Exploring the Compositional Ternary Diagram of Ge/S/Cu Glasses for Resistance Switching Memories

Amorphous semiconductors with tailored ionic and electronic conductivities are central to the operation of emerging resistive memory. However, because of the large amount of potential candidates and compositions, only limited numbers of materials have been tested experimentally. To accelerate the search of efficient solid electrolytes for resistive switching devices, we developed parameters to describe copper-doped germanium sulfides based on ReaxFF, a reactive molecular dynamics framework. The force field was optimized against a training set of first-principle calculations including crystals, amorphous structures, some small molecules, and clusters to describe the atomic interactions among Ge, S, and Cu elements. Based on this novel atomistic model, we studied the mobility of Cu as a function of the ternary composition of amorphous GexSyCuz, and we investigated the corresponding atomic and electronic structures of each solid electrolyte in details. Our analysis led to semiconducting compositions with high Cu mobility and favoring the formation of Cu clusters. Molecular dynamics simulations of switching under an external potential show that devices based on electrolytes with high Cu mobility form thick metallic filaments, and an amorphous copper sulfide phase was observed at the interface. Such an atomistic model is critical to improve our understanding of the atomic mechanism of filamentary growth and can be used to improve retention and endurance of resistive switching devices, which are still limiting their widespread commercial use.

具备定制化离子与电子电导率的非晶半导体，是新兴阻变存储器（resistive memory）正常运行的核心支撑材料。然而，由于潜在候选材料与组分数量繁多，目前仅通过实验测试了有限的材料体系。为加速阻变开关器件高效固态电解质的筛选进程，我们基于反应力场（ReaxFF）这一反应分子动力学框架，开发了用于描述铜掺杂硫化锗的参数集。该力场针对包含晶体、非晶结构、部分小分子与团簇在内的第一性原理计算训练集完成优化，以精准刻画Ge、S、Cu三种元素间的原子间相互作用。依托这一全新的原子级模型，我们研究了Cu的迁移率随非晶GexSyCuz三元组分的变化规律，并详细探究了每种固态电解质对应的原子结构与电子结构。通过分析我们得到了兼具高Cu迁移率与利于Cu团簇形成的半导体组分。外加电势下的开关过程分子动力学模拟结果表明，基于高Cu迁移率电解质的器件会形成较厚的金属导电细丝，且在界面处可观测到非晶硫化铜相。这类原子级模型对于深化我们对导电细丝生长原子机制的理解至关重要，同时可用于优化阻变开关器件的数据保持特性与循环耐久性能——这两项性能目前仍是制约其大规模商业化推广的关键瓶颈。