Fragility crossover in chalcogenide glass-forming liquids: Importance for phase-change memory application
Jiri Orava* & A. Lindsay Greer
Chalcogenide phase-change materials have long been of interest for optical data storage, in which writing is by local melting with a laser pulse to create a glassy mark. Erasure of the mark, heated by a less intense pulse, occurs by nucleation and growth of new crystals within it (for
nucleation-dominated materials, e.g. Ge-Sb-Te alloys) or by crystal growth inwards from the mark perimeter (for
growth-dominated materials, e.g. Ag,In-doped Sb2Te alloys). Transformation mode affects performance: in the growth-dominated case, erasure time depends on mark diameter only, and not on incubation effects. Current interest in phase-change materials focuses on random-access memory, in which switching is by Joule heating, with recording time limited by crystal growth rate. Device scaling and smaller memory cells give extreme transformation conditions: after melting, the cell cools into the glassy state at ~1010 K s–1, and crystallization takes < 10 ns. Operation of such devices is technologically challenging and useful, while opening up interesting fundamental questions.
Phase-change materials are poor glass-formers and conventional kinetic measurements (near the glass-transition temperature
Tg) are far from the regime of fast crystallization relevant for device performance. Recently, it has been found experimentally and confirmed by molecular-dynamic simulations that the nucleation-dominated Ge-Sb-Te glasses can crystallize, around
Tg, up to 105 times faster than would be predicted from temperature dependence of viscosity,
h. Such decoupling, between crystal growth and
h, means that the memory has poor data retention as the glassy mark can crystallize spontaneously at slightly elevated temperatures.
Very recent studies (on the growth-dominated material Ag-In-Sb-Te) give clear evidence for a
fragile-to-strong crossover, similar to what has been claimed in several metallic glass-forming melts. This crossover behaviour may be important in understanding and optimizing memory performance.