Advancing the Study of Grain Growth Dynamics in Thin Films: Real-Time Imaging, Crystallography and Deep Learning at the Frontier

The majority of metals and alloys used in engineering applications, from structural components to microchip interconnects, are polycrystalline in nature. This means that they are composed of many small crystallites, each made of an identical material, but with their atoms arranged in lattices having different orientations. These small, misoriented crystallites join together at planar defects known as grain boundaries. The grain and grain boundary structure of materials has a profound impact on their properties and performance, from strength and ductility to electrical resistivity, and so an understanding of how such structures evolve in time is critically important in designing the next generation of materials.

The manner in which grains grow and evolve is a surprisingly complicated process that, despite decades of study, is still not fully understood, This complexity results from an interplay among many factors—geometry, crystallography, and topology—that conspire to determine how grain boundaries move to lower the energy of the polycrystal. One of the biggest challenges for scientists studying grain growth is in understanding the role of influences such as heat and mechanical stresses in dictating grain dynamics. Indeed, until now, it’s been impossible to capture grain coarsening in real-time and at a high enough resolution to draw meaningful conclusions, and consequently grain growth models often miss the mark because they can’t fully match what is actually happening. Moreover, the traditional way of analyzing grain structures is inherently inefficient, typically based on the hand tracing of grain boundaries, and therefore also limited in how much data can realistically be collected.

To address these limitations, a new research paper published in JOM and led by Professor Katayun Barmak and PhD candidate Matthew Patrick from Columbia University together with Professor Jeffrey Rickman from Lehigh University, used nanocrystalline thin films as their testing ground. This platform was selected as thin films have a simpler structure compared to bulk materials, which makes them easier to study. They also paired transmission electron microscopy (TEM) imaging with MEMS heating systems, allowing them to watch grain growth happen in real-time, and then employed automated analysis tools based on machine learning that obviate the tedious task of manual boundary tracing. This meant that they could analyze thousands of grains, thereby revealing patterns that had heretofore been difficult to spot. For example, they found that, like their much larger-grained microcrystalline counterparts, certain low-energy grain boundaries in nanocrystalline materials are preferred during growth, offering a glimpse into how these systems naturally try to minimize energy.

One of their key experiments involved heating the thin films while watching the grains grow in real time. The team used MEMS-based heating chips, which are incredibly precise, to control the temperature and combined that with real-time computer vision-based software to minimize thermal drift. This strategy allowed them to capture detailed images that highlighted the impact of grain interactions. By pairing these images with occasional precession enhanced diffraction scans, they could also map out the orientations of the grains and crystallographic character of the boundaries of the grains as they evolved. It should be emphasized that the ability to capture grain growth in real time via imaging combined with intermittent crystal orientation mapping to extract boundary crystallography is the key advantage of thin film studies vis-à-vis those in microcrystalline bulk sample, the latter relying solely on intermittent crystal orientation mapping using X-ray diffraction. To analyze the acquired thin film data, the team used a deep-learning algorithm to automate the process of identifying grain boundaries. As noted above, in the past, this kind of work had to be done manually, which was slow and limited the amount of data researchers could gather. The acquisition of data for thousands of grains revealed some fascinating patterns, including the observation that low-energy boundaries were much more likely to remain stable during growth. This observation confirmed a long-standing hypothesis that the populations of energetically favorable boundaries grow as the grain structure evolves. The authors also compared what they noted in thin films to the analogous behavior of bulk materials. While the overall trends were similar, the unique geometry of thin films—particularly their column-like structures—introduced some interesting constraints. For example, grain growth in the films tended to slow down and eventually stagnate at grain sizes near the thickness of the film, forming predictable distributions no matter how long or how hot the films were heated. This was a key finding because it shows how the structure of the material itself influences how grains behave.

In conclusion, the research work of Professor Katayun Barmak and her team is significant because it takes on some of the biggest challenges in understanding grain growth, particularly in nanocrystalline thin films, and approaches them in a fresh, innovative way. They used cutting-edge tools including high-resolution TEM imaging, crystal orientation mapping, automated grain boundary segmentation powered by AI, and MEMS-based temperature control systems to answer questions that have stumped researchers for years.  One of the most exciting parts of this research is its implications for materials design that, until now, has been based largely on trial and error. This study offers new knowledge into how grain boundaries behave, which could help scientists and engineers predict how materials will evolve during processing. For example, knowing how grain boundary character distributions affect growth could allow materials to be designed for specific purposes, like stronger structural components, more efficient electronics, or materials that hold up better in extreme temperatures. Industries like aerospace, semiconductors, and renewable energy could benefit enormously from these advances, especially since polycrystalline materials are critical in those fields. Another major takeaway from this study is that it highlights the utility of thin films as experimental testbeds for grain growth studies.  In particular, while thin films are simpler to study than bulk materials, the team showed that they can still provide valuable insights into how materials behave more generally. This finding opens the door to using thin films for large-scale, dynamic experiments, not just to study grain growth but also to look at other important phenomena such as phase transformations and diffusion. A deeper understanding of these kinetic processes is critical to improving a wide range of materials used in everyday life and high-tech industries alike.

In a statement to Advances in Engineering, Professor Katayun Barmak said “Having painstakingly traced grain boundaries as an undergraduate and again as a graduate student, and then having made generations of graduate students in my group also trace boundaries by hand, I am ecstatic to see it now (very nearly fully) automated. To be able to combine automated boundary tracing with real time imaging and crystallography is a dream come true.”