Authors:

Xu Lu

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Device simulations play a major role in the 21st Century electronics industry. The continuous miniaturization in the Moore's law era has driven semiconductor R&D engineers towards nanodevices with atomic-scale dimensions, necessitating quantum mechanical simulation tools for reliable predictive modeling. This research focuses on simulations being computationally expensive for larger materials, utilizing the mode space approach for basis reduction. To alleviate the limitation of the high cost of nanodevice simulation in quantum mechanical and atomic resolution, this research aims to de velop a machine learning algorithm based on device-specific basis representations that can apply low rank approximations to reduce the simulation costs. In the initial phase, we generate basis representations from band structure calculations using Purdue's quantum code library, NEMO5, on semiconductor nanowire structures. This data is initially assessed by the human eye for quality. Once a sufficient number of data sets is available, a machine learning model will be trained to reproduce basis representations first with the quantum code and later from the device structure directly. Simulation results indicate that it is feasible to predict accurate data for developing nanoscale transistors. After the ML model is developed and preliminary tested, various optimization techniques are applied to ensure the strength of the machine learning algorithm. Stress tests will be performed to evaluate the validity and to inform any required modifications based on outliers. This research aims to create a sufficiently large database of device- specific basis representations to train an AI to run the quantum code library tools fully automatically and predict basis representations without any tools involved. Further improvements will expand the applicability of our ML algorithms to a broader range of nanoscale devices. Keywords: Machine Learning; NEGF; NEMO5; Mode Space; Semiconductors

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Xu Lu