AlGaN-based deep ultraviolet laser diodes (LDs) have attracted considerable interest because of their diverse applications over the last two decades. The optimization of DUV LDs is essential to advancing high-efficiency photonic devices. However, traditional simulation tools are computationally intensive and slow, presenting challenges for iterative development of optoelectronic devices. We propose an AI-driven approach that leverages machine learning (ML) and explainable AI (XAI) to accelerate the design process of DUV LD and enhance understanding of the correlation between key LD performance parameters. Our methodology involves training ML models on a dataset of DUV LD design parameters to evaluate each model’s predictive accuracy. We also integrate XAI to assess input feature importance such as material composition and thickness of epilayers. This framework provides predictions for laser output power (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10825_2025_2352_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="30" /> </InlineMediaObject> <EquationSource Format="TEX">$P_{\text {out}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>P</mi> <mtext>out</mtext> </msub> </math></EquationSource> </InlineEquation>), laser threshold current (<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10825_2025_2352_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="20" /> </InlineMediaObject> <EquationSource Format="TEX">$I_{\text {th}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>I</mi> <mtext>th</mtext> </msub> </math></EquationSource> </InlineEquation>), and optical confinement factor (<InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10825_2025_2352_Article_IEq3.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="12" /> </InlineMediaObject> <EquationSource Format="TEX">$\Gamma$</EquationSource> <EquationSource Format="MATHML"><math> <mi mathvariant="normal">Γ</mi> </math></EquationSource> </InlineEquation>) with R2 values of 73%, 71%, and 80%, respectively, with the best-performing model that is extreme gradient boosting. This model substantially reduces the computational time required for optimum design iteration. These results demonstrate that our AI-based approach outperforms traditional methods in speed and resource efficiency, providing actionable insights into design parameters that align with physical mechanisms. This work establishes a resource-efficient AI framework that accelerates the development cycle of high-performance LDs.

AI-powered deep ultraviolet laser diode design for resource-efficient optimization

期刊论文
发表日期：

2025年7月2日

查看全文

Asima Sarwar,
Muhammad Usman,
Masroor Hussain,
Khurram Khan Jadoon,
Tareq Manzoor,
Shazma Ali

摘要

AlGaN-based deep ultraviolet laser diodes (LDs) have attracted considerable interest because of their diverse applications over the last two decades. The optimization of DUV LDs is essential to advancing high-efficiency photonic devices. However, traditional simulation tools are computationally intensive and slow, presenting challenges for iterative development of optoelectronic devices. We propose an AI-driven approach that leverages machine learning (ML) and explainable AI (XAI) to accelerate the design process of DUV LD and enhance understanding of the correlation between key LD performance parameters. Our methodology involves training ML models on a dataset of DUV LD design parameters to evaluate each model’s predictive accuracy. We also integrate XAI to assess input feature importance such as material composition and thickness of epilayers. This framework provides predictions for laser output power ( $P_{\text {out}}$ $P_{out}$ ), laser threshold current ( $I_{\text {th}}$ $I_{th}$ ), and optical confinement factor ( $\Gamma$ $Γ$ ) with R² values of 73%, 71%, and 80%, respectively, with the best-performing model that is extreme gradient boosting. This model substantially reduces the computational time required for optimum design iteration. These results demonstrate that our AI-based approach outperforms traditional methods in speed and resource efficiency, providing actionable insights into design parameters that align with physical mechanisms. This work establishes a resource-efficient AI framework that accelerates the development cycle of high-performance LDs.