Work

Interpretable data-driven design of Ge-Sb-Se-Te thin films for energy-efficient and multilevel nonvolatile photonic memory

Year: 2026

Type: article

Abstract: The rapid advancement of in-memory computing and nanophotonics demands advanced non-volatile photonic memory materials capable of multi-level switching and exceptional thermal stability for reliable high-density data storage. Chalcogenide-based Ge-Sb-Se-Te systems have recently gained significant attention for their high optical contrast and superior glass-forming ability. Phase-change materials with superlattice-like structures (SLL) have been shown to exhibit structural tunability, resulting in low-power consumption and multi-level transition characteristics. However, the vast design space and the intricate structural evolution–property relationship pose significant challenges, particularly given the scarcity of high-quality experimental data. In this work, we have developed an interpretable data-driven approach for precise design of Ge-Sb-Se-Te SLL thin films that integrates high-quality experimental data with SHapley Additive exPlanations (SHAP) enhanced machine learning. The composition Ge 0.5 Sb 0.15 Se 0.2 Te 0.15 , offering a large refractive index contrast ( Δn > 1.7) alongside a distinctive multi-stage (Fm-3 m to R-3 m) phase transition is screened rapidly. We further demonstrate that the Se → Te → Ge → Sb sequence enables exceptional amorphous stability, while the Se → Sb → Ge → Te sequence enables precise optical control for programmable multi-level photonic memory. This data-driven framework also offers a transferable methodology that may be applicable to the discovery of functional materials with tailored properties.

Source: Materials & Design

DOI: 10.1016/j.matdes.2026.116922