Plate Nº 34 · recorded November 17, 2026
Chemistry & MaterialsReported finding
Flipping Atomic Layers Could Steer Heat in Stacked Computer Chips
A 760-to-1 heat-flow ratio in a flipped MoSSe stack could help manage heat in vertically stacked chips. Japanese researchers found the design using machine learning.
By Priya Raman4 min read730 words
In brief
- A flipped 14-layer MoSSe stack shows in-plane thermal conductivity ~760 times its effective cross-plane value at room temperature.
- The anisotropy ratio is about 35% higher than an ordered crystalline stack of the same thickness.
- Machine learning found the best of 8,192 possible flipping sequences after evaluating only about 3% of configurations.
- The study, led by Xin Wu with senior author Masahiro Nomura, was published in ACS Nano (2026), DOI: 10.1021/acsnano.6c07922.

A material whose ability to conduct heat along its layers is roughly 760 times greater than its ability to conduct heat through them has been identified by researchers in Japan — and they found it by turning atom-thin layers upside down.
The structure, described in a study recently published in ACS Nano, was designed by researchers from the Institute of Industrial Science at The University of Tokyo and their collaborators. They call the design method "flipping engineering": a way of building layered materials that readily carry heat sideways along their layers while strongly blocking heat flow between them.
The finding addresses a practical problem at the heart of modern computing. In three-dimensional integrated circuits, semiconductor chips sit stacked vertically rather than side by side. The arrangement keeps computers getting faster, but heat builds up between the layers and can travel from high-power computing chips into neighboring components such as memory. Engineers therefore need materials that behave like a radiator in one direction and an insulating blanket in the other.
How do you flip an atomic layer?
"Our approach uses Janus atomic layers, whose two faces consist of different atomic species," says the study's lead author, Xin Wu. "Because the top and bottom are different, we can choose the orientation of each stacked layer individually, right side up or upside down, almost like a game of Tetris. This gives us a simple, programmable switch for designing the structure."
Janus layers, named after the two-faced Roman god, have a different chemical composition on each face. The team focused on MoSSe, a molybdenum-based material with sulfur atoms on one face and selenium atoms on the other. Because the two faces differ, each layer in a stack can point either way, and every choice changes how the stack handles heat.
Why did the search need machine learning?
The combinatorics quickly become unmanageable. A 14-layer stack of MoSSe can be arranged in 8,192 different flipping sequences — far too many to test one by one, whether by simulation or experiment.
Instead, the researchers built a machine-learning workflow that combined two tasks: predicting thermal transport (how heat moves through a material, carried by atomic vibrations) and optimizing the stacking sequence. The algorithm searched the space of configurations efficiently and homed in on the best candidate after evaluating only about 3% of the possible arrangements.
The winning structure's in-plane thermal conductivity — its capacity to conduct heat along the plane of its layers — is approximately 760 times its effective cross-plane thermal conductivity at room temperature. In plain terms, heat travels along the material about as easily as it does through a good conductor, while hardly any heat crosses from one layer to the next. This ratio of directional heat flow, known as thermal anisotropy, is about 35% higher than that of an ordered crystalline stack of the same thickness.
What makes the flipped sequence so effective?
"The irregular flipping sequence disrupts the vibrations that carry heat between the layers," explains senior author Masahiro Nomura. "Importantly, it leaves the vibrations that carry heat along the layers essentially untouched, so we can selectively suppress cross-plane heat transport while largely preserving in-plane thermal conductivity."
That asymmetry is the whole point. Heat in crystalline materials travels via lattice vibrations (quantized ripples of the atomic lattice, often described as tiny sound-like waves called phonons). A disordered pattern of flipped layers scatters the vibrations moving perpendicular to the stack, like speed bumps disrupting cross-town traffic, while leaving the along-layer routes largely clear.
What happens next?
The results remain a design principle rather than a manufactured device. Fabrication and experimental validation are challenges for the future, and the study's figures come from the team's predictive workflow rather than from a physical prototype.
Even so, the researchers point to existing layer-by-layer assembly techniques — including methods that can flip individual layers — as a possible route to building such structures. If those techniques prove out, flipping engineering could become a new material design principle for thermal management in three-dimensional semiconductor chips, where every degree of heat matters.
The upside of upside-down, in this case, is a 760-to-1 advantage.
Publication details: Xin Wu et al., "Data-Driven Flipping Engineering for High Thermal Anisotropy in Janus Two-Dimensional Materials," ACS Nano (2026). DOI: 10.1021/acsnano.6c07922
via phys.org/rss-feed/ (Source)
More from Priya Raman
Show full bio
Senior reporter covering people, launches and day-to-day practice in general science news at SciBeat.
432 articles
Nearby plates
- Pressure extends superconductivity through a layered crystal in 3D
- Cornell chip-based laser cooling speeds trapped-ion quantum computers
- NSF Launches ELEGANT to Move Lab Semiconductors Toward U.S. Factories
- New Catalogs Chart Quantum Possibilities of Atomically Thin Materials
- One-Atom Glass Conducts Heat in Quantized Steps, Study Finds