areal density growth - thic · giora j. tarnopolsky © 2003 \march ’03\13 areal density growth...

$: Areal Density Growth - THIC · Giora J. Tarnopolsky © 2003 \March ’03\13 Areal Density Growth TARNOTEK THIC - The Premier Advanced Recording Technology Forum 0 100 200 300 400$
Giora J. Tarnopolsky© 2003 \March ’03\1 Areal Density Growth TARNOTEK

THIC - The Premier Advanced Recording Technology Forum

Areal Density Growth:

Giora J. TarnopolskyTarnoTek - www.tarnotek.com

P.O.Box 519, Palo Alto, CA 94302-0519Phone: (650) 823-6852 Fax: (650) 322-3160

[email protected]

Is it the manifest destinyof the hard-disk drive?

Presented at the THIC Meeting at the Sony Auditorium,3300 Zanker Rd, San Jose CA 95134-1940

March 4-5, 2003



Single-Slide Summary

Areal Density

"HD

D P

rodu

ct D

esira

bilit

y"

There is an optimumAreal Density for whichthe HDD ProductDesirability is highest.

At the optimum AD,the capabilities of theHDD as a high-capacity,high-data-rate device thatassures data perman-ency, ruggedness, andlow manufacturing anddevelopment costs arefully manifested



Two-Slide Talk: Conclusions

Areal Density

Cap

acity

, Effi

cien

cy

EfficiencyAreal DensityProduct Capacity

As the data record-ing density increases,storage efficiency iscompromised.

ECC, servooverhead, mechanism

The users’ capacitywill not grow with the“native” areal density of1/(bit dimensions)

Diminishingperformance returns



Outline

Introduction - at a brisk paceEvolution of useful productsTradeoff: feasibility vs. desirability

Developments 1990 - 20021 Terabit/in2 proposalsCapacity scalingThe value add of A.D.



IntroductionAreal density has grown from 1 Gb/in2 in 1990to ~150 Gb/in2 by November 2002Feasibility assessments of 1 Tb/in2 have beenpublished, stated INSIC Goal10, 50 Tb/in2 have been suggested for HAMR

HDD future predicated oncontinuing growth of the

areal density



IssuesData storage at 1 Tb/in2 is possibleMaterials and systems research for 1 Tb/in2 isscientifically sound

These are not in question!Is it desirable to accomplish

≥ 1 Tb/in2 products?What constraints on user features

does 1 Tb/in2 impose?What is the return on investment?



265 Tb/in2 virtually demonstrated

S. Heinze, M. Bode, A. Kubetzka, O. Pietzsch, X. Nie, S. Blügel,R. Wiesendanger, Science 288 (2000) 1805-1808: Real-Space Imaging

of Two-Dimensional Anti-ferromagnetism on the Atomic Scale

R. Wiesendanger & M. Bode, Solid State Communications119(2001) 341-355: Nano- and atomic-scale magnetism studied byspin-polarized scanning tunneling microscopy and spectroscopy

Single atomic layer of Mn onW(110) forms a two-dimensionalantiferromagnet.

These are magnetic monolayers ofchemically equivalent atoms, whereadjacent atoms at nearest-neighbor siteshave opposite magnetic moments.

The single-atom spin has beenresolved by Spin-Polarized Scan-ning Tunneling Spectroscopy

AD ≥ 265 Tb/in2 for a 12-atom bit

non-magneticW tip

magneticFe tip

Full imagesize is2.7 x 2.2 nm2

4.5 ± 0.1 Å

… and



… the drive is … “just engineering”

Ready for Fry’s?



Nearly Perfect Inventions

Some inventions are born perfectThis assures their permanency …… and defines the domain of development

Examples of perfect inventions are the bicycle,the umbrella, the book, and the disk drive



Gyroscopic effect assures stability of the riderUnder torque T, the bike turns but does not fall

Low ratio of vehicle mass to rider mass~ 15 % (as compared to ~2,200% for car)

EfficientRuggedMass-producedAffordable

Nearly Perfect Inventions: Bicycle

Lr

dtLdr

Tr



2-D travel with only one linear motionHigh volumetric densityRandom accessMass-producedNon-volatileAffordableRugged

Disk Drives - Nearly Perfect Invention

Few-hundred $/boxNo vibration isolation, no T stabilization

These properties define drives



Areal Density and Storage CostsAreal Density BenefitsFewer components

HeadsDisksSpindles

Volumetric density

No AD BenefitData storagemanagement (s/w & h/w)

Data permanencyRuggednessElectronicsInterfacesCapital & Test Equip.



0

100

200

300

400

500

0 80 160 240Capacity (GB)

Reta

il Pr

ice

($)

HITACHIIBMMAXTORSEAGATEWESTERN DIGITAL

Source CDW, 2/2003

Cost vs. Capacity, 7200 rpmChart shows current retailunit prices (not best price)Thus, a 1 Tbyte driveshould cost about $2,000.Right?Not quite!No. 1: The “box” wouldnot sell above

Price =$ (n × 100),where n < 5

No. 2: For larger capacities, the cost ofmanaging the data exceeds the H/W cost

“The price the customer is willing to pay determines allowable costs.” P. Drucker



1 Tbyte Storage Example - early 2003Fantom Drive 960 GB G-Force RAID LX

8 x 120 GB EIDE drives, 7200 rpmHost: Any SCSI host computer, 160 MB/sSpecial vibration or temperature handling: NoneHot-swappable everything (RAID 1,3,5)MTBF: 500,000 hs

Retail on 2/24/03: $7,609.44EIDE drives’ @ Fry’s: $960.00

Non-AD cost: 87%O/S independent, data management, redundancy, profit ...

960GB, 8-drive, 8-bay,Ultra3 SCSI (Ultra160/m),

EIDE backplane,rackmount RAID

(RAID levels 0,1,0+1,3 and 5) hot swappable drives



From 1 to 150 Gb/in2: 1990 - 2002Extraordinary

achievementLog- log plot of track

density and linear densityThe hyperbolae are lines

of constant areal densityThe rays from the origin

are lines of constant bitaspect ratio

Low AD demos hadbetter BER than recentdemos

From about 1 Gb/in2, theBAR has changed from~ 25 to ~ 3 to 4Linear Density (kbpi)

Trac

k D

ensi

ty (k

tpi)

Hitachi PMRFujitsuIBMSEGReadRiteHitachi Long.

© G Tarnopolsky 2002

20

BAR

1

2

4

812

16

250

100

200

806050

30

20

10

20

10

30

4050

75

100

200

Gb/in2

200 1000800600100 400300

P

PP

PP

/ Perp

5

25

15



Linear Density (kbpi)

Trac

k D

ensi

ty (k

tpi)

Hitachi PMRFujitsuIBMSEGReadRiteHitachi Long.WoodMallaryGao & BertramVictora

© G Tarnopolsky 11/2002

20

BAR1

24

8

12 16

600

300500

200

100

50

3020

10

2010

30405075

100

200

Gb/in2

P P

P

/ Perp

5

25

200 1000700100 500300 2000 3000

1k

500

P

P

From 0.1 to 1.0 Tb/in2

At the 2000 MMMConference, PMR resultsbetween 100 to 145Gb/in2, and longitudinalresults at ~ 150 Gb/in2

were shownFour 1 Tb/in2 studiesR. Wood et al., IBMM. Mallary et al., MaxtorGao & Bertram, UCSD-

CMRRR. R. VictoraVictora, UM - MINT, UM - MINTThese studies explore

the regime of ~ 2 Mbpi &~ 500 ktpi



Early Demos, 1 Tb/in2 StudiesC. Tsang et al., IEEE Trans. Mag., vol. 26, p. 1689, Sept. 1990. T. D. Howell et al., ibid., p. 2298.

J. Hong et al., IEEE Trans. Mag., vol. 38, p. 15, 2002. (Fujitsu demo)

Z. Zhang et al., IEEE Trans. Mag. vol. 38, p. 1861, Sept. 2002. (Seagate demo)

F. Liu et at, presented at InterMag 2002, Amsterdam, The Netherlands, April 2002 (ReadRite-MMC demo.)

IBM Travelstar 80GN, Model No. IC25N080ATMR04, Nov. 2002.

47th MMM Conference 2002, announcements by Seagate & ReadRite.

R. Wood et al., IEEE Trans. Mag., vol. 38, p. 1711, 2002.

M. Mallary et al, IEEE Trans. Mag., vol. 38, p. 1719, 2002.

K. Gao and N. Bertram, to be published IEEE Trans. Mag., TMRC 2002, Santa Clara, CA, USA,August 2002.

R. Victora et al., IEEE Trans. Mag., vol.38, p. 1886, Sept. 2002.

M. Kryder et al., presented at TMRC 2002, Santa Clara, CA, USA, August 2002.



0

100

200

300

400

500

600

700

800

900

1000

0 20 40 60 80 100 120 140 160Areal Density (Gb/in2)

Line

ar D

ensi

ty (k

bpi)

0

50

100

150

200

250

300

350

400

450

500

Trac

k D

ensi

ty (k

tpi)

HitachiFujitsuIBMReadRiteSEGSEG PerpRR Perp

P

PP P

PP

P

P

P


P

How the Areal Density Was WonLinear and track

density vs. arealdensity

Track densityincreased by ~ 35,caused most ofthe areal densitygain.

Enabled by theadvent of spinvalve and GMRheads, advancesin head fabric-ation techniques,media SNR

The inflection point inThe inflection point inthe linear density curvethe linear density curvereflects media SNRreflects media SNRbarriers, the onset ofbarriers, the onset ofthermal decay, andthermal decay, andmagnetic spacingmagnetic spacing



Areal densitydemonstrationsare rigorous,comprehensiveassessments ofthe technology

The BER hasworsened withincreasing arealdensity ofdemos

The 100 ~ 200Gb/in2 regime isan importantresearch field

A Moving BER Target

-11

-10

-9

-8

-7

-6

-5

-4

-3

-2

-1

0

0 20 40 60 80 100 120 140 160Areal Density (Gb/in2)

Log(

On-

trac

k B

it Er

ror R

ate)

IBMFujitsuHitachiReadRiteSeagateSeagate PerpRR Perp




0

100

200

300

400

500

600

700

800

900

0 20 40 60 80 100 120 140

Areal Density (Gb/in2)

Line

ar D

ensi

ty (k

bpi)

0

10

20

30

40

50

60

70

80

90

Hea

d-M

edia

Sep

arat

ion

(nm

)

HitachiFujitsuIBMReadRiteSEGSEG Perp

P

P

PP

PP

Head-Media Separation: Downtrack Story

As therecordingwavelengthdecreased, theHMS shrunkfrom ~ 80 nm to~ 10 nm

Extraordinaryefforts put inreduction ofpole-tiprecession,carbon overcoatthickness, flyingheight and lube



0

20

40

60

80

100

120

140

160

180

200

0 20 40 60 80 100 120 140 160 180 200Inverse Flying Heigth (µm-1)

Are

al D

ensi

ty (G

b/in

2 )

0.0

0.3

0.6

0.9

1.2

1.5

1.8

2.1

2.4

2.7

3.0

Loss

Fac

tor 2

π d

/ λ

Areal Density 1990 - 2002Spacing Loss Factor k*d1 Tbit k*d valuesPoly. (Areal Density 1990 - 2002)

1 Tb/in2 k*d arrowsGao & BertramMallary et al.Wood et al

Flying Height et al.: Finite ResourcesAbscissae are

inverse flyingheight, an open-ended scale

100 Gb/in2

demo’s FH = 60 ÅFlying height is

but one componentof the head-mediaseparation

e-(2π•HMS/λ) spacingloss factor. kdplotted

50 nm or ~100 atomic

layers5 nm or

~ 10atomiclayers

0

5

10

15

0 250 500 750 10001/FH (µm-1)

HM

S (n

m)

Head-MediaSeparation

10 Å20 Å



1 Tb/in2 TMR: A Different Future

Distribution of Lattice Constants

0

5

10

15

0 0.5 1 1.5 2Lattice Constants (nm)

Freq

uenc

y

Elements Ferrites Garnets 1-σ TMR

In the 1 Tb/in2

regime, 500 ktpi, the1σ TMR (trackmisregistration) isabout 1.5 nm

The permissible TMRis of the order ofmagnitude of latticeconstants

This is not the caseat 1, 10, or 100 Gb/in2



Capacity will not grow linearly with ADThe capacity of a product will not grow linearlywith areal density

The limited SNR of high AD demands higher ECCoverheadThe limited PES (position error signal) SNR requireshigher servo overheadTMR of O(nanometer) requires smaller arm lengthsand platter diameters

The effective user areal density becomes muchlower that that given by the bits’ dimensions



ScalingCapacity = areal density x area x ECC efficiencyx servo efficiency and also mechanical limits

Capacity is a parametric function of themechanism.

( ));,,(

)1()1(2 22

mechanismRTDLDfCrRTDLDC servoECC

=−×−×−×××= ψψπ



Parametric dependencies: MechanismConstraints: immunity fromexcessive vibration that causesTMR, and access timeNumber of requests N per unit timefrom the host to the drive

Data throughput leads to accesstimes inversely related to the drivecapacity

τ11~~ =

>< timeaccessfiledatausercapacityN

Otherwise, capacitywasted

Mom was right.Mom was right.30,000 rpm30,000 rpm

isn’t fastisn’t fastenough ...enough ...



Parametric dependencies: MechanismAccess time is a function of the applied torqueto the actuator arm. Large torque produceslarge bending moment in the arm and excitesdeflections from nominal positionImpose a desirable access time, determine thenecessary torque, estimate the magnitude of theamplitude of residual vibration after actuationAmplitude of vibration is measured against trackpitch



Actuator arm mechanismis analogous to

helicopter’s blade

Dover Publishers 1994Princeton Univ. Press 1980

Deflection of flexureacceleration

acce

lera

tion

velocity

velo

city

space

time

arc

leng

th

α

l

R

h

R: disk radiusl: arm lengthh: width at pivota: acceleration at headω: angular velocityα: angles: = lα = arc lengthm massIm: moment of inertia w.r.t.

pivot

Torque for actuator sweep of arc lα cm in τ seconds



Torque;

dtdI

dtdLT mact

ω==

;646

222 mlhlmIm ≈

+=

α

l

R

h

.42τ

αω==

la

dtd

2

3

2

2

332

ταρ

τα AlmlTact ==



Deflection at the head position

TPnEIlA

EIlTs

oo

actact ×===∆ 2

52

61

21

ταρ

The flexure deflection by the bending momentof actuation should be compared against thetrack pitch.

E: Young’s modulusIo: moment of inertia

(cross- section)Α: cross section areaρ: density

α

l

R

h



Capacity vs. Areal Density (I)

Disk radius R ~ arm lengthWith a constraint on the amplitudeof vibration of the arm, the capacitygrows slower than linearly with TD:

52

52

22 16

⋅

=

TDAEInl o

αρτ

535

2262 TDLDA

EInC oact ⋅⋅

=

αρτπ

track density (ktpi)

arm

leng

th (c

m) 10 ms

5 ms

10 30 50 100 300 500

6

5

4

3

2

7

per platter



Non-repeatable PerturbationsA different scaling attains if the drive’senvironment creates a random, non-repeatabletorque of rms value Ťnr

. For instance, in a RAIDTnr has a probability distribution, and a 3-σ TMRevent would happen for a 3-σ Tnr occurrenceThe rms arm deflection is

o

nrnr EI

lTs2

21(

=∆



Non-repeatable Arm DeflectionThe non-repeatable arm deflection is equated to the 1-σ TMR,

The arm length is:

For a constant disk radius, the ambient perturbations mustdecrease linearly with track density. If Ťnr does not decrease,then the disk must become smaller.

TDTMRO

EIlTso

nrnr 30

1)1(21 2

=≈=∆ σ(

nr

o

TTDIEl (⋅⋅

⋅=

3022

∆snr



Capacity vs. Areal Density (II)

Disk radius R ~ arm lengthWith a constraint from irreducibleŤnr , the capacity is:

LDTIETDLD

TTDIEC

nr

o

nr

onr ⋅

⋅⋅

=⋅⋅

⋅⋅

⋅= ((

3022

3022 ππ

nr

o

TTDIEl (⋅⋅

⋅=

3022

10 30 50 100 300 500track density (ktpi)

6

5

4

3

2

arm

leng

th (c

m)

1-σ T nr at 100 ktpi



Capacity vs. Areal Density (III)BAR: 7.2 4τ : 10 5 ms1 Tb/in2 may

not be cost-effective forproducing ahigher capacitydrive withtoday’s low priceand ruggedness0

200

400

600

800

1000

1200

1400

0 200 400 600 800 1000

Areal Density (Gb/in2)

Capa

city

(GB/

plat

ter)

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

Dia

met

er (i

nch)

UnrestrictedActuation, 10 tracksAct, 10 tracks,150 GbNon-repeatable torque, 1-sNon-rep, 1-s, 150 GbDiameter (act)



Capacity ~ (Areal Density)α, 0< α<<1The requirement of a rugged, inexpensivedevice leads to a capacity growth significantlyslower than the areal density growthThe limited SNR expected at ~ 2 Mbpi likewiselimits the capacity growth due to an increasingECC overhead, e.g., 35% at 9.5 dB (rms/rms) inWood et al. analysisThere is a technological and economicaloptimum areal density that would enablepowerful, useful products below 1 Tb/in2



The 20 dB, 200 Gb/in2, $200 Challenge

0.5 Terabyte in mobile format, high rpm & data rateDesirable, rugged, affordable, universally used. Raid 5.

ParametersAD 200 Gb/in2

LD 1000 kbpiTD 200 ktpiSNRsys ≥ 20 dBThermal decay < 0.5 %/decadeRotation ≥ 7200 rpmDisk 65 mmCapacity 120 GB/platterTransfer rate ≥ 250 Mbyte/sOperating T 5 - 55 ° COperating shock 200 GNo slider-level microactuator

DriveCapacity 480 GBPlatters 4Heads 8Weight 155 gRetail price <<200 $

SystemCapacity 3.84 TBDrives 8Physical size

Width 9 inHeight 4.25 inDepth 6 in

Price <2,400 $



The disk “tape” challengePsychological adherence to tapeA disk “tape” cartridge is possible and useful200 GB in LTO/DLT cartridge formatNon-op shock: 10-ft drop, any and all axesHot-swappablePrice = Tape cartridge/2Bandwith = 2 x BW(tape)Shelf-life: indefinite

TAPETAPETAPETAPETAPE

DISKDISKDISKDISKDISK



The value ADD of AD GrowthOptimum areal density is that which delivers:

Data PermanencyAbsolute non-volatility // Fixed content // Regulatory compliance

ReliabilityRuggednessSNRsys

Bandwidth and IOPSProduct Longevity

Areal Density Plateau: Extended product cycles

areal density growth - thic · giora j. tarnopolsky © 2003 \march ’03\13 areal density growth...

Documents