CIF:Small:Machine Learning Based Turbo Detection for Two and Three Dimensional Magnetic Recording

Washington State University, Pullman WA

Investigators

Abstract

This project investigates two and three dimensional magnetic recording (TDMR and 3DMR) for next generation hard disk drives. In TDMR, bits are written on two-dimensional patches of a magnetic storage disk, whereas in 3DMR bits are written on multiple disk layers. TDMR is an emerging technology that promises up to an order of magnitude increase in information bits per unit of disk area, without requiring radical redesign of the disk. 3DMR is an even newer technology that has the promise of significant areal information density increases over TDMR. A key problem in TDMR and 3DMR is that, at high densities, some bits are not written to any of the magnetic grains on the disk. Moreover, one must contend with signal dispersion: along-track, across-tracks, and between layers. This project investigates machine learning for turbo detection of TDMR and 3DMR channels at high densities of between 1 and 4 magnetic grains per coded bit. The considered machine learning topics include design of local area influence probabilistic model detectors, recently introduced by the investigators, and design of deep neural network detectors for TDMR and 3DMR. Through established collaborations, the investigators will validate the developed techniques with realistic waveforms and will facilitate technology transfer. The project also includes educational and outreach components. The investigators will work with the Voiland College of Engineering Diversity Programs office to identify potential undergraduate researchers from underrepresented groups to participate in the project. The specific technical objectives of this project are: (i) developing information-theoretic design techniques for deep neural networks, (ii) designing machine learning based media noise predictors for TDMR turbo-detectors, (iii) designing deep neural network detectors that handle both two-dimensional intersymbol interference and media noise, (iv) generalizing the machine learning based detector designs for 3DMR, and (v) evaluating the developed designs with TDMR and 3DMR waveforms derived from realistic micromagnetic simulations, obtained from international collaborators. This work is expected to provide significant progress toward the industry's information density target of 10 Terabits per square inch. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.