Grounded Simulation: Using Simulated Evolution to Guide Embodied Evolution

Abstract

Flash memory is a type of non-volatile memory that is being used at an ever-increasing rate in enterprise level storage devices. It is extremely fast (compared to magnetic disks), requires low power and generates very little heat. Unfortunately, it has a relatively short life span, as it wears out over time, compromising its ability to retain data. Previous work has used GAs to tune the control parameters of Flash to trade retention for endurance and, although successful, was prohibitively costly in terms of time, as all testing had to be done in hardware. This chapter describes the next stages in this work, which use a combination of simulated (faster, but less reliable) and embodied (slower, but more reliable) evolution to produce internal parameter sets for Flash memory that increase the endurance by an order of magnitude while still maintaining an industry accepted level of retention.