Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366712
E. Roddick, Lei Xu, R. Brockie
Some form of heat assistance is likely to be required to achieve >3Tbpsi recording density and continue the long progression of capacity growth of magnetic recording in hard disk drives. To provide useful component design guidance, micromagnetic simulations of heat assisted magnetic recording (HAMR) must include estimates of the areal density of the recording system. In earlier work prepared for Intermag 2020 we investigated the influence of both near-field transducer and medium design on areal density. Exploring a range of medium and NFT designs we demonstrated that once the thermal gradient of the head & medium system is sufficient, the achievable jitter (and hence linear density) is governed by the magnetic properties of the medium. Including read-back parameters, we outlined requirements for HAMR recording systems capable of achieving > 2 Tbpsi in hard disk drives with product margins [1].
{"title":"Energy Barrier Analysis of High Density Hamr Simulations","authors":"E. Roddick, Lei Xu, R. Brockie","doi":"10.1109/TMRC49521.2020.9366712","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366712","url":null,"abstract":"Some form of heat assistance is likely to be required to achieve >3Tbpsi recording density and continue the long progression of capacity growth of magnetic recording in hard disk drives. To provide useful component design guidance, micromagnetic simulations of heat assisted magnetic recording (HAMR) must include estimates of the areal density of the recording system. In earlier work prepared for Intermag 2020 we investigated the influence of both near-field transducer and medium design on areal density. Exploring a range of medium and NFT designs we demonstrated that once the thermal gradient of the head & medium system is sufficient, the achievable jitter (and hence linear density) is governed by the magnetic properties of the medium. Including read-back parameters, we outlined requirements for HAMR recording systems capable of achieving > 2 Tbpsi in hard disk drives with product margins [1].","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127764120","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366713
I. Barsukov, H. Lee, A. Jara, Yu-Jin Chen, A. M. Gonçalves, C. Sha, J. Katine, R. Arias, B. Ivanov, I. Krivorotov
Nanoscale magnets are the building blocks of many existing and emergent spintronic applications, e.g. nonvolatile spin torque memory, spin torque oscillators, neuromorphic and probabilistic computing. Controlling magnetic damping in nanomagnets holds the key to improving the performance of future technologies. Here, we experimentally demonstrate and theoretically corroborate that a ferromagnetic nano-particle (free layer of a magnetic tunnel junction (MTJ) nanopillar) can exhibit spin dynamics qualitatively different from those predicted by the harmonic oscillator model. Nonlinear contributions to the damping can be unusually strong, and the effective damping parameter itself can exhibit resonant dependence on field/frequency [1].
{"title":"Inversion of the Spin-Torque Effect in Mtjs Via Resonant Magnon Scattering","authors":"I. Barsukov, H. Lee, A. Jara, Yu-Jin Chen, A. M. Gonçalves, C. Sha, J. Katine, R. Arias, B. Ivanov, I. Krivorotov","doi":"10.1109/TMRC49521.2020.9366713","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366713","url":null,"abstract":"Nanoscale magnets are the building blocks of many existing and emergent spintronic applications, e.g. nonvolatile spin torque memory, spin torque oscillators, neuromorphic and probabilistic computing. Controlling magnetic damping in nanomagnets holds the key to improving the performance of future technologies. Here, we experimentally demonstrate and theoretically corroborate that a ferromagnetic nano-particle (free layer of a magnetic tunnel junction (MTJ) nanopillar) can exhibit spin dynamics qualitatively different from those predicted by the harmonic oscillator model. Nonlinear contributions to the damping can be unusually strong, and the effective damping parameter itself can exhibit resonant dependence on field/frequency [1].","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132282219","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366716
S. Greaves, R. Itagaki, Y. Kanai
Microwave assisted magnetic recording (MAMR) has the potential to realise large gains in areal recording density. The key component enabling this gain is the spin torque oscillator (STO), which is typically located in the gap between the main pole and the trailing shield of the write head.
{"title":"Effect of spin torque oscillator cone angle on recording performance in microwave assisted magnetic recording","authors":"S. Greaves, R. Itagaki, Y. Kanai","doi":"10.1109/TMRC49521.2020.9366716","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366716","url":null,"abstract":"Microwave assisted magnetic recording (MAMR) has the potential to realise large gains in areal recording density. The key component enabling this gain is the spin torque oscillator (STO), which is typically located in the gap between the main pole and the trailing shield of the write head.","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124447655","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366719
Ahmed Aboutaleb, Amirhossein Sayyafan, B. Belzer, K. Sivakumar, S. Greaves, K. Chan, R. Wood
The hard disk drive (HDD) industry stores data at areal densities close to the capacity limit of the onedimensional (1D) magnetic recording channel [1]. New technologies are emerging to increase density, including heat assisted magnetic recording (HAMR), microwave-assisted magnetic recording (MAMR), and two-dimensional magnetic recording (TDMR). TDMR employs 2D signal processing to achieve significant density gains, without changes to existing magnetic media. Recent encouraging studies [2] –[5] propose multilayer magnetic recording (MLMR): vertical stacking of an additional magnetic media layer to a TDMR system to achieve further density gains. Using a realistic grain flipping probability (GFP) model to generate waveforms [3], [4], we investigate the design of deep neural network (DNN) based methods for equalization and detection for MLMR.
{"title":"Deep Neural Network-based Detection and Partial Response Equalization for Multilayer Magnetic Recording","authors":"Ahmed Aboutaleb, Amirhossein Sayyafan, B. Belzer, K. Sivakumar, S. Greaves, K. Chan, R. Wood","doi":"10.1109/TMRC49521.2020.9366719","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366719","url":null,"abstract":"The hard disk drive (HDD) industry stores data at areal densities close to the capacity limit of the onedimensional (1D) magnetic recording channel [1]. New technologies are emerging to increase density, including heat assisted magnetic recording (HAMR), microwave-assisted magnetic recording (MAMR), and two-dimensional magnetic recording (TDMR). TDMR employs 2D signal processing to achieve significant density gains, without changes to existing magnetic media. Recent encouraging studies [2] –[5] propose multilayer magnetic recording (MLMR): vertical stacking of an additional magnetic media layer to a TDMR system to achieve further density gains. Using a realistic grain flipping probability (GFP) model to generate waveforms [3], [4], we investigate the design of deep neural network (DNN) based methods for equalization and detection for MLMR.","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131974738","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366710
M. Nishikawa, Y. Nakamura, Y. Kanai, H. Osawa, Y. Okamoto
We have previously proposed the waveform equalization using a two-dimensional finite impulse response (TD-FIR) filter [1], [2] and the inter-track interference (ITI) canceller [3] as a signal processing method for shingled magnetic recording (SMR) [4]. In this study, we propose a neural network detector which directly outputs log-likelihood ratio (LLR) as the reliability for the recording sequence from the reproduced waveform and evaluate the channel error rate (CER) performance of the neural network detector in iterative decoding system by computer simulation.
{"title":"A Study on Neural Network Detector in Smr System","authors":"M. Nishikawa, Y. Nakamura, Y. Kanai, H. Osawa, Y. Okamoto","doi":"10.1109/TMRC49521.2020.9366710","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366710","url":null,"abstract":"We have previously proposed the waveform equalization using a two-dimensional finite impulse response (TD-FIR) filter [1], [2] and the inter-track interference (ITI) canceller [3] as a signal processing method for shingled magnetic recording (SMR) [4]. In this study, we propose a neural network detector which directly outputs log-likelihood ratio (LLR) as the reliability for the recording sequence from the reproduced waveform and evaluate the channel error rate (CER) performance of the neural network detector in iterative decoding system by computer simulation.","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124943225","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366711
H. Koike, T. Tanigawa, Toshinari Watanabe, T. Nasuno, Y. Noguchi, M. Yasuhira, T. Yoshiduka, Yitao Ma, H. Honjo, K. Nishioka, S. Miura, H. Inoue, S. Ikeda, T. Endoh
STT-MRAM is now an essential component for future low power consumption electronics. Recently, a number of STT-MRAM developments have been successively disclosed by major LSI vendors [1] –[9], and some of them announced that risk mass-production of STT-MRAM had started. This invited paper reviews, in this opportunity, STT-MRAM circuit design strategies, which cover memory cell design, sense amplifier (S/A) and reference generator (Refgen), and array architecture. Furthermore, as one example of STT-MRAM design, a 128Mb STT-MRAM chip using 40-nm standard CMOS and 3X-nm MTJ technology will be presented [10].
{"title":"Review of STT-MRAM circuit design strategies, and a 40-nm 1T-1MTJ 128Mb STT-MRAM design practice","authors":"H. Koike, T. Tanigawa, Toshinari Watanabe, T. Nasuno, Y. Noguchi, M. Yasuhira, T. Yoshiduka, Yitao Ma, H. Honjo, K. Nishioka, S. Miura, H. Inoue, S. Ikeda, T. Endoh","doi":"10.1109/TMRC49521.2020.9366711","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366711","url":null,"abstract":"STT-MRAM is now an essential component for future low power consumption electronics. Recently, a number of STT-MRAM developments have been successively disclosed by major LSI vendors [1] –[9], and some of them announced that risk mass-production of STT-MRAM had started. This invited paper reviews, in this opportunity, STT-MRAM circuit design strategies, which cover memory cell design, sense amplifier (S/A) and reference generator (Refgen), and array architecture. Furthermore, as one example of STT-MRAM design, a 128Mb STT-MRAM chip using 40-nm standard CMOS and 3X-nm MTJ technology will be presented [10].","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127312680","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366709
Amirhossein Sayyafan, B. Belzer, K. Sivakumar, K. Chan, Ashish James
The hard disk drive (HDD) industry is facing a physical limit on the areal density (AD) of one-dimensional magnetic recording (1DMR) on traditional magnetic media. To increase capacity without media redesign, twodimensional magnetic recording (TDMR) has been introduced. The effective channel model has a media noise term which models signal dependent noise due to, e.g., magnetic grains intersected by bit boundaries. Trellis based detection with pattern dependent noise prediction (PDNP) [1] is standard practice in HDDs. The trellis detector sends soft coded bit estimates to a channel decoder, which outputs user information bit estimates. PDNP uses a relatively simple autoregressive noise model and linear prediction; this model is somewhat restrictive and may not accurately represent the media noise, especially at high storage densities. To address this modeling problem, we design and train deep neural network (DNN) based media noise predictors. As DNN [2] models are more general than autoregressive models, they more accurately model media noise compared to PDNP. The proposed turbo detector assumes a channel model for the k th linear equalizer filter output y(k):
{"title":"Deep Neural Network Media Noise Predictor Turbo-detection System for One and Two Dimensional High-Density Magnetic Recording","authors":"Amirhossein Sayyafan, B. Belzer, K. Sivakumar, K. Chan, Ashish James","doi":"10.1109/TMRC49521.2020.9366709","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366709","url":null,"abstract":"The hard disk drive (HDD) industry is facing a physical limit on the areal density (AD) of one-dimensional magnetic recording (1DMR) on traditional magnetic media. To increase capacity without media redesign, twodimensional magnetic recording (TDMR) has been introduced. The effective channel model has a media noise term which models signal dependent noise due to, e.g., magnetic grains intersected by bit boundaries. Trellis based detection with pattern dependent noise prediction (PDNP) [1] is standard practice in HDDs. The trellis detector sends soft coded bit estimates to a channel decoder, which outputs user information bit estimates. PDNP uses a relatively simple autoregressive noise model and linear prediction; this model is somewhat restrictive and may not accurately represent the media noise, especially at high storage densities. To address this modeling problem, we design and train deep neural network (DNN) based media noise predictors. As DNN [2] models are more general than autoregressive models, they more accurately model media noise compared to PDNP. The proposed turbo detector assumes a channel model for the k th linear equalizer filter output y(k):","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115961808","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366714
Shanwei Shi, J. Barry
A dominant impediment in magnetic recording is pattern-dependent media noise, and its impact will only grow more severe as areal densities increase. The pattern-dependent noise prediction (PDNP) algorithm [1] [2], widely used as an effective strategy for mitigating pattern-dependent media noise in single-track detection, has recently been extended to the multitrack scenario [3]; it uses 2D patterns (spanning multiple tracks) to mitigate both downtrack and crosstrack pattern-dependent noise, based on the architecture shown in Fig. 1. The sampled readback waveforms are filtered by a $2 times 2$ MIMO equalizer with coefficients $mathbf{C}$, whose output $mathbf{y}_{k}=left[y_{k}^{(1)}, y_{k}^{(2)}right]^{T}$ is passed to a 2D -PDNP multitrack detector. Associated with each 2D bit pattern is a signal level vector $mathbf{s}$, a standard deviation diagonal matrix $Lambda$, and a set of matrix-valued predictor coefficients $mathbf{P}_{0}, mathbf{P}_{1}, ldots, mathbf{P}_{N_{p}-1} $. The branch metric of edge e for the 2D -PDNP Viterbi detector is [3]:
{"title":"Minimum-Bit-Error Rate Tuning for 2D-Pdnp Multitrack Detection","authors":"Shanwei Shi, J. Barry","doi":"10.1109/TMRC49521.2020.9366714","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366714","url":null,"abstract":"A dominant impediment in magnetic recording is pattern-dependent media noise, and its impact will only grow more severe as areal densities increase. The pattern-dependent noise prediction (PDNP) algorithm [1] [2], widely used as an effective strategy for mitigating pattern-dependent media noise in single-track detection, has recently been extended to the multitrack scenario [3]; it uses 2D patterns (spanning multiple tracks) to mitigate both downtrack and crosstrack pattern-dependent noise, based on the architecture shown in Fig. 1. The sampled readback waveforms are filtered by a $2 times 2$ MIMO equalizer with coefficients $mathbf{C}$, whose output $mathbf{y}_{k}=left[y_{k}^{(1)}, y_{k}^{(2)}right]^{T}$ is passed to a 2D -PDNP multitrack detector. Associated with each 2D bit pattern is a signal level vector $mathbf{s}$, a standard deviation diagonal matrix $Lambda$, and a set of matrix-valued predictor coefficients $mathbf{P}_{0}, mathbf{P}_{1}, ldots, mathbf{P}_{N_{p}-1} $. The branch metric of edge e for the 2D -PDNP Viterbi detector is [3]:","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128736493","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366718
Jinho Lim, Zhaohui Zhang, A. Garg, J. Ketterson
Magnetization reversals in magnetic recording media are largely carried out by brute force: a field is applied opposite to the existing magnetization direction of some bit that has sufficient magnitude to nucleate a seed that then grows into an oppositely magnetized bit. The fields used are generally quite large, $sim 10$ kG, requiring elaborate magnetic circuitry to keep the fields localized so they do not spill over onto neighboring bits. This situation is to be contrasted with the resonant magnetization reversals performed in NMR spin echo experiments in which r.f. fields of a few Gauss coherently reverse the magnetization in the presence of static fields of a few kG, by applying a so-called Pi pulse; two such pulses restores the original alignment.
{"title":"Simulating Resonant Magnetization Reversals in Nanomagnets","authors":"Jinho Lim, Zhaohui Zhang, A. Garg, J. Ketterson","doi":"10.1109/TMRC49521.2020.9366718","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366718","url":null,"abstract":"Magnetization reversals in magnetic recording media are largely carried out by brute force: a field is applied opposite to the existing magnetization direction of some bit that has sufficient magnitude to nucleate a seed that then grows into an oppositely magnetized bit. The fields used are generally quite large, $sim 10$ kG, requiring elaborate magnetic circuitry to keep the fields localized so they do not spill over onto neighboring bits. This situation is to be contrasted with the resonant magnetization reversals performed in NMR spin echo experiments in which r.f. fields of a few Gauss coherently reverse the magnetization in the presence of static fields of a few kG, by applying a so-called Pi pulse; two such pulses restores the original alignment.","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126071874","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-17DOI: 10.1109/TMRC49521.2020.9366717
Jinlu Shen, B. Belzer, K. Sivakumar, K. Chan, Ashish James
Conventional detection systems in hard disk drives (HDD) typically include a 2D partial response (PR) equalizer that pre-processes the readback signals and shapes the output to a controlled target response, followed by a maximum likelihood (ML) or maximum a posteriori (MAP) detector which outputs log-likelihood ratios (LLRs) to be passed to a channel decoder. Pattern dependent noise prediction (PDNP) algorithm [1] is usually incorporated into the metric computation of the trellis in the ML/MAP detector to combat media noise intrinsic to the magnetic recording (MR) channel. For next generation two-dimensional magnetic recording (TDMR) HDDs, such conventional systems would suffer from impractically large trellis state cardinality when performing multi-track detection, and they may no longer be capable of handling the increased nonlinearities in high density recording channels. This work investigates applying advanced machine learning techniques to TDMR. Convolutional neural networks (ConvNets) are employed in place of the PR equalizer and ML/MAP detector with PDNP to directly process the un-equalized readback signals and output soft estimates. ConvNets are special deep neural networks (DNNs) that assume the inputs are images and perform convolution instead of affine function in the network forward pass [2]. This enables far fewer parameters in ConvNets than regular DNNs of the same depth and therefore allows for deeper networks. The motivation to use ConvNets is the resemblance between data detection problem in MR and typical image processing problems. In MR channels, the write process converts temporal data into spatial patterns recorded on a magnetic medium, which transforms sequential correlation into spatial ISI/ITI. Data detection can be viewed as an image processing problem, proceeding from the 2D image of the shingled bits (see Fig. 1), to higher level abstractions of features by means of convolutional layers that finally allow classification of individual bits. Several variations of ConvNets are compared in terms of network complexity and performance. The best performing ConvNet detector can provide data storage density of up to 3.7489 Terabits/in 2 on low track pitch TDMR channel simulated with a grain flipping probabilistic (GFP) model.
{"title":"Convolutional Neural Network Based Symbol Detector for Two-Dimensional Magnetic Recording","authors":"Jinlu Shen, B. Belzer, K. Sivakumar, K. Chan, Ashish James","doi":"10.1109/TMRC49521.2020.9366717","DOIUrl":"https://doi.org/10.1109/TMRC49521.2020.9366717","url":null,"abstract":"Conventional detection systems in hard disk drives (HDD) typically include a 2D partial response (PR) equalizer that pre-processes the readback signals and shapes the output to a controlled target response, followed by a maximum likelihood (ML) or maximum a posteriori (MAP) detector which outputs log-likelihood ratios (LLRs) to be passed to a channel decoder. Pattern dependent noise prediction (PDNP) algorithm [1] is usually incorporated into the metric computation of the trellis in the ML/MAP detector to combat media noise intrinsic to the magnetic recording (MR) channel. For next generation two-dimensional magnetic recording (TDMR) HDDs, such conventional systems would suffer from impractically large trellis state cardinality when performing multi-track detection, and they may no longer be capable of handling the increased nonlinearities in high density recording channels. This work investigates applying advanced machine learning techniques to TDMR. Convolutional neural networks (ConvNets) are employed in place of the PR equalizer and ML/MAP detector with PDNP to directly process the un-equalized readback signals and output soft estimates. ConvNets are special deep neural networks (DNNs) that assume the inputs are images and perform convolution instead of affine function in the network forward pass [2]. This enables far fewer parameters in ConvNets than regular DNNs of the same depth and therefore allows for deeper networks. The motivation to use ConvNets is the resemblance between data detection problem in MR and typical image processing problems. In MR channels, the write process converts temporal data into spatial patterns recorded on a magnetic medium, which transforms sequential correlation into spatial ISI/ITI. Data detection can be viewed as an image processing problem, proceeding from the 2D image of the shingled bits (see Fig. 1), to higher level abstractions of features by means of convolutional layers that finally allow classification of individual bits. Several variations of ConvNets are compared in terms of network complexity and performance. The best performing ConvNet detector can provide data storage density of up to 3.7489 Terabits/in 2 on low track pitch TDMR channel simulated with a grain flipping probabilistic (GFP) model.","PeriodicalId":131361,"journal":{"name":"2020 IEEE 31st Magnetic Recording Conference (TMRC)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129075073","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：