材料科学
走进人工智能实验室:材料发现的新时代

Much like a growing number of industries, material science is also being aided by artificial intelligence (AI).
Here, machine learning (ML) algorithms analyze vast datasets and identify patterns to suggest promising material candidates in minimal time, while consuming significantly fewer resources than trial-and-error methods.
The traditional manual, human-intensive work that was augmented by the advent of computational systems, which allows for complex calculations, is now being completely revolutionized by automated, parallel, and iterative processes driven by AI, simulation, and experimental automation.
The maturation of AI technology, combined with high-performance computing and hybrid cloud technologies, is helping materials science enter a new paradigm 由 accelerated discovery of new materials, predictive modeling of material properties, and autonomous experimentation.
This paradigm shift enables researchers to transition from trial-and-error approaches to data-driven design, significantly reducing development cycles and paving the way for advanced materials in energy, electronics, healthcare, and sustainability applications.
Recently, researchers from North Carolina State University took a major step further by creating a self-driving lab to achieve a new leap in lab automation and further accelerate the discovery of materials by scientists.
The automated lab collects ten times more data than traditional, manual 方法。 With this move, researchers can conduct real-time, dynamic chemical experiments, saving time and resources while 使 faster breakthroughs.
New lab discoveries won't take years anymore; rather, we are looking at a future where inventions will occur in days.
An AI-powered Lab: Real-Time Learning for Real-Time Discovery
为了 overcome the global challenges in clean energy, human welfare, and sustainability, it is 危急 to make rapid discoveries of advanced functional materials. 药物发现 and synthesis of new materials are also key to innovative technologies 喜欢 batteries, computer chips, solar panels, and 许多 更多。
其结果是, 许多 progress has 被制造 in materials acceleration platforms and self-driving laboratories.
Despite the progress, the capacity of these platforms and labs to explore complex parameter spaces 受阻 by low data throughput. 慢的 data transfers and processing lead to reduced productivity.
Hence, NC State University researchers have “introduced dynamic flow experiments as a data intensification strategy for inorganic materials syntheses within self-driving fluidic laboratories by the continuous mapping of transient reaction conditions to steady-state equivalents.”
发表于 Nature Chemical 工程1, the study details a state-of-the-art self-driving lab that uses real-time experiments to constantly gather data, thereby 制造 材料发现 更快更有效 while reducing costs and environmental impact.
For their work, the study received support from the National Science Foundation and the University of North Carolina Research Opportunities Initiative program.
Now, what does self-driving laboratories (SDLs) even mean? Well, these are robotic platforms that combine ML and automation with chemical and materials sciences to find materials more rapidly. In these ML-assisted modular experimental platforms, a series of experiments, which 选择 by the ML algorithm, are conducted iteratively to achieve the programmed goal.
“Imagine if scientists could discover breakthrough materials for clean energy, new electronics, or sustainable chemicals in days instead of years, using just a fraction of the materials and generating far less waste than the status quo.”
– Paper's co-author, Milad Abolhasani, ALCOA Professor of Chemical and Biomolecular Engineering at North Carolina State University
他补充说:
“This work brings that future one step closer.”
Representing a transformative approach to accelerating scientific discovery, self-driving labs are gaining popularity in the chemistry and materials science fields.
Self-driving labs that utilize continuous flow reactors often rely on steady-state flow experiments, where different precursors are combined before chemical reactions occur, and the mixture continuously flows through a microchannel.
该产品 comes out of it is then characterized by a series of sensors once the reaction is complete.
“This established approach to self-driving labs has had a dramatic impact on materials discovery,” noted Abolhasani, as he shared that this has allowed scientists to “identify promising material candidates for specific applications in a few months or weeks, rather than years, while reducing both costs and the environmental impact of the work.” But it wasn't perfect by any means, with there still being areas for improvement.
In particular, the idling of the system when the chemical reaction 发生 before the resulting material can be 特点是. The waiting time for self-driving labs can be as much as an hour for each steady-state flow experiment.
“We've now created a self-driving lab that makes use of dynamic flow experiments, where chemical mixtures are continuously varied through the system and 被监控 实时”。
– Abolhasani
这是什么 手段 是 they are eliminating the process of running separate samples through the system and testing each 其中 一种 一次 after they've reached a state.
Instead, they have built a system that 只是 does not stop running. The samples continuously move through the system. 本篇 is because “the system never stops characterizing the sample,” and the researcher can “capture data on what is taking place in the sample every half second.”
The integration of dynamic flow experiments within self-driving fluidic laboratories marks a departure from traditional batch experiments.
In contrast to conventional approaches, where isolated data points 聚集在一起 under steady-state conditions, in dynamic flow experiments, microfluidic principles 被利用 HPMC胶囊 快 mapping of reaction conditions.
By creating a constant stream of data, it drastically expands the accessible experimental data.
Abolhasani illustrated that the team now 得到 20 data points about what the experiment produces, starting with one after 0.5 seconds of reaction time, then one after 1 second of reaction time, and so on, unlike the 一种 data point they would 得到 after 10 seconds of reaction time. He added:
“It's like switching from a single snapshot to a full movie of the reaction as it happens. Instead of waiting around for each experiment to finish, our system is always running, always learning.”
Having so much more data can have a tremendous impact on the performance of an AI-powered lab. Data, after all, is key to an algorithm. AI is data hungry, and based on the data it 被喂养, the algorithm makes predictions.
According to Abolhasani:
“The most important part of any self-driving lab is the machine-learning algorithm the system uses to predict which experiment it should conduct next.”
As such, the streaming-data approach allows the ML brain of the self-driving lab to make not just faster but also smarter decisions, “honing in on optimal materials and processes in a fraction of the time.”
The quality of data also determines the accuracy of the predictions. 所以,通过 having more high-quality experimental data, the algorithm can make more accurate predictions, 然后 it 能够 解决 一个问题 更快。
“This has the added benefit of reducing the amount of chemicals needed to arrive at a solution.”
– Abolhasani
To demonstrate the capabilities of their system, the team applied dynamic flow experiments to CdSe colloidal quantum dots. This was used as a testbed due to its status as a well-established inorganic system with not only rich parameter dependencies but also substantial technological potential.
In this case, the team found their lab, which incorporated a dynamic flow system, achieved “an order-of-magnitude improvement in data acquisition efficiency.”
It yielded at least 10x more data than other self-driving labs that utilized steady-state flow experiments. Moreover, once trained, the self-driving lab was able to discover the best candidates on its first try.
This breakthrough, as Abolhasani said, “isn't just about speed,” but about achieving sustainability. By reducing the required number of experiments, the system significantly reduces both chemical consumption and waste, promoting more sustainable research practices. Abolhasani said:
“The future of materials discovery is not just about how fast we can go, it's also about how responsibly we get there. Our approach means fewer chemicals, less waste, and faster solutions for society's toughest challenges.”
AI's Expanding Role in Materials Science: Exciting Recent Discoveries
AI is transforming industries worldwide, 包括 material science, which is fundamental to many technological innovations and societal challenges.
结果, 用法 of AI in the discovery and development of new materials has been 继续 多年 现在,但它有 当然 gained traction in recent years as the technology 成为 more advanced and capable.
“With continued development, we expect robotics and automation will improve the speed, precision, and reproducibility of experiments across diverse instruments and disciplines, generating the data that artificial intelligence systems can analyze to guide further experimentation.“
– Dr. James Cahoon, a co-author of the paper ‘Transforming Science Labs into Automated Factories of Discovery.2“
With that, let's take a look at some key progress made in material science this year across different applications.
For starters, as we recently shared, with the help of AI, scientists have been able to design complex, 3D thermal meta-emitters that can 降低 indoor temperatures and help 保存 能源成本。 The material created using a newly designed ML technique can find applications in residential and commercial buildings, spacecraft, fabrics and textiles, automobiles, and more.
“By automating the process and expanding the design space, we can create materials with superior performance that were previously unimaginable.“
– The study's co-lead, Yuebing Zheng
Developing New Metallic Materials with Superior Strength
A couple of months ago, scientists 报道 using AI to design a new MPEA or multiple principal element 合金,这 旨在 发现 in aircraft components, catalytic converters, and knee replacements.
The newly designed MPEA comes with superior mechanical properties, which Sanket Deshmukh, associate professor in chemical engineering at Virginia Tech, said, “demonstrates how data-driven frameworks and explainable AI can unlock new possibilities in materials design.“
To interpret the analysis made by the AI model, Deshmukh and his team 用过的 SHAP (SHapley Additive exPlanations) analysis, which 允许 them to understand how different elements and their local environments 影响 MPEAs' properties, in turn, providing more insight and accurate predictions.
Besides accelerating the discovery of advanced metallic alloys, Deshmukh believes that integrating ML with evolutionary algorithms and experimental validation can also help us create tools that “can be extended to complex material systems such as glycomaterials – polymeric materials containing carbohydrates.”
Unraveling the Secrets of Dendritic Growth in Thin Films
- 研究3 from Tokyo University of Science (TUS) has developed an explainable AI model that predicts the growth of dendrites (tree-like branching pattern) in thin films, which is a major obstacle in their large-area fabrication and is restricting their commercialization.
By revealing the specific conditions and mechanisms behind dendrite branching, the AI model is helping improve the growth process of thin films. Thin film devices are 危急 in 技术喜欢 半导体。
The novel AI model integrated the machine learning method called principal component analysis (PCA) and the topology technique called persistent homology with free energy analysis.
“By integrating topology and free energy, our method offers a versatile approach to material analysis. Through this integration, we can establish a hierarchical connection between atomic-scale microstructures and macroscopic functionalities across a wide range of materials, paving the way for future advancements in material science.“
– Professor Masato Kotsugi from the Department of Material Science and Technology at TUS
Gaining a Better Understanding of Nanoparticles
A team of scientists from multiple universities 一起来了 至 开发 一个方法4 至 更好 理解 the dynamic behavior of nanoparticles, which are the building blocks of electronics, pharmaceuticals, and industrial materials.
It blended electron microscopy with AI to visualize the structures and movements of molecules at an unprecedented time resolution.
As Peter A. Crozier, a professor of materials science and engineering at Arizona State University, explained:
“Electron microscopy can capture images at a high spatial resolution, but because of the velocity at which the atomic structure of nanoparticles changes during chemical reactions, we need to gather data at a very high speed to understand their functionality.”
To mitigate this noise, they developed an AI method that automatically removes it, “enabling the visualization of key atomic-level dynamics.”
Meanwhile, a research group from Graz University of Technology is taking nanostructure construction to a new level using AI.
For this, they are 发展 a self-learning AI system that autonomously positions individual molecules quickly and in the 右 orientation using scanning tunnelling microscopes, otherwise a difficult and time-consuming process, 允许 这些因素包括原料奶的可用性以及达到必要粉末质量水平所需的工艺。 建设 of “highly complex molecular structures, including logic circuits in the nanometre range.“
The goal is to ultimately build quantum corrals, which are nanostructures in the shape of a gate that can trap electrons, and use them to build logic circuits to study how they work at the molecular level.
Discovering Better Photovoltaic Materials
A sustainable alternative to conventional silicon-based solar cells, perovskite solar cells 显示 great promise as next-generation photovoltaic technology 转换 sunlight into electricity.
Their efficacy can be further increased through molecules that conduct positive charges, but there are millions of different molecules, which means synthesizing and testing all of them. However, by utilizing AI with automated high-throughput synthesis, a team of researchers from the Karlsruhe Institute of Technology (KIT) was able to discover new organic molecules5 in just a few weeks, with only 150 targeted experiments.
The newly discovered materials also increased the efficiency of a reference solar cell by about two percentage points.
For this, the scientists turned to a database with one million virtual molecules and randomly selected 13,000 out of them before choosing 101 of them. 选择的 民政事务总署 the greatest differences in their properties, and they were synthesized - robotic systems before 使用它们 to produce solar cells.
Laying the Foundation for AI-Driven Material Discovery
While all these discoveries are 被制造, for AI to actually make it happen, it needs data. 本篇 includes data about materials as well as data from large-scale simulations.
While many such databases are available, they are pretty isolated, as such, needing a standard “so that users can communicate with all these data libraries and understand the information they receive,“ noted Gian-Marco Rignanese, professor at the Institute of Condensed Matter and Nanosciences at UCLouvain in Belgium.
So, last summer, a major international collaboration 发布 an extended version of the OPTIMADE standard to facilitate AI-driven material discovery.
The OPTIMADE (Open databases integration HPMC胶囊 材料设计) standard is 已备份 by a large international network of institutions and materials databases. With the aim of giving users easier access to leading as well as lesser-known materials databases, a new version of it 被介绍了 that can further accelerate the ongoing AI-driven material discovery.
Investing in AI for Material Discovery
在规划婴儿食品行业的工艺要求时,安全性和可靠性是工艺设计中最重要的方面。 looking into investing in this space, 字母表公司. (GOOGL -0.18%) owned Google 是那个 has released an AI tool called Gnome. It has 报告发现 2.2 million new crystals 它. 然后有 Microsoft (MSFT -0.52%), which has introduced MatterGen and MatterSim to create new candidate materials and validate them.
But there's another AI giant that has launched its own model to elevate the scale and precision of materials research.
NVIDIA®(英伟达™)公司 (NVDA -0.71%)
That's none other than AI darling Nvidia. Late last year, the company 介绍 NVIDIA ALCHEMI (AI Lab for Chemistry and Materials Innovation).
The platform aims to accelerate R&D in chemistry and materials science through the power of AI, and to achieve this, it includes APIs and accelerated inference microservices. 本篇 will enable the creation and deployment of 根 AI models to explore the vast materials universe and suggest new candidates, 和 的发展和 用法 of surrogate models to 实现 之间的平衡 这些因素包括原料奶的可用性以及达到必要粉末质量水平所需的工艺。 成本 of computation 和准确度。 它也会 允许 accessible informatics tooling and pre-trained foundation models HPMC胶囊 fast screening and simulation tools 培训 和 微调 AI模型 HPMC胶囊 new use cases.
Through ALCHEMI, NVIDIA aims to speed up the discovery workflow and “usher in a new era of breakthrough discoveries powering a more sustainable, healthier future.”
NVIDIA®(英伟达™)公司 (NVDA -0.71%)
Nvidia is a $4.22 trillion market cap company, whose shares are trading around $173.5, up 28.8% YTD. Its EPS (TTM) is 3.10, the P/E (TTM) is 55.73, and the dividend yield offered is 0.02%.
NVIDIA Corporation (NVDA) 最新股票新闻和动态
The Score: Norfolk Southern, Citigroup, Nvidia and More Stocks That Defined the Week
更好的人工智能(AI)股票:CoreWeave vs. Nvidia
Trade Tracker: Kevin Simpson buys Apple and sells covered calls on Nvidia
英伟达 AI 芯片新闻、台湾半导体盈利和经济数据推动股市创下新高:每周回顾
What Nvidia H20s mean for China's AI ambitions
Top Republican on China panel objects to resumption of Nvidia H20 chip shipments
结语
As AI, automation, and high-performance computing 一起来, materials science is entering its most transformative era, marking a much-needed shift from human-led 反复试验 to data-driven, autonomous discovery.
Amidst this, AI-powered labs and self-driving experimental platforms are completely changing the way scientists not only discover but also test and optimize materials. Moreover, with initiatives like NVIDIA's ALCHEMI, Google's Gnome, and Microsoft's MatterGen, big tech is also betting on AI to fuel the next wave of innovation!
参考文献:
1. Delgado‑Licona, F.; Alsaiari, A.; Dickerson, H.; Klem, P.; Ghorai, A.; Canty, R. B.; Bennett, J. A.; Jha, P.; Mukhin, N.; Li, J.; López‑Guajardo, E. A.; Sadeghi, S.; Bateni, F.; Abolhasani, M. Flow‑Driven Data Intensification to Accelerate Autonomous Inorganic Materials Discovery. Nature Chemical Engineering, published online July 14, 2025. https://doi.org/10.1038/s44286-025-00249-z
2. Angelopoulos, A.; Cahoon, J. F.; Alterovitz, R. Transforming Science Labs into Automated Factories of Discovery. Sci. Robot. 2024, 9(95), eadm6991. https://doi.org/10.1126/scirobotics.adm6991
3. Tone, M.; Sato, S.; Kunii, S.; Obayashi, I.; Hiraoka, Y.; Ogawa, Y.; Fukidome, H.; Foggiatto, A. L.; Mitsumata, C.; Nagaoka, R.; Varadwaj, A.; Varadwaj, P. R.; Matsuda, I.; Kotsugi, M.; et al. Linking Structure and Process in Dendritic Growth Using Persistent Homology with Energy Analysis. Sci. Technol. Adv. Mater. Methods 2025, 5 (1), Article 2475735. https://doi.org/10.1080/27660400.2025.2475735
4. Crozier, P. A.; Leibovich, M.; Haluai, P.; Tan, M.; Thomas, A. M.; Vincent, J.; Mohan, S.; Morales, A. M.; Kulkarni, S. A.; Matteson, D. S.; Wang, Y.; Fernandez‑Granda, C. Visualizing Nanoparticle Surface Dynamics and Instabilities Enabled by Deep Denoising. 科学 387 (6737),949-954。 https://doi.org/10.1126/science.ads2688
5. Wu, J.; Torresi, L.; Hu, M.; Reiser, P.; Zhang, J.; Rocha‑Ortiz, J. S.; Wang, L.; Xie, Z.; Zhang, K.; Park, B.‑W.; Barabash, A.; Zhao, Y.; Luo, J.; Wang, Y.; Lüer, L.; Deng, L.‑L.; Hauch, J. A.; Guldi, D. M.; Pérez‑Ojeda, M. E.; Seok, S. I.; Friederich, P.; Brabec, C. J. Inverse Design Workflow Discovers Hole‑Transport Materials Tailored for Perovskite Solar Cells. Science 2024, 386(6727), 1256–1264. https://doi.org/10.1126/science.ads0901