SNU College of Engineering Professor In-Suk Choi's Joint Research Team
Designs Geometric Structure Using Machine Learning... Attached Unexpanding Device to 3D Curved Surface -


▲ Professor In-Suk Choi, Professsor Lien, Jhy-Ming, Researcher Yoo-gi Lee
SNU College of Engineering announced on June 10 that a joint research team led by Professors In-Suk, Young-Chang Joo, and Myoung-Gyu Lee of the Department of Materials Science and Engineering, Professor Chang-soon Kim of Graduate School of Convergence Science and Technology, and the joint research team of Professor Lien, Jyh-Ming at George Mason University, has developed a technology that can attach flexible elements to arbitrary three-dimensional curved surfaces without damaging them by utilizing machine learning algorithms.

 
With the active research on the development of free-growing electronic devices that has been recently taking place, there is a limit to the fact that currently commercialized flexible devices cannot be transformed into arbitrary complex forms due to material constraints.
 
Flat flexible TVs, for example, can bend in one direction, but cannot be transformed into a complete sphere. This is because silicon substrates, which are used in most electronic devices such as semiconductors and displays, are easily damaged by small stresses.
 
In response, a joint research team at SNU's College of Engineering succeeded in stably attaching flat-shaped materials, such as steel plates and silicon substrates, to arbitrary three-dimensional curved surfaces using machine learning based on computer-based development plots.
 
The researchers used an algorithm that approximates arbitrary complex three-dimensional curvature to a small grid (mesh) after which it is developed into two dimensions without overlap. At this moment, a machine learning technique called genetic algorithm was used to produce a minimal plot of development, with no overlap in complex three-dimensional shapes approximated to hundreds of grids.
 
The researchers suggested a methodology that produces various electronic devices on flexible substrates that are cut according to the generated development plot and attaches them back to the original three-dimensional curved surface. Since the development plot obtained through the algorithm has a minimized effective area, it can minimize waste of substrate materials that occur during the cutting process.
 
In addition, the three-dimensional geometry used by the researchers is approximated by a sufficient number of many hundreds of grids, in which the folding between the grid and the grid can be approximated by smooth bending in the process of attaching the elements in the form of the produced development plot to the three-dimensional shape.
 
As a result, the researchers proved that materials that break easily even under minute stress, such as silicon substrates, can be attached stably to diverse types of three-dimensional surfaces without local damage due to stress concentration. Furthermore, light emitting devices, including greatly brittle ITO materials, were attached to complex forms of three-dimensional surfaces and showed to operate normally.
 
SNU Professor In-Suk Choi stated that, "This study is significant in that it only uses the machine learning algorithm to provide a methodology that can attach materials in any form without any cracks or gaps." He also added by stating, "Since it is possible to produce various types of devices that are already in use without the development of new materials or processes, it will, in the future, be used in various fields such as wearable/body-attached devices, construction and vehicle interior and exterior design ."
 
The corresponding research paper was published online on the April 10 issue of Science Advances, a science journal.
 
Meanwhile, the research was carried out through support from the National Research Foundation of Korea's Mid-Term Researcher Support Project, Soft Robotics Technology Research Center, SNU's Creative Leading New Researcher Task, and LG Display under LGD-Seoul National University Incubation Program.
 

 

▲ It is not easy to deploy a 3D figure that is approximated by hundreds of grids into a two dimensional plane without any overlaps. The researchers used a machine learning-based algorithm that can control the final shape of the development plot without overlaps after deployment.



▲ In the process of attaching countless grid (mesh) plots to the original three-dimensional shape, the original three-dimensional shape was approximated by bending rather than folding, suggesting a methodology for attaching the original three-dimensional shape without breaking any gaps or damaging materials.



▲ ITO materials used for transparent electrodes such as displays are highly brittle and easily break, even under small stress. The researchers demonstrated that even such highly brittle materials can be used after transforming into arbitrary three-dimensional shapes by controlling only the geometry of the materials without the development of new materials and processes.



▲ Through the use of technology presented by the researchers, silicon substrates used as substrates for various electronic devices can also be transformed into arbitrary three-dimensional shapes by controlling only the geometry. Small analysis simulations as well as actual experiments by the researchers demonstrated that silicon substrates do not get damaged even after attachment to curved surfaces .