12 Nov 2015 The Future of Tape Dr. Mark Lantz Manager Advanced Tape Technologies Principle Research Staff Member IBM Research - Zurich © 2013 IBM Corporation Outline Introduction: The role of tape in the era of big data The Future of Tape – Tape areal density trends and future scaling potential – New world record tape low cost particulate tape areal density demo of 123 Gb/in2 – Technologies enabling the 123 Gb/in2 demo – Tape technology roadmap The Future of HDD: emerging technologies and volumetric scaling The Future of Optical Disk Conclusions 1 © 2013 IBM Corporation The data deluge ~48% CAGR HDD Areal Density Scaling 80% of all files created are inactive – no access in at least 3 months! 2 3 Source: D. Anderson, 2013 IEEE Conf. on Massive Data Storage © 2013 IBM Corporation Tape advantages for long-term storage Very energy efficient: no power needed once data is recorded Very secure: – Data is inaccessible when cartridge is not mounted – Drive level encryption – Portable Very long expected media lifetime (30+ years) Very reliable: Typically no data loss in case of drive failure Main net advantage of tape for archival storage is cost Energy and Storage Systems (1PByte of Data for 1 yr) 500,000 Source: R. Dee, Sun Microsystems lbs of CO2 400,000 300,000 200,000 100,000 4 TA P E O ff lin e M an ua lT A P E V S A M ut om at ed TA P E M A ID X 45 00 S A TA R A ID 0 © 2013 IBM Corporation Tape TCO for Long-Term Archiving – Clipper Group Recent studies from the Clipper Group: 1) Continuing the Search for the Right Mix of Long-Term Storage Infrastructure – A TCO Analysis of Disk and Tape Solutions (15 July 2015) Report # TCG2015006 2) The Impact of LTO-7 on The TCO of Long-Term Storage (15 Sept. 2015) Report #TCG2015008 Investigate 9 year TCO of a 1PB archive that grows to 52 PB (55% CAGR) Main Finding: 6.7x TCO advantage of LTO Tape of Disk 5 © 2013 IBM Corporation Magnetic tape (r)evolution Product / Year: IBM 726 /1952 LTO6 / 2012 TS1150 /2014 Demo 2015 Capacity: 2.3 MBytes 2.5 TBytes 10 TBytes 220 TBytes Areal Density: 1400 bit/in2 2.06 Gbit/in2 6.7 Gbit/in2 123 Gbit/in2 Linear Density: 100 bit/in 385 kbit/in 510 kbit/in 680 kbit/in Track Density: 14 tracks/in 5.35 ktracks/in 13.2 ktracks/in 181 ktracks/in 19.8 cm 6 © 2013 IBM Corporation HDD Areal Density Scaling: Areal density/capacity scaling achieved by shrinking the same basic technology to write smaller and smaller bits on disk 7 Ref: http://www.storageacceleration.com/author.asp?section_id=3670&doc_id=274482 © 2013 IBM Corporation Noise and Magnetic Media Structure ~10 nm Information is encoded in transition edge. Large grains media noise To shrink the size of a bit, we need to shrink the size of the grains 8 If grains become too small, magnetic state is unstable superparamagnetic effect © 2013 IBM Corporation The Superparamagnetic “Limit” Magnetic Media “Trilemma”: Small particles (V) SNR ∝ V HDD has reached the limit of (known) materials to produce larger write fields. Writability Thermal Stability EB ∝ K u ⋅ V H 0 ∝ Ku H 0 < Head Field Technologies to go beyond the superparamagnetic limit: • Two dimensional magnetic recording (TDMR) • Heat Assisted Magnetic Recording (HAMR) • Microwave Assisted Magnetic Recording (MAMR) • Bit Patterned Media (BPM) 9 © 2013 IBM Corporation HDD vs. Tape Areal Density Scaling: IBM-FujiFilm demonstration of 123 Gb/in2 on BaFe tape 123 Gbit/in2 demo ~2025 Goal: Demonstrate the feasibility of tape roadmap for the next 10+ years (Source: INSIC 2012-2022 International Magnetic Tape Storage Roadmap) 10 © 2013 IBM Corporation 2015 Storage Bit Cells and Extendibility Scaled bit cells: Magnified 25x: NAND Flash (3 bits) 17.3 nm x 17.3 nm 2150 Gb/in2 HDD 47 nm x 13 nm 1000 Gb/in2 Optical blu ray (3 layer) ~114 nm x 79 nm 75 Gb/in2 LTO6 Tape 4750 nm x 65 nm ~2 Gb/in2 Jag5 Tape 2210 nm x 49 nm ~6.7 Gb/in2 Demo 140 nm x 37 nm 123 Gb/in2 11 Tremendous potential for future scaling of tape track density Key technologies: improved track follow servo control improved media, reader, data channel © 2013 IBM Corporation Demo Technologies Focus on aggressive track density scaling Require: – dramatic improvement in track following enables track width reduction – reduce reader width from a few microns to 90 nm Ultra narrow reader results in a dramatic loss in read back signal that must be compensated for with – improved media technology require improved writer technology – improved signal processing and coding – improved reader technology 12 © 2013 IBM Corporation Servo pattern design for high areal density demo Main design goal: nm-scale positioning fidelity • Increased azimuth angle increased resolution • Increased pattern density increased servo bandwidth and resolution t s α H d Standard LTO Pattern H = 186 µm, t = 2.1 µm, s = 5 µm α = 6°, d = 100 µm Demo Pattern H = 23.25 µm, t = 1.0 µm, s = 2.4 µm α = 24°, d = 52 µm 4x angle 2x rate Compatible with Future 16 Data Band Tape Format 13 © 2013 IBM Corporation Synchronous servo channel Servo channel decodes the readback signal from the servo pattern and provides position information to the track follow control system Servo channel optimized for p-BaFe improved resolution Optimized servo channel in combination with advanced BaFe media formatted with the 24°demo servo pattern provides nanoscale position information Servo readback signal Servo channel Servo signal ADC Fixed clock frequency Optimum symbol detection Interpolation/ correlation LPOS symbols Reliability estimate Timingbase reference Acquisition, monitoring, and control Lateral-position estimate Tape velocity estimate 14 © 2013 IBM Corporation New H∞ track-follow control system High Bandwidth Actuator Key features – Prototype high bandwidth head actuator – A speed dependent model of the system delay is used for control design – The tape speed is used as a parameter to select the controller coefficients – Disturbance rejection is enhanced at the frequencies of the tape path disturbances Track-follow control system Actuator Response v tape + − 15 PES v tape Track-follow actuator v tape K uy Track-follow controller G + d LTM − D Delay © 2013 IBM Corporation Prototype tape transport & hardware platform Precision flangeless tape path with grooved rollers & pressured air bearings to minimize disturbances TS1140 electronics card for reel-to-reel control and analog front end FPGA Board: System-on-Chip (SoC) -> Servo channels -> Microprocessor for synchronous trackfollow (TF) servo controller FPGA Board Current Driver FPGA Board Servo Readback (LVDS) D/A Serial TF Servo Controller Microprocessor FPU FPGA SoC 16 Servo Servo Servo Channels Servo Channels Channels Channels USB/ ETH Host PC © 2013 IBM Corporation Track-follow performance on BaFe tape Track width computation based on measured position error signal: PES (INSIC method) σPES = standard deviation of position error signal: measure of track following fidelity Track width = 2*√2 * 3*σPES + Reader Width (Reader Width = 90nm) 10 σ-PES (nm) 9 8 7 σ-PES ≤ 5.9 nm over TS1140 speed range 6 5 4 1.0 1.5 2.0 2.5 3.0 3.5 4.0 tape speed (m/s) Reader Width = 90nm σ-PES ≤ 5.9 nm 17 Track width = 140 nm Track density = 181 ktpi © 2013 IBM Corporation Advanced BaFe Media Technology Key technologies for advanced tape media 1. Fine magnetic particles with high coercivity archival lifetime 2. Smooth surface 3. Perpendicular orientation of magnetic particles 18 © 2013 IBM Corporation Metal particle vs. Barium-ferrite particle Metal particle (MP) Shape Barium ferrite (BaFe) magnetization axis Passivation layer Origin of magnetic energy Material Passivation layer Acicular Hexagonal platelet shaped Shape anisotropy Magneto-crystalline anisotropy FeCo alloy BaO(Fe2O3)6 Oxide Needed Not needed The magnetic properties of BaFe particles are NOT affected by its shape. BaFe particles do NOT need an oxide passivation layer because it is an oxide. The size of BaFe particles can be reduced while maintaining high coercivity. 19 © 2013 IBM Corporation TEM image of fine Barium-ferrite particles Latest MP Demo Tape BaFe Volume :2850 nm3 coercivity:189kA/m(2380Oe) Volume : 1600 nm3 coercivity: 223kA/m(2800Oe) 50nm The volume of barium ferrite particle used in the demo tape is 45% smaller than the latest MP, reducing media noise and improving SNR 20 © 2013 IBM Corporation SEM Image of tape surface Latest MP tape 123Gb/in2 demo tape Barium ferrite particles are well isolated and packed with high density. 21 © 2013 IBM Corporation Particle volume vs. coercivity 123 Gb/in2 demo Coercivity (kA/m) 250 BaFe 200 MP 150 100 500 1000 1500 2000 2500 3000 Particle Volume (nm3) • The coercivity metal particles smaller than 3000nm3 decreases with size • The coercivity of barium ferrite particles can be tuned independently of size enabling small particle media with long archival lifetime • BaFe particles as small as 1000nm3 have been developed indicating the further scaling potential of BaFe tape 22 © 2013 IBM Corporation Surface profile Latest MP tape TS1150 JD tape 123Gb/in2 Demo tape Optical interferometry roughness 180 µm 240 µm Ra 2.0nm Ra 1.6nm Ra 0.9nm Ra 2.0nm Rz 34nm Ra 1.8nm Rz 27nm AFM 40 µm 40 µm Ra 2.4nm Rz 40nm Reduced surface roughness of demo tape increases the media SNR 23 © 2013 IBM Corporation Perpendicular orientation Longitudinal orientation (MP tape) Random orientation (TS1140 JC and TS1150 JD tape) Highly perpendicular orientation (123Gb/in2 demo tape) The perpendicular orientation of BaFe particle provides a strong increase in SNR 24 © 2013 IBM Corporation Read/write performance 25 123Gb/in2 demo tape SNR (dB) 20 15 10 Latest MP tape TS1150 JD tape 5 0 150 200 250 Linear density (kfci) 300 The combination of small particle volume, smooth surface and perpendicular BaFe particle orientation provide a major increase in SNR. 25 © 2013 IBM Corporation Enhanced Write Field Head Technology Magnetic Media “Trilemma”: Small particles (V) SNR ∝ V 19 Writability EB ∝ K u ⋅ V H 0 ∝ Ku H 0 < Head Field SNRa (dB) Thermal Stability 18 Std Writer HM Writer 17 16 15 14 IBM developed a new high moment (HM) layered pole write head that produces much larger magnetic fields enabling the use of smaller magnetic particles 26 13 182 235 263 Coercivity (kA/m) 294 Increasing media coercivity © 2013 IBM Corporation Iterative decoding C1 Parity Data Read channel C1 ECC Decoder C2 ECC Decoder C2 Parity 0 10 With EPR4 detection 4·10-2 byte error rate ≈ 10-2 bit error rate -5 10 byte error rate A user byte-error rate of 10-20 can be achievable using two C1-C2 iterations with a byte error rate of ≈ 4·10-2 at the output of the detector N1=240 t1=5 N2=192 t2=12 dr=0 er=0 -10 10 undec C1-o1 C2-o1 C1-o2 C2-o2 C2-o3 capacity -15 10 Require SNRa ≈ 10.5 dB at the input of the detector to achieve a raw bit error rate < 10-2 at the output of the detector 27 -20 10 -1 10 -2 10 channel byte error rate -3 10 © 2013 IBM Corporation Recording performance of BaFe with High moment writer & 90 nm GMR Reader SEM image of GMR reader SNR limit Reader Width = 90nm Byte error rate limit Advanced BaFe supports a linear density of 680kbpi with a 90nm reader and provides an operating margin of ~ 0.5dB SNR 28 © 2013 IBM Corporation Summary of demo results Advanced Perpendicular BaFe medium Linear density = 680 kbpi w/ 90 nm reader (single-channel recording) 1-sigma PES = 5.9 nm, Track density = 181 ktpi (track width = 140 nm) Areal recording density : 123 Gb/in2 61x LTO6 areal density 220 TB cartridge capacity (*) This demonstration shows that tape technology has the potential for significant capacity increase for years to come! (*) 220 TB cartridge capacity, assuming LTO6 format overheads and taking into account the 48% increase in tape length enabled by the thinner Aramid tape substrate used 29 © 2013 IBM Corporation INSIC 2012-2022 Tape Roadmap Note: 2015 INSIC Tape Road Map should be available by the end of 2015. INSIC Roadmap available at: http://www.insic.org/news/2012Roadmap/news_12index.html 30 © 2013 IBM Corporation The Future of HDD 31 © 2013 IBM Corporation HDD Areal Density Scaling: Areal density/capacity scaling achieved by shrinking the same basic technology to write smaller and smaller bits on disk 32 Ref: http://www.storageacceleration.com/author.asp?section_id=3670&doc_id=274482 © 2013 IBM Corporation Technologies to continue HDD scaling Two Dimension Magnetic Recording (TDMR) Heat Assisted Magnetic Recording Use multiple readers for adjacent track interference cancelation ~10% AD gain -Requires new read head technology and new read channel Heat Assisted Magnetic Recording (HAMR) Laser used to locally head media to lower the magnetic field required to write a bit - New media, thermal stability of overcoat/lubricant, - Confinement of heat, low cost integration of laser - Life time of laser / near field transducer Patterned Media Media pattern into islands 1 bit per island=grain, island relative large and hence more thermally stable Bit Patterned Media - Develop low cost nano-patterning technology - Planarization / head flying, timing of write process - Track follow servo TDMR has limited gains, HAMR & BPM are disruptive and require large sustained engineering efforts 33 © 2013 IBM Corporation Next Gen HDD Technological Readiness 2015 TMRC (The Magnetic Recording Conference) Technology Survey (Attended by Scientists and Engineers from HGST, WD, Seagate,….. and Academia) 149 data points, (120 from HDD Industry) BPM MAMR TDMR HAMR HDMR TDMR: 2016-17 HAMR 2018 MAMR: Never BPM: Never HDMR: > 2021 (BPM + HAMR) TDMR will likely ship by 2017 (but only provides small AD gains) HAMR will likely ship by mid 2018 HDD scaling will continue to be slow until at least mid 2018 34 © 2013 IBM Corporation Recent Capacity Scaling of HDD: Volumetric Density Slow down in areal density scaling partially compensated by adding more disks: conventional technology has reached space limit (~5 platters) Helium filled drive less turbulence thinner disks higher capacity – WD 6TB (2013) 6 platters – HGST 10TB Drive (2015) 7 platters w/ shingled magnetic recording Doesn’t scale: No space for more heads and platters! 35 © 2013 IBM Corporation Tape and HDD Projections HDD/Cartridge Capacity (TB) Assumptions: • no room to continue adding platters (not clear if HAMR is compatible with 7 platters) • HDD capacity will be driven by areal density scaling (10% CAGR, 25% CAGR w/ HAMR) (**Seagate projection: 20TB drive by 2020) 100 INSIC Tape Roadmap Enterprise Tape Products HDD Projections HDD Products 10 HAMR? volumetric scaling 2009 2011 2013 2015 2017 2019 2021 Year Cost advantage of tape will continue to grow 36 **Ref: http://www.seagate.com/about/newsroom/press-releases/HMR-demo-ceatec-2013-pr-master/ © 2013 IBM Corporation The Future of Optical 37 © 2013 IBM Corporation The Future of Optical Recording Facebook Blu-Ray storage Rack 1.5PB on ~15000 discs Time to load media ~30+ s 1PB on ~10000 discs Time to load media ~30s 38 HL100-RL Optical Library 0.378PB per rack http://venturebeat.com/2015/03/10/sony-whips-up-a-box-to-store-1-5-petabytes-of-data-on-10000-blu-ray-discs/ http://www.pcworld.com/article/2092420/facebook-puts-10000-bluray-discs-in-lowpower-storage-system.html © 2013 IBM Corporation Optical Recording: Historical Scaling Multi-layer Blu-ray 0.7 GB 4.7 GB 15 GB Scaling the wavelength of light (λ) 39 25 GB Semiconductor laser wavelength limit "Comparison CD DVD HDDVD BD" by Cmglee - Own work. Licensed under CC BY-SA 3.0 via Wikimedia Commons http://commons.wikimedia.org/wiki/File:Comparison_CD_DVD_HDDVD_BD.svg#/media/File:Comparison_CD_DVD_HDDVD_BD.svg 128 GB 4 layers max ~4x capacity © 2013 IBM Corporation Archival Disc Roadmap 2x capacity 2x cost Sony and Panasonic announced on March 10, 2014 © 2013 IBM Corporation Optical Land / Groove Recording 41 © 2013 IBM Corporation Areal Density Trends Areal Density 1000 Proposed roadmap 2 sided, 3 layer 30% CAGR?? 100 BluRay 10 BluRay 3 layer 100GB DVD Gb/in^2 CD (1982) 33% CAGR 1 HDD Tape 0.1 0.01 © 2013 IBM Corporation Optical Media Market Total WW Optical Media Demand JIRA Compliation 14000 12000 10000 Units 8000 6000 4000 2000 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 BluRay 2005 DVD 2006 2007 2008 2009 2010 2011 2012 2013 2014 CD © 2013 IBM Corporation Optical Storage Notes Today BluRay 3-layer media approx. 6x greater $/TB than tape BluRay datarates varies from 10x-20x slower than tape BluRay 3-layer disc library requires ~2x-10x more floor-space than tape (for same capacity) BluRay library access time ~30 seconds, similar to tape BluRay bit error reliability 1E7 worse than tape: Requires erasure coding for data integrity BluRay library temperature and humidity specs similar to tape library BluRay supports small file sizes Tomorrow No advances in raw areal density planned, attempts to squeeze margins, multi-level recording $/TB of optical will be greater multiple of tape than today No plans for more than 3 layers for R/W Blu-Ray Multi-layer disc does not improve $/TB – mfg cost is in deposition layers & stamping E.g. 3-layer disc more than 3x single layer due to media complexity/yield Dual Side plans 2x single sided disk back to back Consumer volumes of BluRay are declining due to bandwidth into homes Consumer volumes decline larger than potential increase in archive volumes © 2013 IBM Corporation Summary: The era of big data is creating demand for cost effective storage solutions Tape remains the most cost-efficient and greenest technology for archival storage and active archive applications Tape has a sustainable roadmap for at least another decade – 123 Gbit/in2 areal density demo shows feasibility of multiple future tape generations – Potential exists for the continued of scaling of tape beyond 123 Gbit/in2 The cost advantage of tape over HDD and optical disk will continue to grow 45 © 2013 IBM Corporation