Reaching a mountain summit does not always require following the same path. Could the same principle apply to growing a forest of carbon nanotubes with prescribed structural features? A study published in National Science Review shows that artificial intelligence can help identify several distinct growth recipes that approach the same combination of array height, density and alignment.
Carbon nanotubes are tiny, hollow carbon structures. When many grow together on a surface, they form an array resembling a miniature forest. The height, density and alignment of this forest influence how the material behaves, making their coordinated control relevant to applications such as thermal management, energy storage and flexible electronics.
Growing these arrays to order is challenging because processing conditions are interconnected. Adjusting a temperature, gas flow or treatment time can change several structural features simultaneously. A modification that brings height closer to its target may move density or alignment further away. With eight processing variables, even ten settings per variable would create 100 million possible combinations.
To navigate this complexity, the researchers developed an AI-assisted inverse-design framework. They first constructed an experimental database containing 200 distinct recipes for water-assisted chemical vapour deposition. A neural network learned how the processing conditions relate to array height, density and alignment. An optimization algorithm then worked backward from a prescribed structural target to search for suitable recipes.
The approach deliberately sought alternatives. For each of three targets, repeated computational searches generated 200 candidate solutions. The researchers selected three recipes with substantially different processing conditions for each target and tested each recipe in three independent growth batches. Of the resulting 27 validation batches, 26 achieved structural descriptor matching scores above 95%. These scores describe how closely the measured array features approach their prescribed targets.
The arrays nevertheless retained differences at a smaller scale. For one target, the three routes produced closely matched overall array features, while the mean nanotube outer diameters were approximately 3.8, 4.8 and 7.2 nanometres. Their wall structures also differed. Similar forests, in other words, need not contain identical trees.
This finding broadens the aim of inverse synthesis from finding one suitable recipe to identifying several experimentally accessible alternatives. The AI framework acts like a navigation aid built from experimental knowledge, allowing researchers to explore routes beyond the immediate neighbourhood of a familiar recipe. Its initial database remains an experimental investment, but the learned relationship can be reused to screen additional targets.
Alternative recipes could eventually help researchers accommodate practical processing constraints while preserving selected structural features. Applying the approach to other catalysts or growth methods would require new experimental data and model retraining.
The study involved researchers from Huazhong University of Science and Technology, and Beihang University. Lei Zhu, a doctoral student at Huazhong University of Science and Technology, is the first author. Prof. Ming Xu is the corresponding authors.
National Science Review
Experimental study