and daughter.
I've known a lot of people that have mentioned they were bored in their free time, but I can't really remember the last time that happened to me. I'm not sure that I know many people who have as many different interests as me, and I would say that this is both a strength and a weakness -
a strength because it means that I love what I do and am a hard worker - and a weakness because I'm not very good at knowing when it's time to settle down and take a break. However, given the choice, I would much rather be busy and involved in work I love over not having any work to do and all the time in the world. That's one of the many reasons I'm very thankful to do what I do.
Current Projects:
Photodetector Optimization:
I am using COMSOL and Ansys HFSS to simulate and optimize photodetector properties, such as their absorption efficiencies as a function of wavelength and the current output of the device. My goal is to optimize a photodetector cell that works via the Schottky barrier effect, where we have a metal contacting an n-type semiconductor. When a photon of sufficient energy strikes the n-type semiconductor, a “hot” electron is generated that can then travel through the metal and generate a small current. There are many uses for this current, including use in optical devices such as cameras or night vision goggles (when the absorption frequency is lower and more in the infrared spectrum). The reason a metal-semiconductor interface is used instead of a semiconductor-semiconductor interface (such as in a typical diode) is because Schottky barrier devices have a faster response time than typical diodes, and also respond more efficiently to lower energy light.
The goal of my research, as I mentioned, is to optimize the absorption characteristics of the device over multiple wavelengths, as well as the current, with the help of a nanoparticle surface coating. I will use neural networks to aid my search, using a similar technique to that used by Hanakata et al., in 2018, where they used neural networks (NNs) in conjunction with molecular dynamics (MD) to help them find graphene sheet structures with a maximum average yield strain (this is a very good paper, for those who are interested). The way they accomplished this was by taking all of the possible graphene sheet configurations (where different sheets had different sections of the graphene cut out) and randomly selecting 100 of them,
continued:
running MD on them and taking the average strain yield that was derived from the simulations. Then, they trained a convolutional neural network on those results, with the graphene shape as input, and the output being the predicted strain yield. Then, the researchers used the neural network to much more efficiently predict the strain yields of all the other possible configurations of graphene (which would be computationally infeasible with MD). They took the top 100 results of this step, and then ran MD again on the 100 configurations of graphene that gave the highest predicted strain yield, followed by further training the NN on these results. This way, the researchers were able to iteratively hone in on the best configuration of graphene, in terms of a high average strain yield, with the aid of a neural network that significantly decreased their time spent searching.
I am using a similar technique to that described by Hanakata et al., using simulations in COMSOL and HFSS via the finite-element method (FEM) to hone in on photodetector designs (where my parameters are nanoparticle/surface characteristics and material types/thicknesses).
jhm5425@psu.edu | 331-575-9616