The chip in your phone contains billions of transistors, each one a pipe roughly 15 to 18 atoms wide through which electrons either flow or don’t. For decades, engineers have known that the walls of those pipes matter enormously (a rough wall scatters electrons, slowing everything down) but no one had actually seen that roughness in three dimensions, at the atomic scale, in a working device architecture. You could infer it. Model it. Worry about it. You could not look at it directly.
That changed this February, when a team at Cornell University published results in Nature Communications demonstrating the first 3D atomic-resolution images of defects inside modern transistors. The technique, multislice electron ptychography, reconstructs a complete volumetric picture of a chip’s interior from the scattering patterns of electrons fired through it. What the researchers found when they finally got a proper look were jagged imperfections at the channel interfaces: irregularities Shake Karapetyan, the doctoral student who led the experimental work, calls “mouse bites”.
Exactly the sort of name that sticks.
The transistor has been around since the late 1940s, and for most of its history it was built flat: a sprawling suburb of logic, with components spreading outward across silicon wafers. That arrangement worked until the chips started running out of horizontal space, and designers began stacking transistors upward into three dimensions, like apartment blocks instead of bungalows. Today’s gate-all-around transistors wrap the electrical gate entirely around a thin silicon channel, squeezing more performance into less volume. But that geometry creates new headaches. “These days, a transistor channel can be only about 15 to 18 atoms wide, which is super, super tiny, and they’re extremely intricate,” Karapetyan says. “At this point, it matters where every atom is, and it’s really hard to characterize.”
The person who perhaps understood that problem earliest is David Muller, an engineering professor at Cornell who co-directs the Kavli Institute at Cornell for Nanoscale Science. Between 1997 and 2003 he worked in the research and development labs at Bell Labs, where transistors were invented, exploring, among other things, just how small a transistor could physically get before physics objected. There, alongside a colleague named Glen Wilk, he helped replace silicon dioxide with hafnium oxide as the material for transistor gates. Silicon dioxide had been leaking too much current at small scales; hafnium oxide didn’t. By the mid-2000s, hafnium oxide had become the industry standard for computers and mobile phones, a quiet revolution embedded in essentially every modern device. Both men eventually left Bell Labs. More than 25 years passed.
Then they started working together again.
This time, Wilk, now vice president of technology at Advanced Semiconductor Materials in Phoenix, helped source the samples. Muller’s group at Cornell contributed the imaging technology they’d spent years developing: the electron microscope pixel array detector, or EMPAD, a device so sensitive that images produced with it hold the Guinness World Record for atomic resolution. Cornell partnered with Taiwan Semiconductor Manufacturing Company, one of the world’s dominant chip producers, who co-funded the project and contributed analytical expertise. The sample structures themselves were grown at imec, a nanoelectronics research hub in Belgium.
What the EMPAD enables is something called multislice electron ptychography. A conventional electron microscope fires electrons through a sample and records the shadow: useful, but essentially flat, a projection rather than a volume. Ptychography works differently. The detector captures the full scattering pattern of electrons as they pass through the transistors, and by comparing how those patterns shift from one scan position to the next, a computational algorithm reconstructs the three-dimensional structure in extraordinary detail. “You can think of this imaging technique like solving a massive puzzle,” Karapetyan says, “both in terms of taking the experimental data and doing the computational reconstruction.”
Muller compares the advance to aeronautics. The electron microscopy his group used at Bell Labs in the late 1990s now seems quaint by comparison. “It was like flying biplanes,” he told the university. “And now you’ve got jets.”
The jets revealed the mouse bites. At the interface between the silicon channel and the surrounding gate oxide, the researchers found roughness in three dimensions: atomic-scale irregularities that formed during the chip’s fabrication. Making a modern chip involves hundreds, sometimes thousands, of sequential steps: etching, deposition, heating. Each step, Karapetyan notes, does something to the structure. The new imaging let the team see precisely what. The silicon in the 5-nanometre-thick channel doesn’t stay put against its interfaces: it relaxes inward, leaving only around 60 per cent of atoms in a configuration resembling bulk silicon. The rest are perturbed, a soft chaos at the margins that earlier, projection-based methods could only infer existed.
“Since there’s 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 says.
That matters most at the frontier. Conventional chips, quantum processors, AI data centre hardware: all are racing toward smaller features and more complex geometries. Quantum computers, in particular, demand a precision of structural control that is still not fully understood. If you cannot see what your fabrication process is actually producing at the atomic level, you are, in some important sense, still flying blind. The mouse bites are small. But now that engineers can finally see them, they can begin to work out which bites to take seriously and which don’t much matter, and start designing the processes that leave the walls of those tiny pipes smoother than anything that came before.
Study link: https://www.nature.com/articles/s41467-026-69733-1
ScienceBlog.com has no paywalls, no sponsored content, and no agenda beyond getting the science right. Every story here is written to inform, not to impress an advertiser or push a point of view.
Good science journalism takes time — reading the papers, checking the claims, finding researchers who can put findings in context. We do that work because we think it matters.
If you find this site useful, consider supporting it with a donation. Even a few dollars a month helps keep the coverage independent and free for everyone.