■Multi-level holographic memory

  Archiving storage system for 8K video signal, that has a very high data-transfer-rate and large capacity, are required. We have been researching on holographic memory using multilevel recording to meet these requirements. In FY 2018, we studied a crosstalk reduction method that is effective for amplitude multilevel recording using four-level modulation and a demodulation technology for reproduced data based on machine learning.
  Two-dimensional data page in which symbol pixels of four type luminance are arranged are recorded and reproduced by laser in amplitude four-level holographic memory. Amplitude four-level recording data in holographic memory are recorded and reproduced by laser irradiation. If the luminance value of symbol pixels while data reproduction is different from the value at the time of recording due to noise, an error occurs with reproduced information. The major cause of noise is leaked light between symbol pixels (crosstalk). To reduce this crosstalk, we made symbol pixels smaller and inserted black symbol pixels between neighboring symbol pixels (to reduce the apparent fill factor) (Figure 6-7). The results of recording and reproduction experiments using data pages created by the proposed method demonstrated that the method can reduce the error rate of reproduction signals to a completely correctable level⁽¹⁾.
  For a demodulation technology for reproduced data, we previously developed a demodulation method using convolutional neural networks and demonstrated its effectiveness for two-level recording. In FY 2018, we conducted numerical simulations to study the application of this method to amplitude four-level recording. The results demonstrated that demodulating 3×3 symbol pixels constituting a modulation block by using convolutional neural networks can reduce demodulation errors to 1/4 those of the conventional hard decision method, which makes 0-1 judgment with a single threshold value. We also increased the size of a modulation block to 5×5 symbol pixels to increase redundancy considering the influence of noise from the surroundings. We confirmed that this can further reduce demodulation errors to about 1/10 those of the hard decision method⁽²⁾.
  Additionally, we conducted fundamental experiments on amplitude multilevel recording with multiple exposure by using a holographic memory prototype drive that we developed in FY 2015.
  Part of this research was conducted in cooperation with Hitachi-LG Data Storage, Inc.

■Magnetic nanowire memory utilizing current-driven magnetic nano-domains

  With the goal of realizing a high-speed magnetic recording device with no moving parts and a high reliability, we are conducting R&D on a magnetic recording device that utilizes the high-speed-motion characteristics of nanosized magnetic domains formed in magnetic nanowires. In FY 2018, we developed a process technology for the integration of a recording head on a magnetic nanowire medium to form magnetic domains accurately. We also studied a new recording condition that can reduce currents necessary for magnetic domain formation.
  The integration of a recording head on a magnetic nanowire medium requires a five-layer structure which consists of a marker layer for position alignment ([1]), magnetic nanowire layer ([2]), insulator interlayer ([3]), recording head layer ([4]) and a top electrode layer ([5]) on the substrate. We employed a method that forms patterns in each layer without misalignment and stacks the layers by depositing thin films and lifting unnecessary parts/constituents off repeatedly for making prototype device. We used a laser lithography system, which can uniformly expose a large area, for [1], [3] and [5], which have relatively large structures, while we used an electron beam lithography system for [2] and [4], which have very minute patterns. Since each layer is formed by using both systems alternately, we used a unique alignment marker to suppress the misalignment of the magnetic nanowire and recording device to 40 nm or less.
  Using the Landau–Lifshitz–Gilbert (LLG) equation, which describes magnetization dynamics and damping in general magnetic materials, we analyzed the process of magnetic domain formation when currents applied to the recording head. The results showed that magnetic domain is not stable due to large polarity fluctuations at the both edge of generated magnetic domains if magnetic domains are recorded by a magnetic field generated by applying a current to a single pole wire as a recording head. To address this problem, we devised a recording method that utilizes sudden magnetic field changes generated by applying currents to two recording poles arranged in parallel in opposite directions (Figure 6-8)⁽³⁾. We found that this method could achieve a high-speed and stable magnetic domain recording and halve the current density per recording head.

■Creation of spin-orbit-torque magnetic memory using topological insulator

  We conducted research on the application of topological insulators to magnetic nanowire memory in cooperation with Tokyo Institute of Technology and the University of Tokyo as a commissioned project from Japan Science and Technology Agency for a strategic basic research programs titled "Creation of Core Technology based on the Topological Materials Science for Innovative Devices." A topological insulator is a new material that can produce strong spin torque (rotating power for local magnetic moments) using currents with aligned spins flowing on the surface, though it is an insulator in its bulk. It is expected that connecting bismuth-antimonide (BiSb), which is a topological insulator having a special crystal orientation, with a magnetic nanowire can largely reduce electric power necessary for the magnetic domain driving in magnetic nanowires to about 1/100. In FY 2018, we investigated a method for epitaxial growth between this bismuth-antimonide and magnetic nanowire materials in good crystallite condition.

 

[References]
(1)	T. Muroi, Y. Katano, N. Kinoshita and N. Ishii: "Superimposed Spatial Guard Interval on Data Page for Reducing Inter-Symbol Interference in Amplitude Multi-Level Recording Holographic Memory," Tech. Dig. ISOM '18, Mo-C-03, 2018, pp.13-14 (2018)
(2)	Y. Katano, T. Muroi, N. Kinoshita and N. Ishii: "Demodulation of Multi-Level Data using Convolutional Neural Network in Holographic Data Storage," 2018 International Conference on Digital Image Computing: Techniques and Applications (DICTA), IEEE Xplore, pp.728-729 (2018)
(3)	M. Kawana et al.: "Micromagnetics simulation of magnetic domain formation in magnetic nanowire in various recording element shapes," 42nd Annual conference on magnetics in Japan, 11pPS-30, p.58 (2018) (in Japanese)