TDMR Performance Gain with Machine Learning Data Detection Channel

Data recovery with cross-track displaced multiple readers (MR), usually referred to as TDMR, is designed to mitigate the performance degradation impact due to transition curvature, PMR track edge effects and inter-track interference [1] . However, relatively large skew angle variation across disk surface from ID to OD makes physical model based two-dimensional channel equalization very challenging [2] . We have developed a machine-learning (ML) based data detection channel utilizing convolutional neural networks (CNN) for hard disk drives and have expanded it for data recovery with multiple readers [3] , [4] . Using sampled waveform directly captured from an actual commercial drive, we have conducted a bit error rate (BER)-based performance comparison between the ML/MR channel and a conventional channel with multiple readers. The study shows that the ML/MR channel can enable significant multi-reader gain.