A large-scale nanocrystal database with aligned synthesis and properties enabling generative inverse design

TL;DR

The study tackles the lack of large-scale, aligned nanocrystal synthesis-Property data by building the NSP database and enabling generative inverse design. It introduces NanoExtractor, an LLM-based information extractor augmented with four data-augmentation strategies, achieving a peak 88% weighted accuracy and outperforming domain-specific and general LLMs. The NSP database enables NanoDesigner to generate viable synthesis routes for PbSe and MgF2, including a non-intuitive 1:1 MgCl2:NaF precursor ratio experimentally shown to suppress byproducts, illustrating a powerful human-AI collaborative framework for nanocrystal discovery. By providing data, models, and protocols, the work lays a foundation for forward prediction and inverse design in nanomaterials and accelerates experimental validation and discovery.

Abstract

The synthesis of nanocrystals has been highly dependent on trial-and-error, due to the complex correlation between synthesis parameters and physicochemical properties. Although deep learning offers a potential methodology to achieve generative inverse design, it is still hindered by the scarcity of high-quality datasets that align nanocrystal synthesis routes with their properties. Here, we present the construction of a large-scale, aligned Nanocrystal Synthesis-Property (NSP) database and demonstrate its capability for generative inverse design. To extract structured synthesis routes and their corresponding product properties from literature, we develop NanoExtractor, a large language model (LLM) enhanced by well-designed augmentation strategies. NanoExtractor is validated against human experts, achieving a weighted average score of 88% on the test set, significantly outperforming chemistry-specialized (3%) and general-purpose LLMs (38%). The resulting NSP database contains nearly 160,000 aligned entries and serves as training data for our NanoDesigner, an LLM for inverse synthesis design. The generative capability of NanoDesigner is validated through the successful design of viable synthesis routes for both well-established PbSe nanocrystals and rarely reported MgF2 nanocrystals. Notably, the model recommends a counter-intuitive, non-stoichiometric precursor ratio (1:1) for MgF2 nanocrystals, which is experimentally confirmed as critical for suppressing byproducts. Our work bridges the gap between unstructured literature and data-driven synthesis, and also establishes a powerful human-AI collaborative paradigm for accelerating nanocrystal discovery.

Figures (5)

• Figure 1: Data augmentation strategies and prompt design of NanoExtractor. (a) Schematic diagram of four data augmentation strategies for raw labels. (b) Two prompt templates designed for training with raw labels and four types of augmented data.
• Figure 2: Data augmentation strategies and prompt design of NanoExtractor. (a) Schematic diagram of four data augmentation strategies for raw labels. (b) Two prompt templates designed for training with raw labels and four types of augmented data.
• Figure 3: Performance evaluation of NanoExtractor. (a) NanoExtractor output for the test set sample (Test set_1) and (b) the reference output. (c) Weighted average test set scores evaluated by human experts across varying training epochs. (d) Comparison of weighted average scores on the test set for NanoExtractor against chemistry-specialized (ChemDFM, ChemLLM, SciLitLLM) and general-purpose (GPT-5.2, Grok-4) LLMs.
• Figure 4: Statistical overview of the NSP database. Statistics on (a) reaction types, (b) product properties, and (c) product names recorded in the NSP database. (d) Probability of reactant combinations for CsPbBr3 nanocrystals in the NSP database.
• Figure 5: Generative inverse design and experimental validation. (a) Schematic diagram of inverse design for nanocrystals using NanoDesigner with the NSP database as training data. (b) The suggested synthesis route for MgF2 nanocrystals by NanoDesigner. (c) Photographs of MgF2 nanocrystals as colloids, ethanol dispersions, and dried colloids. (d) TEM image of MgF2 nanocrystals. (e) XRD patterns of PbSe and MgF2 nanocrystals. (f) TEM images of PbSe nanocrystals, the inset shows size distribution statistics.