Download to read offline
![Transistor Production
Figure: It is growing every year [IEEE Spectrum]
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-8-320.jpg)
![Frequency [ITRS 2013]
Figure: the rate has slowed down
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-11-320.jpg)
![Power Density Trend [ITRS 2015 — A Kahng, UCSD ]
Figure: Peak power density trend
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-14-320.jpg)
![Power Consumption [Borkar et al]
Figure: Processor power consumption
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-15-320.jpg)
![Heat Flux [IBM — ITRS 2015 meeting]
Figure: Heat flux: Watt/area with respect to devices - bipolar vs CMOS
There is a need of new device technology to tackle the heat flux!
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-16-320.jpg)
![Exascale Computing: Trends in Top500.org
Power and Energy will pose a major challenge [Borkar 2011].
GWatt consumption is expected.
Figure: Performance developement graph of top500 supercomputer
And, it is projected to go beyond Exaflop! Do we need it?
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-17-320.jpg)
![Performance per Watt Trend [top500.org]
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-19-320.jpg)
![Memory Wall: Hindrance in performance
Here is a more recent result examining the memory wall problem
for CPU,GPU, and TPU!3
Figure: The gap is growing exponentially still..
3
Amir et al, AI and Memory Wall, arXiv:2403.14123v1 [cs.LG] 21 Mar 2024
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-21-320.jpg)
![Emerging Memory Technology [ITRS 2015 meeting]
Solution: The idea is to bring in hierarchy system!
Figure: Focus is on non-volatile memory
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-22-320.jpg)
![On die Cache [Borkar et al]
Figure: It will increase to mitigate the memory wall problem
- The upcoming technology is system in package (SiP) moving
from the existing the SoC
- Addressing the problem of memory wall- memory bandwidth and
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-23-320.jpg)
![Emerging Architecture
von Neumann
The traditional microprocessor
Non von Neumann
Cellular Automata
Co-located memory-logic [procesor-in-memory,
memory-in-logic, nonvolatile logic]
Reconfigurable computing
Cognitive computing [neuro morphic, machine learning]
Statistical and stochastic computing [statistical inference,
approx computing]
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-24-320.jpg)
![Emerging Architecture: Data and learning
In Design or test production
GPU architecture (NVIDIA)
Tensor Processing Unit [Google ML processing for Search and
Language ]
Microsoft Azure with Stratix 10 (intel) FPGA
XPU by Xiling for Machine learning
IPU (Intelligent Proc Unit) by Graphcore
DPU (Data flow computnig) by Wavecore
DLU by Fujitsu
Jaynarayan Tudu
What Next in Computer Architecture: Problems and Research D](https://image.slidesharecdn.com/computer-architecture-research-250820070224-51120c76/85/What-next-in-Computer-Architecture-Problems-and-research-direction-25-320.jpg)
Uploaded byJaynarayan Tudu
What next in Computer Architecture: Problems and research direction
The slides contain a future direction on the computer architecture research.
What next in Computer Architecture: Problems and research direction
- 1. What Next inComputer Architecture: Problems and Research Direction Jaynarayan Tudu Computer Science and Engineering Indian Institue of Technology Tirupati, India August 20, 2025 Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 2. VLSI Past andLesson Learnt It all started with Moore’s Law The corollary of Moore’s Law Performance Power, Energy and Temperature (the Dennard Scaling) Reliability Complexity of the Design Evolution in computing Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 3. CMOS Transistor Figure: nmosand pmos transistors 1 1 Taken from: CMOS VLSI Des, Weste and Hariss Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 4. Transistor and ICHistory Figure: The first transistor and IC built at AT&T Bell Lab 2 2 Taken from: CMOS VLSI Des, Weste and Hariss Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 5. Transistor Scaling: Moore’sLaw Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 6. Manufacturing Cost: Moore’sLaw Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 7. Transistor count Figure: Itsis now in billion era Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 8. Transistor Production Figure: Itis growing every year [IEEE Spectrum] Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 9. Transistor Technology Trend Figure:Scaling driven technology Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 10. Trends in Functionalityin Microprocessor Figure: Microprocessor Functionality The highest till date: IBM P6 at 5.0 GHz. Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 11. Frequency [ITRS 2013] Figure:the rate has slowed down Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 12. Intel Microprocessor Trend:Clock frequency TheRevolutionContinues IntelcontinuestodeliveronthepromiseofMoore’s Lawwiththeintroductionofpowerfulmulti-core technologies,transformingthewaywelive,work, andplayonceagain. 108KHz 2,300 Numberoftransistors 10µ Manufacturingtechnology 10µ Manufacturingtechnology 6µ Manufacturingtechnology 3µ Manufacturingtechnology 3µ Manufacturingtechnology 1.5µ Manufacturingtechnology 1.5µ Manufacturingtechnology 1µ Manufacturingtechnology 0.8µ Manufacturingtechnology 0.6µ Manufacturingtechnology 0.25µ Manufacturingtechnology 0.18µ Manufacturingtechnology 90nm Manufacturingtechnology 65nm Manufacturingtechnology 65nm Manufacturingtechnology 90nm Manufacturingtechnology 65nm Manufacturingtechnology 45nm ManufacturingTechnology 0.13µ Manufacturingtechnology 0.18µ Manufacturingtechnology Intel®4004processor Introduced1971 Initialclockspeed ThegroundbreakingIntel®4004 processorwasintroducedwith thesamecomputingpower asENIAC. 500-800KHz 3,500 Numberoftransistors Intel®8008processor Introduced1972 Initialclockspeed 2MHz 4,500 Numberoftransistors Intel®8080processor Introduced1974 Initialclockspeed TheIntel®Pentium®4processor ushersintheadventofthe nanotechnologyage. 5MHz 29,000 Numberoftransistors Intel®8086processor Introduced1978 Initialclockspeed 5MHz 29,000 Numberoftransistors Intel®8088processor Introduced1979 Initialclockspeed 6MHz 134,000 Numberoftransistors Intel®286processor Introduced1982 Initialclockspeed 16MHz 275,000 Numberoftransistors Intel386™processor Introduced1985 Initialclockspeed 25MHz 1,200,000 Numberoftransistors Intel486™processor Introduced1989 Initialclockspeed 66MHz 3,100,000 Numberoftransistors Intel®Pentium®processor Introduced1993 Initialclockspeed 200MHz 5,500,000 Numberoftransistors Intel®Pentium®Proprocessor Introduced1995 Initialclockspeed 300MHz 7,500,000 Numberoftransistors Intel®Pentium®IIprocessor Intel®PentiumIIXeon®processor Introduced1997 Initialclockspeed 500MHz 9,500,000 Numberoftransistors Intel®Pentium®IIIprocessor Intel®Pentium®IIIXeon®processor Introduced1999 Initialclockspeed 1.5GHz 42,000,000 Numberoftransistors Intel®Pentium®4processor Introduced2000 Intel®Xeon®processor Introduced2001 Initialclockspeed 1.7GHz 55,000,000 Numberoftransistors Intel®Pentium®Mprocessor Introduced-2002 InitialClockSpeed 1GHz 220,000,000 Numberoftransistors Intel®Itanium®2processor Introduced2002 Initialclockspeed 3.2GHz 291,000,000 Numberoftransistors Intel®Pentium®Dprocessor Introduced2005 Initialclockspeed 2.93GHz 291,000,000 Numberoftransistors Intel®Core™2Duoprocessor Intel®Core™2Extremeprocessor Dual-CoreIntel®Xeon®processor Introduced2006 Initialclockspeed 2.66GHz 582,000,000 Numberoftransistors Quad-CoreIntel®Xeon®processor Quad-CoreIntel®Core™2Extremeprocessor Introduced2006 Intel®Core™2Quadprocessors Introduced2007 Initialclockspeed 1.66GHz 1,720,000,000 Numberoftransistors Dual-CoreIntel®Itanium®2processor9000series Introduced2006 Initialclockspeed > 3GHz 820,000,000 Numberoftransistors Quad-CoreIntel®Xeon®processor(Penryn) Dual-CoreIntel®Xeon®processor(Penryn) Quad-CoreIntel®Core™2Extremeprocessor(Penryn) Introduced2007 Initialclockspeed 1975 TheAltair8800microcomputer,based ontheIntel®8080microprocessor,was thefirstsuccessfulhomeorpersonal computer. 1972 TheIntel®4004processor,Intel’s firstmicroprocessor,poweredthe Busicomcalculatorandpavedthe wayforthepersonalcomputer. 2003 Intel®Centrino®mobiletechnologybrought highperformance,enhancedbatterylife, andintegratedWLANcapabilitytothinner, lighterPCs. 1994 Intelchipspoweredalmost75percentofall desktopcomputers. Moore’sLaw In1965,Intelco-founderGordonMoorepredictedthat thenumberoftransistorsonachipwoulddoubleabout everytwoyears.Sincethen,Moore’sLawhasfueleda technologyrevolutionasIntelhasexponentiallyincreased thenumberoftransistorsintegratedintoitprocessors forgreaterperformanceandenergyefficiency. TheRevolutionBegins Throughouthistory,newandimprovedtechnologies havetransformedthehumanexperience.Inthe20th century,thepaceofchangespedupradicallyaswe enteredthecomputingage.Fornearly40years Intelinnovationshavecontinuouslycreatednew possibilitiesinthelivesofpeoplearoundtheworld. 1989 TheNationalAcademyofEngineering namedthemicroprocessoroneoften outstandingengineeringachievements fortheadvancementofhumanwelfare. 2005 Dual-coretechnologywasintroduced. 1995 Releasedinthefallof1995,theIntel®Pentium®Pro processorwasdesignedtofuel32-bitserverand workstationapplications,enablingfastcomputer-aided design,mechanicalengineeringandscientific computation. 2001 TheItanium®processoristhefirst inafamilyof64-bitproducts fromIntelandisdesignedfor high-end,enterprise-class serversandworkstations. 2006 Intellaunchedfourprocessorsforservers undertheXeon5300brand,andanother processorundertheCore2Extreme seriesforhighperformancecomputing. These"quad-core"processorsshow improvedperformanceoverotherswith justoneortwoprocessingcores. 2007 Inthesecondhalfof2007,Intelbeganproduction ofthenextgenerationIntel®Core™2andXeon processorfamiliesbasedon45-nanometer(nm) Hi-kmetalgatesilicontechnology. 1998 TheIntel®PentiumIIXeon processorsfeaturetechnical innovationsspecificallydesignedfor workstationsandserversthatutilize demandingbusinessapplications. TheIntel®8008processorwas twiceaspowerfulastheIntel® 4004processor. TheIntel®8080processor madevideogamesandhome computerspossible. TheIntel®8086processorwas thefirst16bitprocessorand deliveredabouttentimesthe performanceofitspredecessors. ApivotalsaletoIBM'snewpersonal computerdivisionmadetheIntel® 8088processorthebrainsofIBM's newhitproduct--theIBMPC. TheIntel®286wasthefirst Intelprocessorthatcould runallthesoftwarewritten foritspredecessor. TheIntel386™processorcouldrunmultiple softwareprogramsatonceandfeatured 275,000transistors—morethan100times asmanyastheoriginalIntel®4004. TheIntel486™introducedtheintegrated floatingpointunit.Thisgenerationof computersreallyalloweduserstogofrom acommandlevelcomputerintopointand clickcomputing. TheIntel®Pentium®processor,executing 112millioncommandspersecond,allowed computerstomoreeasilyincorporate"real world"datasuchasspeech,sound, handwritingandphotographicimages. ThePentium®Proprocessordelivered moreperformancethanprevious generationprocessorsthroughan innovationcalledDynamicExecution. Thismadepossibletheadvanced3D visualizationandinteractivecapabilities. TheIntel®Pentium®IIprocessor’ssignificant performanceimprovementoverprevious Intel-Architectureprocessorswasbasedon theseamlesscombinationoftheP6 microarchitectureandIntelMMXmedia enhancementtechnology. TheIntel®Pentium®IIIprocessorexecuted InternetStreamingSIMDExtensions, extendedtheconceptofprocessor identificationandutilizedmultiple low-powerstatestoconservepower duringidletimes. TheIntel®Pentium®Mprocessor,theIntel® 855chipsetfamily,andtheIntel® PRO/Wireless2100networkconnectionare thethreecomponentsofIntel®Centrino® processortechnology.Intel®Centrino® processortechnologywasdesigned specificallyforportablecomputing. TheIntel®Pentium®Dprocessorfeatures thefirstdesktopduel-coredesignwithtwo completeprocessorcores,thateachrunat thesamespeed,inonephysicalpackage. Intel®Core™2Duoprocessoroptimizes mobilemicroarchitectureoftheIntel® Pentium®Mprocessorandenhancedit withmanymicroarchitectureinnovations. Intel®Centrino®ProandIntel®vPro™ processortechnologyprovideexcellent performancefromtheDual-CoreIntel® Core™2Duoprocessor. Dual-CoreIntel®Itanium®2processor9000series outperformstheearlier,single-coreversionofthe Itanium2processors.Withmorethan1.7billion transistorsandwithtwoexecutioncores,these processorsdoubletheperformanceofprevious Itaniumprocessorswhilereducingaveragepower consumption. TheunprecedentedperformanceoftheIntel® Core™2Quadprocessorismadepossiblebyeach ofthefourcompleteexecutioncoresdelivering thefullpowerofIntelCoremicroarchitecture. TheQuad-CoreIntel®Xeon®processorprovides 50percentgreaterperformancethanindustry- leadingDual-CoreIntel®Xeon®processorinthe samepowerenvelope.Thequad-core-based serversenablemoreapplicationstorunwitha smallerfootprint. Intel’snextgenerationIntel®Core™2processor family,codenamed"Penryn",contains industry-leadingmicroarchitecture enhancements. Further,newSSE4instructions forimprovedvideo,imaging,and3Dcontent performanceandnewpowermanagement featureswillextend“Penryn”processorfamily leadershipinperformanceandenergyefficiency. TheIntel®Itanium®2processoristhe successorofthefirstItaniumprocessor. ThearchitectureisbasedonExplicitlyParallel InstructionComputing(EPIC).Itistheoretically capableofperformingroughly8timesmore workperclockcyclethanotherCISCandRISC architectures. 1976 Anoperatorinanearlybunnysuitshowshowa4-inch waferispreparedforapositiveacidspin. 1981 TheIntel®8088microprocessorwas selectedtopowertheIBMPC. 1982 Within6yearsofitsrelease,anestimated 15million286-basedpersonalcomputers wereinstalledaroundtheworld. Note:Numberoftransistorsisanapproximatenumber. 45nm Manufacturingtechnology Figure: Clock frequency has been consistently in rise! Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 13. Performance Growth: 1978to 2018 Figure: Growth of performance over last 40 years (where is the Moore’s law) 1978 to 1986: Doubling every 3.5 years 1988 to 2002: Doubling every 2 years 2002 to 2010: Doubling every 4 years 2010 to 2014: Doubling every 8 years 2014 to 2018: Doubling every 20 years Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 14. Power Density Trend[ITRS 2015 — A Kahng, UCSD ] Figure: Peak power density trend Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 15. Power Consumption [Borkaret al] Figure: Processor power consumption Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 16. Heat Flux [IBM— ITRS 2015 meeting] Figure: Heat flux: Watt/area with respect to devices - bipolar vs CMOS There is a need of new device technology to tackle the heat flux! Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 17. Exascale Computing: Trendsin Top500.org Power and Energy will pose a major challenge [Borkar 2011]. GWatt consumption is expected. Figure: Performance developement graph of top500 supercomputer And, it is projected to go beyond Exaflop! Do we need it? Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 18. Power Trend inTop500 Figure: Nothing comes free... more performance consumes more power! The result is recent one - 2024 November. Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 19. Performance per WattTrend [top500.org] Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 20. Memory Wall: Hindrancein performance Tjhe old story of memory wall on CPU and Memory! Figure: Obsever the flat cpu performance post 2005.... The real issue is the gap! Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 21. Memory Wall: Hindrancein performance Here is a more recent result examining the memory wall problem for CPU,GPU, and TPU!3 Figure: The gap is growing exponentially still.. 3 Amir et al, AI and Memory Wall, arXiv:2403.14123v1 [cs.LG] 21 Mar 2024 Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 22. Emerging Memory Technology[ITRS 2015 meeting] Solution: The idea is to bring in hierarchy system! Figure: Focus is on non-volatile memory Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 23. On die Cache[Borkar et al] Figure: It will increase to mitigate the memory wall problem - The upcoming technology is system in package (SiP) moving from the existing the SoC - Addressing the problem of memory wall- memory bandwidth and Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 24. Emerging Architecture von Neumann Thetraditional microprocessor Non von Neumann Cellular Automata Co-located memory-logic [procesor-in-memory, memory-in-logic, nonvolatile logic] Reconfigurable computing Cognitive computing [neuro morphic, machine learning] Statistical and stochastic computing [statistical inference, approx computing] Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 25. Emerging Architecture: Dataand learning In Design or test production GPU architecture (NVIDIA) Tensor Processing Unit [Google ML processing for Search and Language ] Microsoft Azure with Stratix 10 (intel) FPGA XPU by Xiling for Machine learning IPU (Intelligent Proc Unit) by Graphcore DPU (Data flow computnig) by Wavecore DLU by Fujitsu Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 26. Trends in IoTand AI Figure: IoT in use, market capital indicates user trends and applicability - Find out the trend for AI applications and computing! Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 27. Summary and ActionItems Going beyond Moore’s Law Sustainable computing - Power and energy remains a challenge Increasing clock frequency in CPU a well GPU Increasing number of Cores in CPU and GPU Exploring accelerator level parallelism NVIDIA: What Is Energy-Efficient Computing? Sustainable computing is an end-to-end approach that spans the entire lifecycle of technology, designed to enhance energy efficiency and boost carbon mitigation. From weather forecasting to chip manufacturing, accelerated computing holds the potential to realize green computing objectives, while saving costs. Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 28. Summary and ActionItems Figure: World economic forum: Global impact of sustainable computing - energy is a problem Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 29. Reference and ReadingMaterials J. L. Hennessy and D. A. Patterson, Computer Architectures: A Quantitative Approach, Morgan Kaufmann Publishers, 5th Edition. J.P. Shen and M.H. Lipasti, Modern Processor Design, MC Graw Hill, Crowfordsville, 2005 Current Literature (from ISCA, Micro, HPCA, ICCD, and IEEE Trans. on Computers, IEEE Architecture Letters) The other resources: http://pages.cs.wisc.edu/ arch/www/ Prof. Onur Mutlu: https://users.ece.cmu.edu/ omutlu/ Almost all universities have very strong research group SPEC Benchmark: https://www.spec.org/benchmarks.html Simulators: simplescalar, sniper, gem5, gpgpu-sim, tejas DBLP: https://dblp.uni-trier.de/ Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 30. Research in India IIScBangalore: Prof Matthew Jacob, Prof. R Govind Rajan, Dr. Uday Bandagolu, Dr. Arka Basu, Prof. S K Nandi. IIT Bombay: Prof. Virendra Singh, Prof. Sachin Patkar, and Prof. Madhava Desai, Dr. Biswabandan Panda. IIT Madras: Prof. V Kamakoti, Dr. Rupesh Nasre, Prof. Madhu Mutyam IIT Delhi: Prof. Pritiranjan Panda, Prof. Smruti Sarangi IIT Kanpur: Prof. Mainak Chaoudhuri, IIT/IIIT Hyderabad: Dr. Mittal, Dr. Lavanya Jaynarayan Tudu What Next in Computer Architecture: Problems and Research D
- 31. Thank You Jaynarayan Tudu WhatNext in Computer Architecture: Problems and Research D