Student: Magdalina Ustimova

Supervisor: Dr. Somayeh Sadeghi-Kohan

Student: Kai Arne Hannemann
Supervisor: Jan Dennis Reimer

Abstract:

Simulating gate-all-around (GAA) transistors using neural networks is a growing area of research that aims to overcome the limitations of traditional simulation methods for GAA devices. GAA transitors are a promising technology for enhancing the performance of electronic devices, however, their simulation is difficult due to the complex 3D geometry and varying material properties of the device, particularly when it comes to the gate capacitance which is a key characteristic for this type of transistors. Traditionally, GAA devices are simulated using SPICE simulation, however, these simulations can be computationally demanding and may not be able to capture the timing behavior for designs with millions of transistors. Additionally, traditional Switch-level simulators are not able to accurately capture the timing behavior of GAAs. This highlights the need for new, accurate simulation techniques that brigde the gap between Switch-level and elctrical-level simulations. This work thus investigates the feasibility of new efficient neural network-based simulation techniques, which can be run on GPUs, to improve the simulation of GAA transistors.