부산대학교 도서관

Deep neural networks (DNNs) have been used to investigate geometrical scaling of heat-assisted magnetic recording (HAMR). Our investigation considers areal density capability (ADC) sensitivity with respect to distinctive design features. We study HAMR performance by numerical scaling of key parameters including thermal profile spot size and reader dimension. DNNs are trained based on a full recording system model, which includes the writer, the reader, and the channel detector, to accelerate ADC optimization through geometrical scaling. In this article, we show that by providing adequate training data through sampling the entire design space, a deep feedforward neural network can learn the behavior of HAMR and accelerate BER estimation for triple track recording.