For the first time, a Cornell University research team in the United States directly "saw" the structural defects hidden inside advanced chips at the atomic scale, and vividly called these tiny irregular shapes "mouse bites." This imaging breakthrough provides a new tool for the debugging and yield improvement of future high-end chips.
This new technology relies on a high-resolution three-dimensional electronic imaging method, developed by Cornell University in collaboration with Taiwan Semiconductor Manufacturing Company (TSMC) and semiconductor equipment manufacturer ASM. It can reconstruct the internal structure of the transistor at the nanometer or even atomic scale and directly locate microscopic defects that affect performance and reliability. The research results were published in the journal Nature Communications on February 23. The first author of the paper is Shake Karapetyan, a doctoral student at Cornell University.
David Muller, the project leader and Samuel B. Eckert Professor at Cornell's Duffield School of Engineering, said that it is almost impossible to directly see the atomic structure of these defects with existing methods, and the new method will become a key characterization tool for "debugging" and "troubleshooting" during the chip development stage. Since everything from smartphones and cars to artificial intelligence data centers and quantum computers rely on advanced chips, this development is expected to have a wide impact across the entire information industry chain.
In modern semiconductor devices, the transistor is the core unit that controls current switching, and its channel area is like a micro-pipe for electrons to "walk". Muller described that if the inner wall of this "pipe" is rough, it will hinder the flow of electrons, so accurately measuring the roughness of the channel wall and distinguishing which areas are "good" and which are "bad" becomes particularly critical at the atomic scale. The transistor channels in today's high-performance chips are only about 15 to 18 atoms wide, and their structures are so complex that any slight deviation can cause measurable performance differences. Karapetyan bluntly stated that at such a size, almost "every atom's position matters," and accurately characterizing these structures has been a problem.
Looking back at the early days of the development of semiconductor technology, most transistors were laid out in a planar manner and spread out laterally on the surface of the chip. As sizes continue to approach physical limits, the industry is turning to three-dimensional stacking structures, where devices are "standing up" vertically to form increasingly complex 3D architectures. Muller recalled that while working at Bell Labs from 1997 to 2003, he studied the physical factors that limited the extreme shrinkage of transistors. Today, the feature sizes of these 3D structures are smaller than the resolving power of many traditional characterization methods, making diagnosing performance issues increasingly difficult.
The evolution of advanced electron microscopy technology has laid the foundation for solving this problem. Muller and current ASM Vice President of Technology and Cornell alumnus Glen Wilk collaborated on research on using high dielectric constant hafnium oxide (HfO₂) as a gate material to replace silicon dioxide with severe leakage in small sizes during their time at Bell Labs. This work later promoted the popularization of hafnium oxide in computer and mobile phone chips. The paper they published that year on the use of electron microscopy to characterize related materials was widely read in the semiconductor industry.
Today, the "propeller aircraft" Muller calls has been upgraded to a "jet fighter", which is embodied in electron positron diffraction imaging (electron ptychography) technology. This method relies on an electron microscopic pixel array detector (EMPAD) developed by his research group. It records the scattering pattern generated when the electron beam passes through the transistor, and then calculates and reconstructs the subtle changes in the pattern between adjacent scanning points to obtain ultra-high-resolution images. EMPAD's precision is so high that it was recognized by the Guinness Book of World Records as achieving the highest resolution atomic-level imaging to date.
With the support of TSMC and its corporate analytical laboratory and nanoelectronics research center Imec, Muller and Wilk have reunited after 25 years to apply EMPAD technology to contemporary cutting-edge semiconductor structures. Karapetyan compared this process to solving a "super large puzzle", which requires collecting massive experimental data and completing complex computational reconstruction.
By processing and analyzing the data, the team was able to track the spatial position of individual atoms and quantify the subtle undulations at the transistor channel interface. They collectively referred to these tiny pits and roughness as "rat bite" defects. The defects were formed during optimized material growth steps during device fabrication, and the samples used for testing came from Imec's process lines. Karapetyan pointed out that the preparation of modern devices often requires hundreds or even thousands of chemical etching, deposition and heat treatment steps. Each step will have an impact on the final structure. In the past, one could only rely on projection imaging to "guess" what is happening inside, but now one can directly "see" structural changes after several key steps. This gives process engineers the opportunity to more finely adjust process parameters such as temperature and verify their structural results in real time.
The research team believes that this ability to directly visualize atomic defects will have a potential impact on almost all technological forms that rely on advanced chips, including conventional applications such as smartphones, laptops, and large data centers, as well as next-generation quantum computing systems that require extremely high precision in material structures. Karapetyan said that with such a set of tools, there will be greater room for performance in both basic scientific research and process engineering control in the future.