Welcome to my portfolio! I am Saeed Akbarshahi, passionate about extracting insights from data and using machine learning to solve complex problems. With a background in Computational Physics, I have developed a strong foundation in deep learning, data analysis, predictive modeling, and statistical inference. This portfolio showcases a selection of projects that demonstrate my skills in data cleaning, visualization, machine learning, and more.
Here you will find projects ranging from exploratory data analysis to advanced machine learning models. Each project is contained in its directory with a detailed README explaining the problem, approach, results, and insights.
-
Enhancing Semiconductor Fabrication: A Deep Learning Approach to Lithography Mask Optimization:
Lithography mask optimization is a crucial step in the semiconductor manufacturing process, where the goal is to design masks that can accurately and efficiently create the desired patterns on silicon wafers. Machine learning (ML) methods can significantly contribute to this area by improving the precision, speed, and efficiency of mask design, considering the complex interactions between light and the mask patterns. This project aims to benchmark various machine learning methodologies, comparing their effectiveness in enhancing the precision, speed, and efficiency of lithography mask design amidst the intricate interactions between light and mask patterns.
-
Predicting Semiconductor Defects Using Machine Learning:
This project leverages machine learning techniques to predict patterning defect occurrences in semiconductor manufacturing, aiming to enhance yield and streamline the rework process by utilizing inline measurements of scanner exposure dose and focus. In the pursuit of minimizing yield loss in semiconductor production, we developed a classifier system capable of predicting defect modes before post-etch inspection. By analyzing inline scanner data, our model identifies potential defects immediately after the exposure step, thereby facilitating timely intervention and contributing to the overall efficiency and reliability of the manufacturing process.
-
Economic Turbulence and Tech Giants: A Study of R&D and Operational Spending Impact on Profits:
This project examines how spending on research and office costs affected the profits of big tech companies during the 2008 economic downturn and the COVID-19 crisis. It aims to reveal spending patterns during tough economic times.
-
The Role of Active Learning in Machine-Learned Potential Energy Surfaces:
Recent progress in computational chemistry has been significantly enhanced by machine learning (ML), especially in the development of potential-energy surfaces (PES). This advancement aims to narrow the gap between the high accuracy of ab initio methods and the efficiency of classical force fields (FFs). The innovative approach of these ML techniques is to estimate potential energy directly from chemical structures without the need for fixed chemical bonds or a priori knowledge of interactions, instead leveraging the statistical relationships derived from training data. In this project, we focus on employing active learning methods to investigate the number of samples required to develop reliable machine-learned interatomic potentials for water. Our objective is to achieve precision and efficiency in modeling the unique properties of water.
-
Identification of a Fe-Dependent Optical Mode in CuAl1−xFexO2:
Delafossites are promising candidates for photocatalysis applications because of their chemical stability and absorption in the solar region of the electromagnetic spectrum. For example, CuAlO2 has good chemical stability but has a large indirect band gap, so efforts to improve its absorption in the solar region through alloying are investigated. The effect of dilute alloying on the optical absorption of powdered CuAl1−xFexO2 (x = 0.0−1.0) is measured and compared to electronic band structure calculations using a generalized gradient approximation with Hubbard parameter and spin. A new absorption feature is observed at 1.8 eV for x = 0.01, which red-shifts to 1.4 eV for x = 0.10. This feature is associated with transitions from the L-point valence band maximum to the Fe-3d state that appears below the conduction band of the spin-down band structure. The feature increases the optical absorption below the band gap of pure CuAlO2, making dilute CuAl1−xFexO2 alloys are better suited for solar photocatalysis.
-
Phonon Dynamics in Anisotropic Dilute Delafossite:
Here we report a new versatile approach for the calculation of phonon modes which is applicable to anisotropic, dilute alloys with allowance for a large variety of alloying elements. This approach has significant advantages over previously reported methods, especially for the lattice dynamics of such complex alloys. We use this approach to model the effects of Fe-doping on the vibrational modes in dilute alloys of CuAl1−xFexO2 (x = 0, 0.01, 0.05, and 0.10) delafossite powders. These samples were structurally characterized with X-ray diffraction (XRD) combined with Rietveld refinement to measure their lattice parameters and Raman and FTIR spectroscopies to measure optical phonon frequencies. To compare experimental results from XRD with calculations for lattice parameters, we use a disordered supercell for x in the range 0−0.1. Both results agree well with Vegard’s law. For the phonon calculations, an approach using a disordered supercell is not feasible because it is too computationally expensive. Instead, we developed our weighted dynamical matrix (WDM) approach that uses a straightforward ordered supercell for forceconstant calculations of the CuAlO2 and CuFeO2 parent end points, and combines them using a WDM approach leading to an effective medium for vibrational mode calculation in random alloys. Computationally, when Fe is substituted for Al (increasing x), an increase in the bond length is observed leading to a red-shift in the peak positions in all of the phonon modes vs x, in agreement with the experimentally observed trend.
-
- Programming Languages: Python, SQL
- Data Analysis and Visualization: Pandas, NumPy, Matplotlib, Seaborn, PowerBI
- Machine Learning: Scikit-learn, PyTorch
- Browse through the project directories listed above to view individual projects.
- Each project directory contains a README with an overview of the project, data sources, methodology, results, and key findings.
- Code, datasets (if publicly shareable), and any additional resources are also provided within each project's directory.
I am always interested in learning about new opportunities, collaborating on projects, or simply connecting with fellow data science enthusiasts. Feel free to reach out!
- LinkedIn: www.linkedin.com/in/saeed-akbarshahi
- Email: [email protected]
Thank you for visiting my data science portfolio. I look forward to connecting with you!