Electron Microscope Detects 'Mouse Bite' Defects in Transistors Just 15 Atoms Wide

A modern high-performance chip packs billions of transistors onto a slice of silicon. Each of those transistors has a channel - the path electrons travel through - that may be just 15 to 18 atoms wide. At that scale, a single misplaced atom is not merely a flaw. It is a structural problem that can degrade how the chip performs. Until recently, there was no way to see those flaws directly.

Imaging the Walls of a 15-Atom Pipe

David Muller, the Samuel B. Eckert Professor of Engineering at Cornell, has a useful analogy for what goes wrong. "The transistor is like a little pipe for electrons instead of water," he said. "You can imagine, if the walls of the pipe are very rough, it is going to slow things down. And so measuring how rough the walls are and which walls are good and which walls are bad is now even more important."

Muller lab, in collaboration with Taiwan Semiconductor Manufacturing Company (TSMC) and Advanced Semiconductor Materials (ASM), has now developed a way to see exactly that roughness, in three dimensions, at atomic resolution. The technique is called electron ptychography. An electron microscope pixel array detector - co-developed by Muller group - collects detailed scattering patterns of electrons after they pass through a transistor. By comparing how those patterns change across scan positions, the researchers can reconstruct an image with extraordinary clarity. The detector has produced the highest-resolution images in the world, a fact recognized by Guinness World Records.

The research, led by doctoral student Shake Karapetyan, was published February 23 in Nature Communications.

Mouse Bites in the Channel

When the team reconstructed the full three-dimensional structure of transistor channels and tracked individual atomic positions, they found what Karapetyan termed "mouse bites" - interface roughness arising from defects that formed during the chip fabrication process. The sample structures were grown at nanoelectronics hub Imec.

"Fabrication of modern devices takes hundreds, if not thousands, of steps of chemical etching and deposition and heating, and then every single step does something to your structure," Karapetyan said. "Before you used to look at projective images to try to figure out what was really going on. Now you have a direct probe to actually see after every single step and have a better grasp of what happened."

For chip developers, that kind of direct feedback is what separates debugging from guessing. A fabrication engineer who can see, in atomic detail, what a particular temperature or deposition step does to the channel surface can make targeted adjustments rather than iterating blindly through hundreds of process variants.

A Reunion 25 Years in the Making

The collaboration has a backstory that spans decades. Muller worked at Bell Labs from 1997 to 2003, where he and Glen Wilk - now vice president of technology at ASM - explored replacing silicon dioxide as the gate material in transistors. Silicon dioxide leaked too much current at small scales. Their proposed alternative, hafnium oxide, eventually became the industry standard for computers and cell phones in the mid-2000s. "The papers we published on how to use electron microscopes to characterize these materials - a lot of the semiconductor folks had read those very, very carefully," Muller said.

More than 25 years after that collaboration, Muller and Wilk reunited to apply the newer EMPAD-based imaging to modern semiconductors with TSMC support. "When we got back into this project, that was very clear. And the microscopy has gone a very long way. Back then, it was like flying biplanes. And now you have got jets."

From Phones to Quantum Computers

The capability has broad potential reach. Modern computer chips appear in everything from smartphones and laptops to AI data centers and automobile electronics. Quantum computers are a particular target: they require extraordinary structural control of materials at scales where every atom matters, and current understanding of what those structures look like after fabrication remains incomplete.

"Since there is really no other way you can see the atomic structure of these defects, this is going to be a really important characterization tool for debugging and fault-finding in computer chips, especially at the development stage," Muller said. "I think there is a lot more science we can do now, and a lot more engineering control, having this tool."

Co-authors include Steven Zeltmann of PARADIM at Cornell, and Ta-Kun Chen and Vincent Hou of TSMC. The research was funded by TSMC; microscopy facilities were supported by CCMR and PARADIM, funded by the National Science Foundation.