ECE171 DigitalCircuits

Prof.MarkG.Faust MaseehCollegeofEngineering andComputerScience

Lecture3
Topics
IEEE754FloatingPoint Binary i Codes d
BCD ASCII,Unicode GrayCode 7SegmentCode Moutofncodes

Seriallinecodes

FloatingPoint
Needtorepresentrealnumbers Fixedpointtoorestrictiveforprecisionand range Notunlikefamiliarscientific scientificnotation notation +6.022x1023
Sign Mantissa Exponent
3

Normalizedmantissa singledigittoleftofdecimalpoint

IEEE754FloatingPointStandard
Binary +1.101011x212
Sign Exponent

Sign Exponent Significand

Mantissa

Anormalized(binary)mantissawillalwayshavealeading1 sowecanassumeitandgetanextrabitofprecisioninstead

Fraction

Exponentsarestoredwithabiaswhichisaddedtotheexponentbeforebeing stored(allowsfastmagnitudecompare) Universallyusedonvirtuallyallcomputers Severallevelsofprecision/range:Single,Double,DoubleExtended

IEEE754FloatingPointStandard
SinglePrecision 32bits bits,Bias=127
1 8 23 Significand 0

Exponent+Bias 31 30 23 22

F=1Sign x(1+Significand)x2(ExponentBias)
5

IEEE754FloatingPointStandard
Example
1 8 F=1Sign x(1+Significand)x2(ExponentBias) 23

1 10000001
31 30

01000000000000000000000
=0.2510
= 1.012 x2(129127) = 1.012 x22 = 1012 = 5.0 0

129

23 22 = 1.012 x2(129127) = 1.25x2(129127) = 1.25 1 25x22 = 1.25x4 = 5.0

IEEE754FloatingPointStandard
Willcommonlyseetheseexpressedashex
1 8 23

1 10000001
31 30 23 22

01000000000000000000000
0

C0A00000
7

IEEE754FloatingPointStandard
SomeSpecialCases
Zero( (noassumedleading g1) )
Exponent=0,Significand=0

NaN( (NotaNumber), ),e.g. g

Denormalizednumbers
Exponent=0,Significand 0

IEEE754FloatingPointStandard
Whydoweneeddenormalizednumbers?
Addresses Add gapcaused dby b implicit i li itleading l di 1 Smallestpositivenumber(a)is1.000000x2126 Nextnumber(b)is1.000001 1 000 001x2126= (2126 +2149)

Gaps!
SExponent Significand 1823
00000000000000000000000000000000 00000000000000000000000000000001 00000000000000000000000000000010 00000000000000000000000000000011 00000000000000000000000000000100

0 a

=0 1 00000000000000000000001x2127 =1.00000000000000000000001 =1.00000000000000000000010x2127 =1.00000000000000000000011x2127 =1.00000000000000000000100x2127

Distancefromzerotosmallestpositivenumberis1.00000000000000000000001x2127~=2127 9 Distancetonextnumberis0.00000000000000000000001x2127=223 x2127=2150

IEEE754FloatingPointStandard
DenormalizedNumbers
Solution Nonnormalizedform Exponent=0,Significand 0 ImplicitExponentof126 F=1Sign x(Significand)x2(126) Smallest S ll positive i i number b ( (a) )=0.000001 0 000 001x2126=2149 Nextsmallestnumber(b)=0.000010x2126=2148

10

BCD(BinaryCodedDecimal)
Usedinearly4bitP Simpledisplays 4Bits/DecimalDigit

11

ASCII
AmericanStandardCodeforInformation g (pronounced: (p Askey y) Interchange Forencodingtext Universallyused
EBCDIC(oldIBMmainframestandard)died Unicode U i d f fori international t ti l( (nonRoman) R )languages l

12

ECE171 1000101 1000011 1000101 0110001 0110111 0110001 0000000 AsaCstring

13

ReflectiveGrayCode(RGC)
Adjacentcodesdifferbysinglebit
UnitDistanceCode

Oftenusedininterfacing gmechanicalsensors

14

ReflectiveGrayCode

15

ReflectiveGrayCode (Conversion)

16

ReflectiveGrayCode (Conversion)

17

7SegmentCode

18

1outofncodes
8devices 23 =8 therefore3 bits(wires) suffice ifdevoted8 wires usedinSCSI diskdrives, PCI enumeration 000 001 111

19

SerialDataTransmission&Storage
Parallel
Storage:eachbitofwordread/writtensimultaneously Transmission:eachbithasseparatesignalpath

Serial
Reducecosts Simplifydesign Higherspeed(LVDS,skew)

Applications
USB PCIExpress E
20

SerialDataTransmission&Storage
Clock
Determinesrateatwhich h hbits b aretransmitted d(bit (b rate) ) bittime=clockperiod(1/bitrate)=bitcell

Sync
Determinesstartofbyte(orpacket)

DataFormat
Determinedbylinecode

21

CodesforSerialDataTransmission andStorage
NRZ NonReturntoZero NRZI NonReturntoZeroInvert( (onones) ) Transitionbased Differentialsignaling:USB

RZ ReturntoZero BPRZ BipolarReturntoZero DCbalanced MLT3(100BaseTEthernet) Guaranteesatransitionineverybitcell Facilitatesclockrecovery Requireshigherbandwidth Originalcoaxbased10MbpsEthernet moutofncodes(e.g.8B10B):Gigbabit Ethernet 22

Manchesterencoding

OthertechniquesforDCbalancing(andedgedensity)

WhySerial?
Parallel Serial
+

DeviceA

DeviceB

DeviceA

DeviceB

10bidirectionalwiresat250Mbps

pairunidirectionalwiresat2500Mbps(2.5Gbps)

23

Traditional ParallelBus
Device A Device B

Usedinlowtomedium100MHzrange Issues Boardtracelengtheffectonskew Clockskewacrossdevices Faster F t data d t rate t squeezes eye

SourceSynchronousBus
DeviceA DeviceB

Usedin200MHzto1.6GHzrange Cl ksignal Clock i lis i f forwarded d dwith i h d data Designimpact: Boardlayouttracklengthmismatch stilladdstoskew Eliminatesskewerrortermcausedbyclock domainskew Allowsfastercycletimesthanparallel
25

EmbeddedClock
Data Clock

Clocksignalembeddedwith ithdata

Clocksignalisembeddedwithdata Receiveduseddigitalphaselockedloop(DPLL)to recoverclockanddeterminewherebittimesare Edgedensitymustguaranteedbyencodingscheme Examples PCIExpress, Express USB, USB SerialRapidIO, RapidIO Infiniband IntelsnewQuickPath Interconnect 26

EdgeLockTechnique(Tracking Receiver)
Device A
DeviceAsendspulsetraintoDeviceB

Device B

DeviceBlocksontoedges tobeinsyncwithpulsestream

27

EnsuringEdgeDensity:mofncodes
Some8bitcodewordshavetoofew1s(or0s)toensureedgedensity sufficienttorecoverclock 8b10encoding(developedbyIBMin1983)
Use10bitcodewords,butonlyuseasubsetoftheavailable210 codewordsthat haveabalancednumberof0sand1s Benefits
Ensureedgedensity AvoidDCbiasatreceiverfromimbalance
Running gdisparity p yforunbalancedcodewords

256datacharacters
All8bitbytes 12controlcharacters(INIT,etc)
DeviceA Parallel Data Parallel Data TXFIFO 8B/10B Serializer Encoder 8B/10B Deserializer Decoder
+ _

8 bit byte

10 bit code

Table lookup DeviceB Deserializer 8B/10B Decoder 8B/10B Encoder RXFIFO Parallel Data Parallel Data

RXFIFO

+ _

Serializer

TXFIFO

28

ErrorDetectionandCorrection
Errorsoccurduring datastorage/retrievalandtransmission
Noise, Noise crosstalk, talk EMI, EMI cosmicrays, rays impuritiesinICmaterials Morecommonwithhighspeeds,lowervoltages

Usemoutofn codestodetecterrors
Not N t all llpossible ibl codes d areused d(valid) ( lid) Errors inused(valid)codes(hopefully)produceunused(invalid)codes

Example:Luhn Algorithm
Creditcards,IMEI(InternationalMobileEquipmentIdentity) Startfromright,doubleeveryseconddigit Addalldigits Addafinalcheckdigittoensuresumismultipleof10

29

DifferentialSignaling
Differential,pointtopoint
Complementarysignalstransmitted Receiverdetectsvoltagedifference betweenlines Lowamplitudes(200mV 400mVtypical),highspeeds Goodnoiseimmunity
Pairroutedtogether noisecancelsout

LVDS LowVoltageDifferentialSignaling
ANSI/TIA/EIA6441995standard(signalingonly,notprotocolorconnectors) 3.125Gbps,+/ 350mV Gbps atmWs Highspeed&lowpowerconsumption FibreChannel GigabitEthernet FibreChannel, Ethernet,HDMI HDMI,DVI, DVI USB

WellsuitedtoNRZIencoding onlycareabouttransitions signals cross

30