Stacking semiconductor chips vertically in three-dimensional integrated circuits offers a promising pathway to keep computers accelerating, but the compact design introduces severe thermal management challenges. In these multi-layered architectures, operational heat accumulates rapidly within the tight vertical junctions between stacked components. Without an efficient escape mechanism, this trapped thermal energy can migrate directly into neighboring hardware, including sensitive memory modules and adjacent computing units. Managing these localized temperature spikes requires advanced materials that can selectively conduct heat away along their horizontal planes while serving as effective insulating barriers against vertical heat transfer.
To address this challenge, researchers from the Institute of Industrial Science at the University of Tokyo, along with their collaborators, have introduced a computational design strategy known as flipping engineering. Described in a study published in ACS Nano, this method relies on two-dimensional layered materials engineered to conduct heat along their planes while strongly impeding heat flow between adjacent sheets. By reorienting the directional layout of individual atom-thin layers, the framework creates a structured pathway capable of functioning simultaneously as an efficient directional radiator and an insulating protective blanket for compact electronics.
The engineering approach centers on Janus atomic layers, which are specialized two-dimensional sheets characterized by distinct atomic species on their opposing surfaces. Because the top face possesses a different chemical identity than the bottom face, each individual layer in a vertical stack can be oriented either right side up or upside down. Lead author Xin Wu likened this layer-by-layer orientational selection to a game of Tetris, noting that treating the upward or downward orientation of each sheet as a programmable switch establishes a flexible, modular foundation for tailoring overall thermal transport across the finished stack.
In their demonstration, the research team focused on molybdenum sulfoselenide, or MoSSe, a compound constructed with sulfur atoms coating one face and selenium atoms coating the other. When configuring a 14-layer stack of this asymmetric material, the two possible orientations per layer yield 8,192 distinct structural arrangements. Because fabricating and measuring each individual permutation experimentally or evaluating every iteration with exhaustive physical calculations is computationally impractical, the team implemented a machine-learning workflow designed to accelerate the structural search.
The researchers developed an automated pipeline that integrated thermal transport prediction algorithms with stacking optimization routines to isolate high-performing candidates from the large pool of possibilities. By evaluating only about 3% of the 8,192 total configurations, the machine-learning system identified an optimized layer sequence exhibiting exceptional thermal directional behavior. At room temperature, the predicted in-plane thermal conductivity of this specific structure was approximately 760 times greater than its effective cross-plane thermal conductivity, achieving a thermal anisotropy ratio roughly 35% higher than that observed in an ordered crystalline stack of identical thickness.
The physical mechanism responsible for this selective heat steering involves vibrational dynamics across the atomic lattice. Senior author Masahiro Nomura explained that introducing an irregular sequence of flipped orientations disrupts the specific atomic vibrations responsible for transferring thermal energy vertically from one layer to the next. Crucially, this structural variation across the vertical axis leaves the in-plane vibrational modes that transmit heat laterally largely undisturbed, enabling the material to strongly suppress cross-plane thermal transport without sacrificing its capacity to spread heat across the horizontal surface.
While physical synthesis and direct experimental confirmation remain key tasks for ongoing research, the authors noted that existing layer-by-layer fabrication techniques, including specialized methods capable of flipping individual atomic monolayers during assembly, provide a viable technological route to realize these optimized stacks. By demonstrating how data-driven algorithms can efficiently navigate structural design spaces, the study establishes a foundational material design principle to mitigate internal heat accumulation in future three-dimensional semiconductor devices.
The work was conducted at the Institute of Industrial Science at The University of Tokyo (UTokyo-IIS), one of the largest university-attached research institutes in Japan. Founded in 1949, the institute comprises over 120 research laboratories across nearly all engineering fields and includes approximately 1,200 members, divided between roughly 400 staff members and 800 students. Throughout its history, the organization has focused on closing practical gaps separating specialized academic research from commercial engineering implementations.
By combining machine learning predictions with atomic-scale flipping, the research team demonstrated that computational design can rapidly identify configurations that suppress vertical heat leakage while preserving lateral conduction. This data-driven framework provides a targeted approach for evaluating multi-layer Janus materials, offering a concrete design foundation for engineers working to resolve severe thermal bottlenecks in next-generation stacked semiconductor architectures.
Subject of Research: Thermal management in three-dimensional semiconductor chips using flipped Janus atomic layers
Article Title: Flip side: Upside-down atomic layers could steer heat in stacked computer chips
Article References: Flip side: Upside-down atomic layers could steer heat in stacked computer chips. Original publication (Bibliographic details unavailable.)
Image Credits: AI Generated
DOI: Not provided
Keywords: Janus atomic layers, thermal management, three-dimensional integrated circuits, thermal conductivity, machine learning, molybdenum sulfoselenide, thermal anisotropy, semiconductor chips, heat dissipation, stacking optimization, cross-plane thermal transport, in-plane thermal conductivity, Institute of Industrial Science, University of Tokyo, two-dimensional materials, atomic layer assembly, ACS Nano
News Source: Bethany Barker. (November 21, 2026). Flipping Janus Atomic Layers Boosts Directional Heat Flow in Stacked Semiconductor Chips. Scienmag.



