Advanced microprocessor systems
FirstTransistorMicroprocessorEvolution
DiscreteTransistors
Bipolar
FET
PlanarTransistors
BJTFET
1971 1972
10,000nm 10,000nm
808629000transistors3000nm
1978 1985
80386275000transistors1500nm
1993 1995
Pentium3.1Mtransistors800nm
PentiumPro5.5Mtransitors500nm
2006
PentiumDPresler362Mtransistors65nm
2007
Core2DuoWolfdale411Mtransistors45nm
2011 2016
8-corei7Haswell2.6Btransistors22nm
22coreXeonBroadwell7.2Btransistors14nm
HilevelCrosssection
Microprocessorcircuitsonchip
Clock
Graphics
Network Comm.
Mass Storage
A/VI/O Control
ROM/FLASH
CPU
Peripherals
Power
SCSIIDE/ATA/SATAFDDI
USBIEEE488Bluetooth
Hypertransport/PCI express
10/100bT/Gb/802.11
PCI express RAML1/L2/L3
Advanced Microprocessor System Block Diagram
DDR1,2,3,4-SDRAM
1.8 – 3.8 GHz
Dual SLI
Wireless, Ethernet, Optical
SATA, RAID, DVD, Blueray
Keyboard, Mouse, scanner, USB
Up to 1KW, water cooling
3D sound,Joystick,
CPUAdvancementMechanisms:
1)Advancesinprocessorarchitecture:
RISCPipeliningSuperscalarOut-of-orderexecution
RISC- asystemthatusesasmall,highly-optimizedsetofinstructionsthattypicalyexecuteinoneclockcycle,ratherthanamorespecializedsetofinstructionsthatmayrequireseveralclockcycles.TherearemanytraitsassociatedwithRISC.Forexample,mostmachinesareimplementedwithmultipleinternalbusessimilartoaHarvardarchitecture.AnothercommontraitisthatRISCsystemsusetheload/storearchitecture,wherememoryisnormallyaccessedonlythroughspecificinstructions,ratherthanaccessedaspartofotherinstructionslikeanadd.
Pipelining- atechniqueusedinthedesignofcomputerstoincreasetheirinstructionthroughput.Ratherthanprocessingeachinstructionsequentially,eachinstructionissplitupintoasequenceofstepswhichareexecutedwithinasmalloffsetfromoneanother.Thus,differentstepscanbeexecutedconcurrently(bydifferentcircuitry),andalmostinparallel.
Superscalar- aformofparallelismcalledinstruction-levelparallelismwithinasingleprocessor.ACPUexecutesmorethanoneinstructionduringaclockcyclebysimultaneouslydispatchingmultipleinstructionstoredundantfunctionalunitsontheprocessor.
Out-of-orderexecution– ACPUtechniqueinvolving:
• fetchinginstructionsinacompiler-generatedorder
• Theinstructionsaredynamicallyscheduled
•Inbetweentheymaybeexecutedinsomeotherorder
•Independentinstructionsbehindastalledinstructioncanpassit
•Instructionsarereshuffledbackintothecorrectorderforwriteback stage
CPUAdvancementMechanisms:
2)Advancesinprocessormanufacturing:Lithographyprocesses:
Opticale-beamX-rayExtremeUV
OpticalLithography-aphotograhic processusedtopatternpartsofasemiconductorfilmorthematerialof asubstrate.Ituseslighttotransferageometricpatternfromaphotomask toalight-sensitivechemical"photoresist”,onthesubstrate.Aseriesofchemicaltreatmentstheneitherengravestheexposurepattern,orenablesdepositionofanewmaterialinthedesiredpattern.
e-beamLithography- isthepracticeofscanningafocusedbeamofelectronstodrawcustomshapesonasurfacecoveredwithanelectronsensitivefilmcalledaresist.Thepurpose,aswithphotolithography,istocreateverysmallstructuresintheresistthatcansubsequentlybetransferredtothesubstratematerial,oftenbyetching.Usingthistechnique,sub10nmstructurescanbecreated.
X-rayLithography- usesX-raystotransferageometricpatternfromamasktoalight-sensitivechemicalphotoresist onasubstrate.Aseriesofchemicaltreatmentsthenengravestheproducedpatternintothematerialunderneaththephotoresist.Usingthistechnique,sub1nmstructurescanbecreated.
ExtremeUVLithography- anext-generationlithographytechnologyusinganextremeultraviolet(EUV)wavelength,currentlyexpectedtobearound13.5nm.Usingthistechnique,sub20nmstructurescanbefabricated.
CPUAdvancementMechanisms:
3)Advancesinprocessorimplementations:
SemiconductorcomputingOpticalcomputingMolecularcomputingQuantumcomputing
Timelineofcomputationalimplementations27
00BC
1900
AD
1945
AD
1960
AD
Mechanical Electronic Semiconductor
2014
AD
1945
AD
Optical
1970
AD
1994
AD
Molecular
Quantum 2014
AD
2014
AD
2014
AD
Semiconductorcomputing- Siliconisonlyoneoutofmanydifferentsemiconductormaterials.A combinationoftheelementsgalliumandarsenicformscrystalswhichpermitelectronstomovefasterthaninsilicon,sothatthismaterialissometimesusedwhenextremespeedisimportant.Themainreasonthatsiliconisusedincomputersisbecauseitiseasier,andthereforelessexpensive,tomakecomplicatedcircuitsoutofsiliconthanforanyothermaterial.Computercircuitsalsorequiresomepartstobemadeoutofinsulatorsinadditiontothepartsthataresemiconducting.Withsilicon,itiseasytomakeagoodinsulatorbyaddingsomeoxygentoproducesiliconoxide.Theaveragecostisnowmuchlessthanonepennyper10,000transistors.
Opticalcomputing- usesphotonsproducedbylasersordiodesforcomputation.Photonspromisetoallowahigherbandwidththantheelectronsusedinconventionalcomputers.Photoniclogicistheuseofphotons(light)inlogicgates(NOT,AND,OR,NAND,NOR,XOR,XNOR).Switchingisobtainedusingnonlinearopticaleffectswhentwoormoresignalsarecombined.
Molecularcomputing- aformofcomputingwhichusesDNA*,biochemistryandmolecularbiology,insteadofthetraditionalsilicon-basedcomputertechnologies.Itisinitsinfancyandiscurrentlythesubjectofresearch.
*AbioengineerandgeneticistatHarvard’sWyssInstitutehavesuccessfullystored5.5petabits ofdata— around700terabytes— inasinglegramofDNA,smashingthepreviousDNAdatadensityrecordbyathousandtimes.
Quantumcomputing- acomputationdevicethatmakesdirectuseofquantum-mechanicalphenomena,suchassuperpositionandentanglement,toperformoperationsondata.Quantumcomputingisstillinitsinfancybutexperimentshavebeencarriedoutinwhichquantumcomputationaloperationswereexecutedonaverysmallnumberofqubits. Bothpracticalandtheoreticalresearchcontinues todevelopquantumcomputersforbothcivilianandnationalsecuritypurposes,suchascryptanalysis.
28
Advanced Microprocessor System Block Diagram
Colorkey:Red=processorBlack=I/OLt.gray=singlechipDarkgray=containedonchip
OMAP4430 die
Bump Underfill
Package size
Footprint (top side)
Footprint(bottom side)
intro-003
Top memory MCP package
Memory interface
Bottom PBGA547 OMAP4430POP package 0.4-mm pitch
12 x 12 mm
intro-004
Public Version
Package-On-Package Concept www.ti.com
1.4 Package-On-Package ConceptThe OMAP4430 die uses flip-chip technology. The OMAP4430 package-on-page (POP) device supportsmemory stacking using a POP implementation.The OMAP4430 die provides two LPDDR2 interfaces. Each interface supports up to two chip-selects, soup to four LPDDR2 memory dies are supported. Those interfaces are available only on device top ball out.The two stacked memory packages are directly connected to the two LPDDR2 EMIF4D interfaces of theOMAP4430 die.Figure 1-3 shows the concept of the POP solution, and Figure 1-4 shows stacked memory package on thePOP device.
Figure 1-3. POP Concept
Figure 1-4. Stacked Memory Package on the POP Device
Two types of LPDDR - SDRAM memories are supported in POP package: S4 and S2 with size up to 2GBand 32-bit data width.The POP device includes feedthroughs. The feedthroughs are defined from the bottom ball-grid array(BGA) to the stacked memory. The purpose of some of the feedthroughs is to provide power supply to thestacked memories.
270 Introduction SWPU231R–July 2010–Revised March 2011
© 2010–2011, Texas Instruments Incorporated
Stackedmemorypackage-on-package(POP)device
QuadCore14nm1.6GHzBaseFrequency2.4GHzBurstFrequency