Modeling Nitrogen Diffusion Under Extreme Conditions for Hyperdoped Silicon Using Machine Learning
Discipline: Mathematics and Statistics
Session: 2
Room: 4 - Hanover F
Rogelio Lopez - Elizabeth City State University
Co-Author(s): Qimora Mason and Abdennaceur Karoui
Nitrogen diffusion in semiconductors is essential for the development of next-generation optoelectronic devices. N-type hyperdoping introduces intermediate energy bands in the silicon bandgap, which enhances optical absorption and increases the efficiency of infrared photodetectors and solar cells. However, nitrogen diffusion in defect-rich silicon under extreme doping conditions is not well understood due to the complex interactions of vacancies and interstitial defects in forming chemical complexes and facilitating the diffusion of these species. Traditional numerical methods often struggle to accurately capture these nonlinear, defect-mediated mechanisms, highlighting the need for innovative approaches.
We hypothesize that a new hybrid computational framework integrating molecular dynamics (MD), machine learning (ML), and finite element analysis (FEA) can yield accurate diffusivity values for the various nitrogen species in hyperdoped silicon. This method utilizes MD simulations to identify defect-mediated diffusion pathways and calculate activation energies for the diffusion of different nitrogen species. The ML component employs neural networks to predict diffusivity across a broad range of thermodynamic conditions, capturing complex, nonlinear interactions. ML aids in processing a wider array of parameters than traditional MD calculations and can analyze saturated solid solutions with diverse chemical complexes. FEA integrates these predictions to solve diffusion equations under realistic boundary conditions relevant to ion implantation used in hyperdoping, enabling macroscopic validation of the diffusivity results.
Preliminary MD findings suggest that vacancy dynamics are the dominant factor in nitrogen diffusion at processing temperatures of 800–1200°C, primarily in the form of VN and N2 interstitials, while VN2 plays a lesser role. The diffusivity is significantly enhanced at these temperatures, highlighting the importance of lattice defects in nitrogen transport and providing insights for optimizing hyperdoping strategies in optoelectronics and quantum computing.
In conclusion, this hybrid framework offers a scalable approach to obtaining realistic and accurate diffusivity solutions, overcoming the limitations of traditional methods that only consider ideal crystals. These findings emphasize the critical role of lattice defects in nitrogen transport and their implications for hyperdoping strategies.
Future research will extend this framework to multicomponent doping systems, exploring nitrogen-oxygen interactions and validating predictions through various experimental characterizations, including Fourier Transform Infrared (FTIR) spectroscopy and Raman spectroscopy.
Funder Acknowledgement(s): The authors thank the NNSA funding agency Award# DE-NA0004112, the NSF EiR Award number 2401243, and the Consortium for Nuclear Security Advanced Manufacturing enhanced by Machine Learning.
Faculty Advisor: Abdennaceur Karoui, abkaroui@ecsu.edu
Role: Research Assistant/Presenter