A machine-learned superionic, Li-metal-compatible, and cost-effective halide electrolyte for all-solid-state Li metal batteries

Halide solid electrolytes are promising for high-voltage cathodes, yet a combination of high ionic conductivity, robust Li metal compatibility, and cost-effectiveness, is fundamentally challenging to achieve. Here, assisted by a generative machine learning approach, we identify a Ti/F co-doping strategy in the cost-effective Li2ZrCl6 system to improve both ionic conductivity and Li metal compatibility. Substituting Zr with the more electronegative Ti induces a powerful inductive effect to weaken Li+-anion coupling, thereby significantly reducing the a-b plane migration barrier and transforming the native 1D ionic transport into a 3D percolating network. The resulting electrolyte, Li2.1Zr0.9Ti0.1Cl5.7F0.3, achieves a room-temperature ionic conductivity of 2.9 mS cm⁻1 and robust Li metal compatibility with long-term cycling stability exceeding 2000 h at 1 mA cm⁻2 in Li symmetric cells. Moreover, the electrolyte exhibits remarkable air stability under dry-room condition and significantly low materials cost, validating its great potential for large-scale production. This unique combination of high ionic conductivity and robust Li metal compatibility enables the all-solid-state Li metal batteries with uncoated LiNi0.8Mn0.1Co0.1O2 cathode to realize stable cycling performance (86.5% capacity retention after 1000 cycles at 1 C in mold-type cells) and high areal capacity (3.6 mAh cm⁻2 in pouch-type cells), demonstrating a viable path towards practical, high-energy-density devices. This work establishes a new paradigm for rapidly and efficiently screening key target materials for practical device applications.