This study employs Density Functional Perturbation Theory (DFPT) to elucidate the influence of biaxial strain on the optical Raman peaks and electronic band gap of monolayer WS2. By analyzing phonon and electronic dispersions, this work establishes a direct correlation between induced strain and the modifications that arise in the band gap and optical phonon modes, providing a strategic framework for tailoring material properties. The computational approach involves varying the a and b lattice parameters from -2% to 2% in 0.5% increments. Self-consistent field (SCF) and phonon calculations are conducted to derive phonon frequencies and construct band structures along the lattice’s high-symmetry paths. Using a 2x2x1 WS2 supercell, the findings indicate that biaxial strain reduces the direct band gap at the K-point of the Brillouin Zone. Furthermore, the A1g and E2g Raman modes exhibit pronounced blue and red shifts, respectively, under tensile strain. These results deliver nuanced insights into the strain-dependent electronic and vibrational behaviors of WS2, paving the way for the development of next-generation optoelectronic and photonic devices with highly tunable characteristics.

I took part in a research fellowship funded by the National Science Foundation in which I was given the opportunity to work in the Materials Data Science field. Under the supervision of one of my professors, I worked on a computer vision project in which I was tasked with using Instance Segmentation to improve on current methods of detecting small, spherical particles in Scanning Electron Microscope images of a gas-atomized nickel superalloy powder. In order to label the ground truth images, I used VGG Image Annotator to manually map out thousands of particle instances across 5 images. I then trained the model using a Meta AI implementation of Mask R-CNN to achieve a 6% increase on detection recall. I presented my findings at an undergraduate research symposium to other students and researchers in the Materials Data Science field.

Physicians are required to be almost 100% accurate during their jobs, especially when it comes to making final diagnoses. When you have something as serious as a cancer diagnosis, physicians are often under a lot of pressure to give the correct diagnosis, sometimes with very limited time and information. Breast cancer is the most common disease affecting women in many parts of the world, which has led to efforts to be able to detect tumors in the breast as early as possible. In many sub-Saharan African countries, we see issues associated with the technology used for classification of these tumors. While breast cancer is more common is developed countries (> 80 cases per 100,000 people) than it is in developing countries (< 30 cases per 100,000 people), the mortality rates in developed countries is much lower because of advanced research and technology. My approach to this problem is to train a neural network to be able to classify these tumors with very limited data, as if we were trying to simulate the conditions of a physician working with very little data themselves. The dataset that I am working with only contains 9 parameters given a score from 1-10 based on the severity. In a real setting, there are several different variables that can be generated from studying a tumor and using only 9 has shown yield very accurate results. Most research in tumor classification uses images of the tumors as part of the model; however, for my research I wanted to focus more on numerical data, as I believe this will be a better simulation of having limited resources.