Computing with Quantum Valleys to Cut AI's Energy Cost

Overview

Every AI query generates heat. That heat requires water to cool. By 2030, AI data centers are projected to consume up to 12% of all US electricity — and most of that electricity is still generated from fossil fuels. The long-term solution is not just cleaner energy sources. It is more efficient chips.

Valleytronics is an approach to computing that exploits a quantum mechanical property of ultra-thin two-dimensional materials — called the valley degree of freedom — to perform computations with dramatically less energy than conventional silicon chips. In our work published in ACS Applied Electronic Materials (2025), we demonstrated that a WTe₂/MnSe₂ van der Waals heterostructure achieves a colossal valley splitting of 236 meV — one of the highest values ever reported for a 2D/2D system, equivalent in effect to applying an external magnetic field of 787 Tesla. This valley splitting is stable at room temperature and tunable by strain, opening a realistic path toward room-temperature valleytronic devices.

Technical Highlights

• System: monolayer 2H-WTe₂ on monolayer MnSe₂ (van der Waals heterostructure), lattice mismatch 3.1%
• Valley splitting: 236 meV in the most stable (A2) configuration — among the highest reported for any 2D/2D vdW system
• Mechanism: magnetic proximity effect via interfacial orbital hybridization; induced magnetic moment on W atom (0.041 μ_B in A2)
• Berry curvature: confirms broken time-reversal symmetry responsible for lifting valley degeneracy
• Curie temperature: 457.57 K at 6% tensile strain (well above room temperature), with valley splitting retaining 191 meV
• Stability confirmed dynamically (phonon spectrum), thermally (AIMD at 300 K), and mechanically (elastic constants)
• Device design: anomalous valley Hall effect device proposed for experimental realization; spin-valley indices switchable via MnSe₂ magnetization direction
• Applications pathway: quantum computing, neuromorphic computing, integrated photonics

Figure 8: Band structure of the WTe₂/MnSe₂ heterostructure with MnSe₂ substrate magnetization oriented along the (a) +z and (b) −z directions. The yellow arrow in the inset indicates the magnetization direction of the MnSe₂ substrate, with the Fermi energy set to zero. Panels (c) and (d) illustrate schematic experimental configurations for realizing the anomalous valley Hall (AVH) effect under +z and −z substrate magnetization, respectively. Panels (e) and (f) show the experimental setup for the AVH effect based on valley-dependent optical selection rules under +z and −z substrate magnetization.

Supply Chain & Policy Relevance

Post-silicon computing materials like valleytronic semiconductors do not appear on any FEOC supply chain list. WTe₂ and MnSe₂ require tungsten (primarily from China, but also significant US deposits), manganese (globally distributed, including US sources), and selenium (US, Bolivia, China). None of these carry the same supply chain vulnerabilities as the cobalt/lithium/Xinjiang inputs of the current AI hardware paradigm. At the hardware level, more energy-efficient chips reduce the water and electricity demand of every data center running on them. This is the materials science answer to the AI energy problem.

Key Publication