Scholarlink

<p>We present a machine learning approach for the inverse design of organic semiconductor materials from benzene and thiophene-based polycyclic aromatic compounds (PACs). Inverse design is an efficient approach to materials discovery that aims to design materials with preset properties. However, it is complex due to the non-uniqueness and nonlinearity of property-to-structure relationships. We demonstrate the potential of this approach through the inverse design of PACs to achieve target HOMO-LUMO gaps, a key property for organic semiconductors, ranging from 1.36 eV to 4.37 eV with an error of 0.15 eV within Density Functional Theory uncertainty. The model uses goal-conditioned reinforcement learning with chemical domain knowledge, allowing addressing design goals directly. To incorporate practical aspects such as chemical accessibility, the model can include soft constraints, such as minimizing ring count to favor smaller structures. Thus, our framework addresses key inverse design challenges while allowing prioritization of more optimal or diverse candidates.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Inverse Design Using Goal-Conditioned Reinforcement Learning for Organic Semiconductor Materials from Benzene and Thiophene-based Polycyclic Aromatic Compounds

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月12日

查看全文
  • Tri M. Nguyen,
  • Thanh N. Truong

摘要

We present a machine learning approach for the inverse design of organic semiconductor materials from benzene and thiophene-based polycyclic aromatic compounds (PACs). Inverse design is an efficient approach to materials discovery that aims to design materials with preset properties. However, it is complex due to the non-uniqueness and nonlinearity of property-to-structure relationships. We demonstrate the potential of this approach through the inverse design of PACs to achieve target HOMO-LUMO gaps, a key property for organic semiconductors, ranging from 1.36 eV to 4.37 eV with an error of 0.15 eV within Density Functional Theory uncertainty. The model uses goal-conditioned reinforcement learning with chemical domain knowledge, allowing addressing design goals directly. To incorporate practical aspects such as chemical accessibility, the model can include soft constraints, such as minimizing ring count to favor smaller structures. Thus, our framework addresses key inverse design challenges while allowing prioritization of more optimal or diverse candidates.