Inc. 3165 Kifer Road Building-B Cafeteria Santa Clara
Reliability physics of the complex memory sub-system of modern, robust solid state storage devices (SSDs) under throughput acceleration stress can be analyzed leveraging Machine Learning – towards understanding their inherently designed fault-tolerance schemes that mitigate expected memory degradation mechanisms through their reliable warranty life as a system. Given the strength of multiple designed error-management schemes effectively countering multiple memory degradation mechanisms under stress, the developed empirical data based Machine Learning framework allows inferential and predictive assessments on reliable SSD design at system-level, in a quantitative and pro-active manner. Such Machine-Learned quantitative assessments on the system-level health of individual devices can be utilized towards assessing dynamic throughput stress impact on design, managing qualification reliability assessments and/or associated decision-making on the reliability of individual and populations of solid-state storage devices/systems. In this talk, the first published description of this research will be discussed, along with the context of independent research affirming the fundamental physics and related Machine Learning application on SSD device health.