Loading...
Simulating vertically & horizontally aligned NR-LEDs via COMSOL and device optimization via artificial neutral networks
Citations
Altmetric:
Date
2025-12
Abstract
Light emitting diodes (LEDs) convert electrical energy into optical radiation when current is injected. Electrons and holes are injected into the system from the n and p contacts respectively. Non-radiative processes such as Auger recombination, exciton-phonon coupling and Förster energy resonance transfer (FRET) occur before the carriers can radiatively recombine at the active region. Additional mechanisms involving LED droop include carrier leakage, carrier delocalisation, phase-space filling and the quantum stark effect. These processes can reduce the internal quantum efficiency of LEDs. It is crucial to identify these pitfalls to minimise the effects of these processes and to assist in device design optimisation to find a more efficient lighting device and save more energy. Artificial neural networks (ANNs) are considered analogous to the functionality of biological neurons in animals. ANNs are interconnected assemblies of units or nodes. The importance of their work is projected into photonics, in areas such as photonic crystals, plasmonic nanostructures, silicon photonic devices as well as metamaterials.
This project intends to achieve the following: Firstly, to produce a simulation model of three Cadmium Selenide/Cadmium Sulfide based LED designs using COMSOL Multiphysics – the monolayer, the vertically aligned nanorod (VNR) and the horizontally aligned nanorod (HNR) models. These models will assist in the characterisation of the current density versus Voltage (J-V) curve, the spontaneous emission recombination rate, energy levels and the electron-hole concentrations of the device. The VNR and HNR models will be aligned with experimental designs. Following this, data from the monolayer model will be used for three neural network (NN) models. These are the feedforward, reverse and tandem NN models. This project seeks to highlight the use of tandem neural networks compared to simpler feedforward neural network (FFNN) models regarding the issue of the prediction of JV curves based on the thickness variation of geometrical layers from the monolayer model.
Supervisor
Description
Publisher
University of Limerick
Citation
Collections
Files
Keywords
ULRR Identifiers
Funding code
Funding Information
Sustainable Development Goals
External Link
License
Attribution-NonCommercial-ShareAlike 4.0 International