硬盘结构及基本知识课件

By Patiwat KamonpetBasic Disc DriveDisc Drive OverviewDisc Drive BasicsMagnetic Recording BasicsRecording ChannelComputer SystemDisc Drive OverviewTodays PC ArchitectureIO BusLogicISA BusOther peripheralsCPUPentium ProMemoryPCI BridgeChipVideo GraphicsAdapter CardInterfaceAdapter CardMonitorIDE or SCSIDisc DriveCableRibbon CablePCI Board EdgeConnectionPCI Board EdgeConnectionPCI BusLocal System BusWired onMother BoardWired onMother BoardWired onMother BoardFilesCollection of BytesTextDocumentComputer InstructionsPictureetc.Sequence of BlocksStored in referenced by a rather than location on disk.Files are managed by the computers operating system.The disk drive has no awareness of files.Storing Files on Disc Drive ComputerDisc DriveHere are 3 Block of DataStart Storing in Location 5ControllerInterfaceAdapterFile:Letter.DOCLETTWR.DOCANOTHER.DOC01234567891011121314151617181920DIRECTORYTransfer Ratein Mega Mytes per second(MBps)How to Access FilesDirectory=A List of and LacationsLETTER.DOC PROGRAM.EXE ANOTHER.DOC.Block location on disk5,6*,7*1024,1025*,1026*,1027*12,13*,14*,15*,16*.The operating system in the computer keeps track of the directory*In DOS,the directory keeps track of the location for only the 1st block of each file.The Table,or FAT,keeps track of the location of the other blocks.HDA ComponentsDISCCIRCULATE FILTERCLAMP RINGOD LIMIT STOPBOTTOM POLEVCMPCCPREAMP CHIPID LIMIT STOPHEADFLEXUREARMPIVOT CARTRIDGEBEARINGTOP VIEWDisc Drive BasicsPCB ComponentsHOST CONTROLLERVCM&SPINDLE CONTROLLERREAD/WRITE CHANNELMICROCONTROLLERSERVO CONTROLLERSRAMDRAMSHOCK SENSORSPINDLE CONNECTORHDA CONNECTORSHOCK ICMass Storage Architecture Using Disc DrivesRead/WriteChannelPositionSystemSPMControlSpindle Spindle MotorMotorVCM (Voice VCM (Voice Coil Motor)Coil Motor)ControllerInterfaceAdapterMemoryCPUPC-AT System PC-AT System Bus(ISA)Bus(ISA)SCSI Ribbon SCSI Ribbon CableCableEmbebbed on mother board or add-in cardBlock DefinitionsREAD/WRITE Detects bits from the signal coming from the CHANNEL head(analog)and converts them into digital bitsPOSITION SYSTEMSeeks to and keeps the heads positioned over the correct track of data on the disk(E-Block-VCM-Servo)SPM CONTROL Keeps the disk rotating and at the proper speed CONTROLLERRecognizes the digital data coming from the Read Channel and organizes it into blocks of bytesUsing Recording Head To Magnetize A FilmFilm MotionCurrentWriting Data On A Magnetic FilmFilm MotionCurrent ReversedTransition ResultsTrackTrack=A strip of data written on a magnetic filmEach bits value is sampled at regular interval:1 when magnetic transition presents0when magnetic transition does not presentTrack WidthSampling PeriodWrite Other Tracks by Moving the HeadFilm MotionTrack DensityTrack WidthTrack PitchTrack Density =Number of tracks that fit in one inch(TPI)Bit Density(Linear Density)Bit LengthBit Density =Number of bits that fit in one inch of track(BPI)Arial Density1”1”Areal Density=The amount of data that can be stored in 1 square inchAD =BPI*TPIReading Data Back by MR Read HeadRun constant current through MR stripe,Measure the resistance.Magnetic field from filmpicked up by stripeField variation in stripechanges the resistanceMR stands for MagnetoResistance.Film MotionProblem with MR StripeThe MR stripe detects the field from a transition a long way away.Solutions:Space the transitions far apart Detect several overlapping bits at a time Use shieldsShielded MR HeadShields permit only the MR stripe to only see the media below the gap.The Voltage Being Picked Up is Not Very HighWall Plug220 VoltsComputer Signals3-5 VoltsFlashlight Battery1.5 VoltsEKG waves on your skin0.01 VoltsTV Signal(picked up by antenna)0.0008 VoltsSignal From Recording Head0.0003 Volts0.0003V0.075VPre-amplify the read signal very close to the headInductive Write MR Read HeadIntegrated Inductive Write MR Read HeadTrack WidthReader GapMagnetic SpacingHead WidthTrack width is determined by head width(approximately equal).Bit length is determined by reader gap and spacing from gap to media and many others.What Controls Density?The rate at which data is read or written through the headmeasured in Million bits per second(Mbps)As Bit Density Increases,So Does Data Rate!Dont confuse data rate with transfer rate,the rate at whichdata transfers over the interface(in Megabytes per secondor MBps)Film MotionData RateMagnetic Storage On A Disc DriveCircular TracksVoice Coil Motor movesthe head in and outSpindle Motor drives the discat constant RPMCalculate Data Rate0.9 r Too big to deal with We break each track into chunks called sectors:Most common sector Size =512 Bytes(1024 and 2048 bytes common)Typical Sectors Per Track =50 to 256 (determined by bit density)Breaking tracks into sectors used up some space-Formatting Efficiency(5%-15%)Constant Angular Recording(CAR)RidRodRadiusData RateRidRodRadiusRidRodRadiusVelocityBPILess dataZone Bit RecordingRidRodRadiusBPIRidRodRadiusRidRodRadiusVelocityData RateZoneMaximize CapacityZoneZone TableConstant Angular Recording CapacityCapacity=number of tracks bits per tracknumber of tracks=TPI (Rod Rid)bits per track=BPI Rid 2RidCaptacity=TPI (Rod Rid)bits per trackConstant Angular Recordingbits per track=constantRidRodRadiusbits per trackAreaZoning Max CapacityZoned Recordingbits per track=2r BPIRidRodRadiusbits per trackCapacity Improvement=(Rod Rid)2 Rid 50%for 3.5”FFZoning Practical CapacityRidRodRadiusbits per trackCapacity Improvement=(Rod Rid)2 Rid(1-N-1)N=number of zones(4 in this example)4 zones 38%improvement8 zones 44%improvement4 zones 47%improvement4 zones 48%improvement Typical zoned drive has 16 zonesFor 3.5”FF drives,the limit to zonings improvement is about 150%Magnetization Curve of MediaHHcDHM Squareness:Coercive-Squareness:Remanence:Saturation magnetization:Coercivity:Slope at Coercivity:Magnetic Recording BasicsLongitudinal Recording Write FieldHeadHeadHx=2000OeHx=2200OeLines of constanthorizontal fieldintensityGap180020002200240026002800The Write BubbleInside write bubbleField Hc of 2000OeStrong enough to magnetize mediaOutside write bubbleField Hc of 2000OeStrong enough to magnetize mediaHeadHeadGap200022002400260028001800Media LayerHc=2000OeWriting a Transition?Media motionTransition written at the trailingEdge fo the write bubbleThis region is magnetized first to the leftand then again to the rightWriting a Transition200022002400260028001800HM Media motionThe media in this area sees1200 Oe in the new direction,Stays magnetized in the old direction!The media in this area sees2400 Oe in the new direction,Being magnetized in the old direction!HcM=0Real Transitions are Blurry!200022002400260028001800HM Media motionIt takes distance on the mediato change the direction of magnetizationThis is called“Transition Length”Transition LengthTransition LengthMHhHcxHMHcxMPrevious state of medium-50%50%Hdtransition length(2a)Horizontal Component of Head FieldDemagnetization Field from the TransitionDemagnetization Field from a TransitionMHdatransition length parameterx+MMHdHdTMrA recorded transition generates demagnetization fieldHdWilliams-Comstock Model of a Recorded TransitionMHdHHcxHMHcDHCalculating The Transition LengthwhereTransition Length Parameter500 Magnetic Spacing3”Media Thickness200 Write Field Gradient Factor(0.75)300 Oe/”Media Coercivity2200 OeRemanence Magnetization7500 GCoercive Squareness80%Typical ValuesFrom Williams-Comstock ModelWriting Shorter Sharper TransitionsMedia motionCloser Head-Media Spacing(HMS)Thinner Media LayerShorter Write Gap LengthTighter Media Switching Field Distribution(all the media switch at the same H)Write FieldGradientMediaSquarenessHigh Write Field Gradient(closer bubbles)200022002400260028001800HMTransition LengthHigh Media Squareness(how steep M-H curve)Reading with a GMR Read HeadBMMBMMvvPhysical Mechanism of GMR EffectM3dFermi levelM4sConduction bandTwo current modelFor normal GMR materialss-d scattering yields energy loss:significantly contributes to resistivity.The number of available 3d states at Fermi surface is different for different spins Physical Mechanism of GMR EffectLow resistance stateMMMMHigh resistance stateScattering of spin electrons occurs within a mono-layer from the interface.Parallel State:Antiparallel StateGMR Read Head Transfer CurveM2M1M2M1q qNon-magneticconductive layerCharacterizing Magnetically Isolated PulsesdT2aPW50GWhereTransition ParameterShield-to-Shield SpacingMagnetic SeparationMedia ThicknessFrom Williams-Comstock modelAchieving Desirable Isolated PulsesHigh Peak AmplitudeIncrease flux by increasing Mr(Remanence Magnetization)Increase flux by increasing media thicknessDecrease magnetic spacingLonger read gap lengthNarrow Pulse WidthDecreasing magnetic spacingShorten read gap lengthDecrease media thicknessReduce self-demag by increasing coercivityIncrease write head field gradient in head construction(dont use too much current)readingwritingNeed trade-offsRecording ChannelRecording ChannelChannel write dataInputuser dataECC encoderChannel encoderEqualizerDetectorECC decoderChannel decoderoutputuser dataAnalog readback signal11010110110110110101101011011011011010Data Writing Processwrite current NRZIclock“Data”magnetic mediumTData Reading Process S N S N S N S N S N IVTThe Read/Write ChannelWriteCircuitPreampEncoderDecoderReadChannelData To RecordWrite ClockData Read BackRead Ref.ClockFromConrollerToConrollerHDAPCB20 mA200 Vpp50 mVppTTL,ECLTTL,ECL1011101110111011Pre-amps Write Circuit:H-Bridge DriverVccRdampHeadPredriverWrite DataWrite GatePre-amps Read Circuit:Differential Pre-ampVV+-Single-endedDifferentialCommon-mode noiseis rejected!NoiseNoiseThe Read Channel S N S N S N S N S N ObjectiveOutput a digital pulse corresponding to the peak of each transition on the mediaMEDIAReadSignalDerivedClockRead ChannelOutputTPeak DetectorThresholdDetectorDifferentiatorZero CrossingDetectorANDRead-backpulse101BitcellBitcellBitcellDetection Window=TNeed timingRecovery circuitTiming Recovery:Phase Locked Loop(PLL)PhaseDetectorIntegratorVCOFrom PeakdetectorClockPeak DetectorOutputVCO OutputVCOvery veryearlyVCOveryearlyVCOearlyVCOslightlyearlyVCOOn TimeVCOslightlylateVCOlatePulseMissingVCOOn TimeSlowDownSlowDownSlowDownSlowDownDontChangeSpeedUpSpeedUpDontChangeDontChangePhase DetectorOutputInter-Symbol Interference(ISI)Linear Superposition of pulsesReadbackWaveformWrite Current Pulses interfere with each other when written close together:Amplitudes are reduced and Timing is distortedUser Density(UD)=PW50/TTPW50Peak Detector Output ExamplesPeak DetectionAnalog Read-back WaveformThreshold Detector OutputDifferentiated DataZero Crossing Detector OutputDetected DataRead-back Waveforms at Different User DensitiesUD=0.75UD=1.5UD=2.0Asynchronous DetectionPeakDetectionPLLDetector has NO knowledgeof the bit timingPLL knows the bit timingNo communication to DetectorSynchronous DetectionPeakDetectionPLLDetector has knowledge of when a pulse may occur(bit timing)Can make a here/not-here decisionMakes better decisionsSignal to Noise Ratio(SNR)can be lowerSampled Detector allows for post compensationModel and remove ISI as an error sourceSynchronous Channel:Sampled Peak Detection0.0 -0.2 -0.9 0.0 0.9 0.1 -0.1 -0.8 0.7 -0.8 -0.1 0.1 0.9 0 -0.9 -0.2 0.00 0 -1 0 1 0 0 -1 1 -1 0 0 1 0 -1 0 00 0 1 0 1 0 0 1 1 1 0 0 1 0 1 0 050%-50%Detection ThresholdReceived SamplesTargetvaluesDetectedDataSynchronous Channel:Sampled Peak Detection50%-50%Detection Threshold-0.2 -0.4 -0.8 0.0 0.7 0.3 -0.2 -0.7 0.3 -0.7 -0.2 0.3 0.8 0 -0.8 -0.4 -0.2 0 0 -1 0 1 0 0 -1 0 -1 0 0 1 0 -1 0 00 0 1 0 1 0 0 1 0 1 0 0 1 0 1 0 0Received SamplesTargetvaluesDetectedDataMissing 1 transitionOr have one toomany transitionsSequence DetectionWe know certain sequences shouldnt exist.Make use of the fact!Step 1:Determine the rule for which sequences existFor sampled peak detection=polarity of pulses must alternateStep 2:Compare the observed samples with the expected samplesfrom all possible sequences.Choose the closest sequence.Closest=sequence with minimum squared errorClosest=most likely sequence Maximum LikelihoodStep 1:Rule for Possible SequencesPreconditionedThe TrellisEach path through the trellis corresponds to a possible data sequenceEach path through the trellis predicts a possible sequence of samples to observesTrellis ExamplePRMLPartial Response Maximum LikelihoodBinary data transmission method used in communications signal processing used to detect data in a noisy environment Originally used with deep space probesClass 4 applied to magnetic recording channels4 in PR4 refers to the class of partial response system used for magnetic recording channelsTwo relatively independent partsPartial Response-Method for equalizing the readback signal to achieve a sampled three level outputMaximum Likelihood-Sequence Detection Partial ResponseClass 4 Partial ResponseFilter or equalize until a transition gives the following waveformTarget more than one non-zero sample per pulseEach sample only contains part of the pulse(response)2 non-zero samples(call them+1)All other samples=0PR4 Equalized Isolated and Dibit PulsesIsolated PulseDibit PulseExample Class 4 Partial Response Waveform0 0 1 0 0 0 0 0 1 1 0 1 0 1 1 1 0 0 1 0 1 1 1 0 1 1NRZIPR4 Eye-patternAll waveforms at clock points pass through one of three points corresponding to sample values of-1,0,and 1.Sampling once each bit period results in three level outputTrellis Diagram for Class 4 Partial ResponsePR4 State DiagramRead-back Waveforms at Different User DensitiesUD=0.75UD=1.5UD=2.0Magnetic Channel SpectrumPW50/T=2PW50/T=3PW50/T=1PW50/T=1/2PW50/T=1/3At low recording densities the spectral energy is concentrated near one half of the channel clock rate frequencyAt higher recording densities most of the signal spectrum is below half of the channel clock rate frequencyLimit channel bandwidth to 1/2T without losing informationWhy go to higher order Partial ResponsesPR4PW50/T=0.5EPR4E2PR4PRML Read ChannelAGC-Automatic Gain Control maintains required constant signal level(VGA&Gain Control)Low Pass Filter-Coarse equalizationFIR-Finite Impulse Response filter for fine equalizationADC-Analog to Digital Converter samples equalizer outputViterbi Detector-Compares and selects maximum likelihood sequencePRML Write ChannelRandomizer-Reduces probability of occurrence of worst case repeated patterns that are detrimental to the timing and gain loopsEncoder-Encodes user data into channel dataPrecoder-Limits error propagation,Compensates for response of the equalized channelWrite Precomp-Write precompensation for nonlinear effects(not for ISI)Write PrecompensationCompensate for nonlinear effects-demagnetization field from previously written transition shifts transition earlyMHdHHcxRLL CodingRun Length Limited Codes(RLL)defined as(d,G/I)d=Minimum number of 0s separating 1sG=Maximum number of 0s separating 1s in global outputI=Maximum number of 0s separating 1s in each interleaved controls intersymbol interferenceCodes for PRML channels d=0ISI is constructively used and not a problem as in peak detectionG influences timing and gain correction update rates and sets lowest frequency of channel which must not be distorted by filterI influences the Viterbi path memory by minimizing the number of states between crossoversCode RateCode Rate-number of recorded bits written on disc for each user bit to be recordedCodeRate1,72/30,4/48/90,6/616/17Lower rate codes require higher frequency clocksHigher rate codes result in lower frequencies which requires a lower filter bandwidth that filters out more noiseHigher rate codes are more complexUser Density=Channel Density*Rate