FET: EAGER: Physics-Informed AI-Models for Efficient Computation of Stochastic Electromagnetic Fields in Realistic 3D Nano-Scale Optical Interconnects

The advancement of computing, communication, and storage technologies is pivotal for the progress of science and national welfare. This project aims to enhance the analysis and design optimization of on-chip optical interconnect technologies, crucial for achieving ultra-high bandwidth and speeds in modern computing infrastructures. Through the development of an innovative computational electromagnetics (CEM) framework, this project addresses the limitations of current models in accurately simulating on-chip optical channels.

These channels are vital for data transfer in systems ranging from server farms to smartphones, enabling cutting-edge applications such as Cloud Computing, Artificial Intelligence (AI), and the Internet of Things (IoT). The project's outcomes promise to significantly advance technological fields, improving communication, education, healthcare, agriculture, energy efficiency, mobility, and national security, thereby contributing to the United States' competitiveness in key economic sectors.

Furthermore, the project underscores a commitment to educational enhancement and diversity by integrating the project’s key results into graduate curricula and fostering the inclusive recruitment of students from underrepresented groups.

This award focuses on creating a novel computational electromagnetics (CEM) framework that incorporates physics-informed artificial intelligence (AI) to efficiently perform signal/power integrity simulations of realistic three-dimensional, nano-scale silicon-on-insulator (SOI) channels exhibiting random surface roughness. Addressing the challenges of computational expense and limited model fidelity in existing approaches, the project proposes three specific aims: 1) Formulating an energy-conserved Hamiltonian electromagnetic macro-model based on finite-difference time-domain (FDTD) methods; 2) Developing physics-informed AI models as Hamiltonian deep neural networks to expedite the generation and utilization of the macro-model in FDTD; 3) Integrating these AI models into the simulation process to efficiently compute stochastic electromagnetic fields and assess the impact of random surface roughness on the signal/power integrity of realistic SOI channels.

This project is poised to advance and transform CEM and signal/power integrity analysis by enabling accurate, fast, and stable modeling of optical interconnects, facilitating the design of exa-scale bandwidth data-transfer networks. The project’s AI-CEM framework may be extended beyond optical SOI channels to model a broad range of complex optoelectronic devices, thus fostering design innovations in interdisciplinary areas like communication, sensing, and quantum computing.

The AI-enabled Optical Interconnect Designer Tool (AIeOIDT) will be developed and available as open-source software to encourage collaboration and innovation.

This project is jointly funded by the Foundations of Emerging Technology (FET) Cluster of the CCF Division in the CISE Directorate, and the Established Program to Stimulate Competitive Research (EPSCoR).

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.