I have worked in several different institutions in China and Europe. I started my research career in the Institute of Metal Research (IMR), Chinese Academy of Sciences in Shenyang. After briefly working at Shanghai Jiao Tong University, I moved to Europe where I worked at RWTH Aachen University for three years, and then at the department of physics at Uppsala University for two years. Then I moved back to China and joined the department of materials science at Xiamen University. Since 2013, I have been a professor at the School of Materials Science and Engineering at Beihang University in Beijing.

Credit: Jianguo Li

All those changes have been driven by my research interests. For example, I started my research on experimental synthesis and characterization of the MAX phase. MAX materials refer to a family of ternary layered carbides or nitrides, where M is an early transition metal, A is an A-group element (such as gallium, aluminum, silicon, and so forth) and X is carbon and/or nitrogen. One of the interesting aspects of MAX materials is that MAX phases can present metallic and ceramic properties. However, I found myself puzzled by some research questions that cannot be well explained via experimental techniques — such as why MAX materials have good electrical conductivity, and why they are not brittle as other ceramics. I then turned those queries into theoretical models by learning ab initio calculations by myself for the MAX phase. This experience convinced me to devote more effort into theoretical calculation and modeling, which then led me to my position at Aachen where I expanded my experiences in bulk and powder MAX materials to MAX thin films. At Uppsala University, I even spent all my time in theory, and at the same time, I started to have a great interest in phase-change materials. After two years in Uppsala, I joined Xiamen University where I started my research group working on theories of phase-change materials, thus recognizing the importance of materials calculations. Around the time when the US government announced the Materials Genome Initiative, Beihang University invited me to lead a project on materials genome engineering and that is where I am now. Overall, I have been working on many different things spanning structural materials, semiconductors and 2D materials and have to solve various challenges, but each time it is a fruitful learning experience.

My main research topics include phase-change materials, 2D materials, and lately the development of materials informatics tools. I started work on chalcogenide phase-change martials when I was at Uppsala University. Although phase-change memory achieved successful commercial applications, many fundamental questions on phase-change materials remain unknown, such as the mechanism of rapid phase transition between crystalline and amorphous states. It was a great challenge to directly observe the structure evolution during device performance due to the ultra-small volume and ultra-fast transition speed for phase-change materials. I revealed the atomic structure of crystalline and amorphous phases as well as the transition mechanism of phase-change materials with help from first principles and molecular dynamics simulations. Furthermore, by using high-throughput calculations and experiments, we developed a phase-change material — Y-doped Sb₂Te₃ — that exhibits comparable properties to the commercially used Ge₂Sb₂Te₅ compound.

I started working on 2D materials at Beihang University, mainly focusing on the 2D MXene phases that can be derived by etching the MAX phases. This large category of 2D materials has many interesting research questions to be answered. The first report of experimental synthesis of the MXene phase in the laboratory dates back to 2011. The first MXene material that I worked on was the 2D magnetic Cr₂C, which was one of the first works on 2D MXene magnets. I also worked on a theoretical screening of MXene materials for important energy applications, such as photocatalysis and water splitting. We also introduced the criteria for high-throughput screening of candidate materials for photocatalysts. It is worth mentioning that, as a computational materials scientist, it is always an exciting moment when your predictions have been verified by experiments. For instance, in 2017, we theoretically proposed a new category of 2D boride materials — MBene. One year later, an experimental research group in the US successfully synthesized the MBene phase. Personally, I feel very happy to see that theoretical predictions can lead to discoveries of new materials.

As the science research paradigm changes from the third (computation and simulation) to the fourth (data-driven) phase, we endeavor to develop a universal material’s design platform. Under the support of the Chinese Materials Genome Engineering Project, we developed a platform named ALKEMIE (Artificial Learning and Knowledge Enhanced Materials Informatics Engineering).

ALKEMIE is the product of a fruitful collaboration among physicists, chemists, materials scientists, mathematicians, and software engineers. In ALKEMIE, we have developed multiple modules and applications to address various material challenges, such as the high-throughput computation module (Matter Studio), the data storage module (Matter DB), the machine learning algorithm module (Matter AI), and the graph neural network machine learning potential module for cross-scale simulations (Potential Mind). In addition, we are also currently developing a large language model application, named MatGPT, based on the open-source LLaMA and GLM models. To meet different user needs, we have developed local installation packages running in client-service mode, web clients in ALKEMIE cloud database, and command-line interface tools for developers, among others. In the future, we hope that ALKEMIE can serve as an interdisciplinary collaboration platform, accelerating research and development in new materials — including phase-change materials, energy materials, 2D materials, and biomedical materials — based on our user-friendly graphical interface. Benefiting from the high compatibility of the ALKEMIE platform, multi-scale computational workflows can be realized according to our development framework, thereby accelerating the discovery of new materials.

It is important to develop new methods and theories that can explain experimental observation and further help experimental design. For instance, it remains challenging to predict the properties of materials across different scales. Machine learning-based methods can potentially provide some solutions. For instance, machine learning interatomic potentials extending the scale of molecule dynamics simulations can be helpful in bridging atomic-scale insights to device-scale simulations. In addition, machine learning methods that utilize human knowledge and materials datasets can provide an alternative pathway to discover new materials with desired functionalities. For example, a dedicated generative pre-trained transformer (GPT) model for nano/2D materials would greatly reduce the primary screening time of finding new candidates for specific applications. Normalizing different materials databases and transforming the data to be readily applicable to novel machine learning methods still have a long way to go. We propose that the 3M (multicomponent, multiscale, and multistage) challenges in materials data should be handled to maximize the power of machine learning for materials research. Finally, it is always crucial to enhance communications and collaborations with experimentalists to understand their requirements so that we can provide better insights for explanation of the experiments.

One important consideration is to maintain a balance between the specificity and generality of tasks for such platforms. I feel that the most important aspect is modular design. For general applicability or specific software, modularization allows different software to be integrated through a unified design philosophy and workflow framework. For example, ALKEMIE has developed a universally first-principles computational workflow for VASP, QE, or Abinit through modularization, to meet different computational needs for various material systems. For specificity tasks, such as phase-change memory materials, we have pre-trained machine learning potential models for multicomponent systems like Ge–Sb–Te, which can be directly utilized through interfaces without retraining. Also, the platform should be extendable and implement easy-to-use application programming interfaces (API) and user-friendly operation graphical user interfaces (GUI). For specific or general applications, user-friendly operation GUI allow users to maintain consistent operating logic, reducing the barriers caused by switching between different software. Moreover, easy-to-use API provide development guidelines and convenience for customizing specific needs for different tasks. Finally, continuous development and updates are essential. We should constantly optimize the software by identifying issues through user feedback and data analysis. ALKEMIE has been continuously developed and evolved since 2016, and we have been constantly developing new features such as large language models for materials by the data-driven artificial intelligence paradigms.

Openly accessible data and code are crucial for computational materials science in the data driven paradigm. The biggest challenge for open data and open code is that in the field of computational materials science, or more broadly speaking in the materials science community, the culture of open source has not been fully built yet. Thus, publishers and funding agencies need to work on policies to encourage the authors to share the data and code in addition to the scientific findings. As for the materials data, in addition to be findable and accessible, it should also be interoperable and reusable, also known as the FAIR principle. The FAIR principle means that data sharing does not simply entail uploading the XLS file or output of the software to an online repository, but efforts to include metadata and additional information making the data more understandable are also required. In this sense, to save the time and effort of the data contributor, detailed but simple instructions to share FAIR data should be made for specific fields.

In my opinion, open code is more common than open data. But the issue is that the code might become unusable without an appropriate amount of data. Thus, efforts on open data and open code can be fused, as we did in the ALKEMIE platform consisting of data generation, data management, and data mining modules, which should be helpful to further harness the power of the open-source culture in computational materials science. Another suggestion is that, for the open-source code, in addition to the user instructions and test examples, the application boundary of the code — for instance, if a user re-applies the code in other systems — needs to be clear. Finally, how to make sure the owners of the data and code get credit and recognition when these resources are reused is also a big challenge. Proper practices for identifying and citing these intellectual properties are required.

I started my career as an experimental researcher, which gave me a good understanding of what our experimental collaborators need for their research. One of my research topics is on phase-change materials, which can be applied to several applications, such as random-access memory (RAM). There are numerous research inquiries concerning phase-change RAM across various scales, including defect/doping at the atomic level, phase stability at the mesoscale, and device fabrication. Addressing these issues necessitates collaboration among materials scientists, chemists, and physicists, as well as cooperation between computational and experimental scientists. My research experience has equipped me with a skill set for productive collaboration.

In general, one of the main challenges is that research targets and outcomes are different. For instance, physicists and chemists care more about micro- or atomic-scale insights, such as chemical bonding; however, electrical engineers are more interested in the overall device performance. The industry collaborators are more sensitive to the fabrication process and the cost. Overall, I feel that the development of a common (scientific) language between collaborators can be helpful, and to bridge your own research questions to those that your collaborators care about. For instance, to collaborate with engineers, we had to better answer the question for them: why are those atomic-level insights important for the device performance?

First, data sharing and standardization, as both academia and industry often possess large amounts of materials data related to materials properties, synthesis methods, and processing techniques. By sharing data and establishing common standards for data formatting, analysis, and representation, both parties can pool resources and avoid redundant efforts, fostering the development of more accurate models and predictions.

Second, I feel that academia and industry can collaborate on many research projects, which enables the integration of academic expertise with industrial requirements, leading to practical solutions and advancements. They can also collaborate on translational research projects converting basic research into practical applications. Academia has expertise in fundamental research and can transfer knowledge and technologies to industry. Collaborative efforts can facilitate the transfer of materials-informatics tools, techniques, and methodologies to industry, aiding in the development of innovative products. Organizing joint workshops, conferences, and seminars facilitates knowledge exchange, dialogue, and networking between academia and industry. Such events create opportunities for researchers, practitioners, and industry professionals to share ideas, present research outcomes, and collaborate on common challenges.

Finally, academia can contribute to industry’s skill requirements by offering educational programs and training initiatives in materials informatics. Collaborative efforts in curriculum development and internships programs can bridge the skill gap, equipping the workforce with industry-relevant knowledge and expertise.

Overall, academia and industry can combine their unique strengths, knowledge, and resources in materials informatics, accelerating the development and application of data-driven approaches and enhancing materials discovery, design, and innovation.

This question reminds me of an experience I had at a materials conference. At the time, I was being recognized in the community for the MAX phase calculations. In the conference, one speaker looking at my name tag said: “Oh, there is a very famous Dr Sun from Aachen, and he performed a lot of calculations …” I told him that “it is not he, it’s she …”.

I feel that computational research is one of the best options for women scientists, since it offers the flexibility in terms of working location and time. The flexibility allows one to spend more time with kids and family, if needed. For early-career women scientists, my first suggestion is to always be positive. For instance, I have been in a situation with very little start-up funding. I always stay positive by seeking more collaboration, which I feel has been the key to walking me out of many difficult situations. In addition, hard work definitely pays off, regardless of the gender.