EXTREME SCALE DESIGN AUTOMATIONarchive2.Cra.org/ccc/files/docs/esda/panel-iccad-14.pdfvalidation,...
Transcript of EXTREME SCALE DESIGN AUTOMATIONarchive2.Cra.org/ccc/files/docs/esda/panel-iccad-14.pdfvalidation,...
EXTREME SCALE DESIGN AUTOMATION CCC/SIGDA Workshop Series
Alex K. Jones
University of Pittsburgh
Special Session Presentation
SiGdaspecial interest group on
design automation
Distribution A – Approved for Public Release; Distribution Unlimited
Classic Moore’s Law: Made New Designs Possible, Old Ones Lucrative
Silicon Process Technology
Intel386™ DX
Processor
Intel486™ DX
Processor
Pentium® Processor
Pentium® II, III
Processors
Pentium® 4
1.5µ 1.0µ 0.8µ 0.6µ 0.35µ 0.25µ 0.18µ
Time
30 engineers, ½ yr
500+ engineers, 5 years 5
Source: Bob Colwell Intel, DARPA
SURVEY: WILL MOORE’S LAW END?
Source: EE Times Survey March 2014
• Traditional (Dennard) scaling ended 10 years ago (sub 80nm)
• Industry roadmap will continue to find patches to silicon (5nm) – Seems the cost scaling
proposition may have ended
EDA is sDll chasing a mul$-‐dimensional moving target
SiGdaspecial interest group on
design automation
EXTREME SCALE SILICON ROADMAP
EUV Saves Costs versus Double Patterning
450mm Increases Throughput to Save Cost
2023?
DELAYED FD-‐SOI?
All around gate transistors?
Will scale to 10nm? 7nm? 5nm?
What is clear? EDA will become increasingly important
at the Extreme Scale
SiGdaspecial interest group on
design automation
CCC/SIGDA WORKSHOPS ON EXTREME SCALE DESIGN AUTOMATION
Series of three workshops:
• Workshop 1: Emerging Technologies and Workforce Continuity – March 7-8, 2013 Pittsburgh
• Workshop 2: Extreme Scale Chips and Industry Research – June 2-3, 2013 Austin (Collocated with DAC)
• Workshop 3: Achieving Sustainable Collaborations Through Abstractions, Methodologies, and Benchmarks – February 20-21, 2014 Tampa
The purpose of this workshop series was to take an introspec8ve look at the EDA field while crystalizing
a vision for both the near and long term.
SiGdaspecial interest group on
design automation
WHERE SHOULD EDA INVEST?
Major Focus Areas
• Extreme-scale Electronic Design Automation (ESDA) – First Big Data Discipline, 1015 devices – Focus on System-level Design and Verification
• Emerging Technologies (Post-CMOS/Hybrid) – Develop Full System Flows – Technology choices: more than ad hoc demonstrations
• New Markets (DAoT) – Near term: CPS/IoT, Cyber-secure systems (hardware) – Medium term: Biology/Medical Technology
• Cross Cutting – Abstractions,
Metrics, Benchmarks
– Synergy with Computer Architecture
– Education and Workforce
SiGdaspecial interest group on
design automation
EXTREME SCALE DESIGN AUTOMATION Continuing Onward: Next-Generation Electronic Systems
SiGdaspecial interest group on
design automation
ESDA IS AN EXCITING TIME FOR EDA
EDA’s Objectives “at a glance” ITRS Productivity Curve
Figure 1: The design productivity gap [6]
198
119
85
198
9
1993
199
720
01
200
5
200
9
2013
2017
2021
2025
2029 Time
Technology Capabilities2x/36 months
HW Design ProductivityFilling with IP and Memory
HW Design Productivity
HW Design Gap
LogGates/Chip
Gates/Day
Figure 3: Key aspects of modern research and development in EDA.
Level of Abstraction
Quality (of optimization) of result = QOR; PPAY= power, performance, area and yield
Paradigm domain-specific, stochastic and approximate computing, etcApplication
Time-scale of the research lifecycle in years How are we doing?
EDA Improvement PotenDal
This uncertain, and exci8ng environment is reminiscent of the beginnings of the EDA era prior to the stability provided by Dennard
scaling.
SiGdaspecial interest group on
design automation
Uncertainty of the current environment is alleviated by the large global demand for IC
products
Figure 1: The design productivity gap [6]
1981
1985
1989
1993
1997
2001
2005
2009
2013
2017
2021
2025
2029 Time
Technology Capabilities2x/36 months
HW Design ProductivityFilling with IP and Memory
HW Design Productivity
HW Design Gap
LogGates/Chip
Gates/Day
WHAT ABOUT “LEGACY” TECHNOLOGY NODES?
130nm design starts dominate the market
• Barriers – “Good enough solutions” – Fear new methods break tools
• Unintended side effects • Push button flows still achievable
– Inadequate investment
• 130nm designs are popular – Reduces upfront cost – Increases yield
Closing the produc$vity gap at 130nm could approximate
current technology capabili$es
An EDA advancement demonstrated in a tool that finds superior solu8ons is par8cularly valuable to effec8vely u8lize legacy technology nodes. SiGdaspecial interest group on
design automation
HYBRID POST-CMOS ELECTRONICS A Changing Landscape: New Technologies for Integrated Circuits
SiGdaspecial interest group on
design automation
ROLE OF EMERGING TECHNOLOGIES
Replace Si-CMOS? • There are many emerging
technologies – Most are niche – Likely hybrid solutions
• Challenges – Integration – Stochastic behavior – Models and
abstractions – Full system
demonstrations
11
processed in order to evaluate and understand low-level (e.g., atomistic) behavior and to explore all reachable system states is huge. There may need to be better statistical methods for simulations involving big data.
Flexible Models: There will be a need for more flexible, modular, and/or extendable tools that can easily incorporate different behavior from various emerging technologies. A building-block approach, where each block is easily modifiable may be particularly advantageous.
High-level Abstractions: The benefits of high-level prototyping, estimation, synthesis, and verification can only be unlocked with appropriate high-level design and EDA abstractions. Some emerging technologies are particularly dependent on new abstractions and models–from logic to system level.
Physical Layout: New devices and circuit styles typically require rethinking traditional notions of physical layout, synthesis, extraction, and verification.
3.2 New Design Processes for New Technologies
Focusing on high-level analysis may require rethinking how systems are specified, designed, and implemented. If we take a current system that has been optimally designed using traditional silicon technology, and then simply replace each silicon component with an emerging-technology-equivalent, it may not yield any significant benefits (and may even be worse). This may not be surprising, since design decisions at the highest level are often the result of constraints dictated at the technology level. Instead, if we have the capacity to accomplish design exploration at the highest level first, unconstrained by technology, we may end up with a completely new way of computing that could allow us to better exploit the best properties of new technologies. Only then will it become apparent what kind of design automation tools would be most beneficial for these new technologies. We need new models that give designers the flexibility to change the high-level structure.
TECHNOLOGY PROPOSED USEOptical interconnect, optical devices High-performance, high-bandwidth communications
Terahertz (RF) circuitsAutomotive radar, security, high-bandwidth wireless communcation
Microelectromechanical systems (MEMS) and Nanoelectromechanical (NEMS)
Mechanical filter/switches, wideband antennas, gyroscopres, energy harvesting, data storage, sensors
Spintronics/multiferroicsModeling synapse, physical brain/biomimetric behavior
Flexible electronicsWearable computing, body tracking, glucose monitoring
Qbit technologies Quantum computing/annealing/optimization
Phase-change memories (including memristors) DNA memory (having long retention)
Microfludicis Lab-on-a-chip, cooling
Steep slope devices Ultra low-power computing
Superconductors Ultra low-power computing, ultra high performance
Carbon-based electronicsUltra low-power computing, high performance, monolithic 3D ICs
Figure 4: Emerging technologies for the EDA community.
EDA should not determine which new technologies to pursue.
EDA should address new technology specific challenges for EDA flows.
SiGdaspecial interest group on
design automation
THE HYPE CYCLE OF INNOVATION
Expectations
time
InnovationTrigger
Expectations
First-generationproducts, high price, lots of customization needed
time
InnovationTrigger
Expectations
Mass mediaHype begins
First-generationproducts, high price, lots of customization needed
time
InnovationTrigger
Expectations
Mass mediaHype begins
First-generationproducts, high price, lots of customization needed
time
InnovationTrigger
Peak of InflatedExpectations
Expectations
Negative press beginsMass media
Hype begins
First-generationproducts, high price, lots of customization needed
time
InnovationTrigger
Peak of InflatedExpectations
Expectations
Negative press beginsMass media
Hype begins
First-generationproducts, high price, lots of customization needed
time
InnovationTrigger
Peak of InflatedExpectations
Trough ofDisillusionment
Expectations
Negative press beginsMass media
Hype begins
First-generationproducts, high price, lots of customization needed
Second-generationproducts, some services
time
InnovationTrigger
Peak of InflatedExpectations
Trough ofDisillusionment
Expectations
Negative press beginsMass media
Hype begins
First-generationproducts, high price, lots of customization needed
Second-generationproducts, some services
time
InnovationTrigger
Peak of InflatedExpectations
Trough ofDisillusionment Slope of Enlightenment
Expectations
Negative press beginsMass media
Hype begins
First-generationproducts, high price, lots of customization needed
Methodologies and best practices developing
Second-generationproducts, some services
time
InnovationTrigger
Peak of InflatedExpectations
Trough ofDisillusionment Slope of Enlightenment
Expectations
Negative press beginsMass media
Hype begins
First-generationproducts, high price, lots of customization needed
Methodologies and best practices developing
Third-generationproducts, out of thebox solutions, product suitesSecond-generation
products, some services
time
InnovationTrigger
Peak of InflatedExpectations
Trough ofDisillusionment Slope of Enlightenment
Expectations
Negative press beginsMass media
Hype begins
First-generationproducts, high price, lots of customization needed
Methodologies and best practices developing
Third-generationproducts, out of thebox solutions, product suitesSecond-generation
products, some services
time
InnovationTrigger
Peak of InflatedExpectations
Trough ofDisillusionment Slope of Enlightenment
Plateau ofProductivity
Expectations
Negative press beginsMass media
Hype begins
First-generationproducts, high price, lots of customization needed
Methodologies and best practices developing
High-growth adoptionphase starts: 20 to 30percent adoption
Third-generationproducts, out of thebox solutions, product suitesSecond-generation
products, some services
time
InnovationTrigger
Peak of InflatedExpectations
Trough ofDisillusionment Slope of Enlightenment
Plateau ofProductivity
Source: Gartner
Generalized Pattern of Research and Commercialization From Conception to Productive Use
EDA Minimizes the Trough of Disillusionment
EDA Maximizes the Slope of Enlightenment
EDA Can Lead to Concrete Evalua$on of New
Technologies
NSF NEB Program–EDA should be included
SiGdaspecial interest group on
design automation
THE DESIGN AUTOMATION OF THINGS Looking Forward: New Markets and Applications
SiGdaspecial interest group on
design automation
THE POWER OF DESIGN AUTOMATION
Success of Electronic DA • Facilitated unprecedented
exponential advancement of Si/CMOS
• EDA separates – Design – Construction – Optimization
• Clear abstraction & predictive models of low-level behavior allows: – High level analysis – Optimization – Verification
DA Techniques Beneficial to Areas • Require analysis before
construction • Lack appropriate abstractions • Rely on both optimization and
analysis to meet specs • Can make low cost alternatives to
existing high cost products • Need efficient assessment of
outcomes prior to construction
Biological and Medical Technologies
Cyber-‐physical Systems
Cyber-‐secure Systems
SiGdaspecial interest group on
design automation
NEW MARKETS WHERE DESIGN AUTOMATION CAN BE TRANSFERRED
Identified Potential
CPS/loT
Automotive Energy
Robotics
Medical Technology
High Barrier
DFS/Verification
Emerging
DF Wearables
Low Barrier
Identified Potential
High Barrier
Emerging
Low Barrier
Identified Potential
CPS/loT
High Barrier
Emerging
Low Barrier
Identified Potential
CPS/loT
Automotive
High Barrier
Emerging
Low Barrier
Identified Potential
CPS/loT
Automotive Energy
High Barrier
Emerging
Low Barrier
Identified Potential
CPS/loT
Automotive Energy
Robotics
High Barrier
Emerging
Low Barrier
Identified Potential
CPS/loT
Automotive Energy
Robotics
High Barrier
Emerging
DF Wearables
Low Barrier
Identified Potential
CPS/loT
Automotive Energy
Robotics
High Barrier
DFS/Verification
Emerging
DF Wearables
Low Barrier
SiGdaspecial interest group on
design automation
NSF CPS– EDA iden8fied but missing
NSF STARSS– EDA also missing
NSF EFRI– BioDA poten8al
topic
CROSS CUTTING CHALLENGES Abstractions, Metrics, and Benchmarks
Education and Workforce Synergies with Computer Architecture
SiGdaspecial interest group on
design automation
SiGdaspecial interest group on
design automation
EDUCATING IN EDA “EDA? But I want to save the planet!”
“EDA? But I want to do something cool!”
• Students want to impact society
• EDA is not perceived to “change the world” • EDA is TWO levels of
indirecDon away from “cool”
“EDA? That’s too hard! I have to learn physics and algorithms and
stuff!”
“What do you mean, build the tools to make the chips that enable the
smartphone? Can’t I just write apps for Google?”
Industry University Partnerships!
Ac8vity creates excitement
CURRICULUM AND STRATEGY
• I teach the same old courses the same way, it works!
• Students can’t learn this stuff before they know “the basics”
• Students will see the importance of MY course
• We should call this course what it is: “formal methods of verificaDon” X BeZer marke8ng to
students
Make EDA fun (e.g., Crowdsourcing)
Cri8cal Mass & MOOCs
SiGdaspecial interest group on
design automation
CONCLUSIONS
SiGdaspecial interest group on
design automation
DIMENSIONS OF FUTURE EDA ACTIVITIES
Electronics: Hybrid CMOS with Emerging
Technologies
New Markets:Cyber-physical, Cyber-secure, and Bio-medical Technologies
Traditional EDA Tool-kitImmediate Need: EDA for
scaled CMOS + product ready tech
Immediate Need: EDA applied to near fields – automotive,
robotics, and energy
EDA Approaches on Big Data
Transformative: Big data research–system level design and verification
Transformative: EDA big-data methodologies applied to far
fields – synthetic bio, systems bio, medical devices
Electronics: Hybrid CMOS with Emerging
Technologies
Electronics: Hybrid CMOS with Emerging
Technologies
New Markets:Cyber-physical, Cyber-secure, and Bio-medical Technologies
Electronics: Hybrid CMOS with Emerging
Technologies
New Markets:Cyber-physical, Cyber-secure, and Bio-medical Technologies
Traditional EDA Tool-kit
Electronics: Hybrid CMOS with Emerging
Technologies
New Markets:Cyber-physical, Cyber-secure, and Bio-medical Technologies
Traditional EDA Tool-kitImmediate Need: EDA for
scaled CMOS + product ready tech
Electronics: Hybrid CMOS with Emerging
Technologies
New Markets:Cyber-physical, Cyber-secure, and Bio-medical Technologies
Traditional EDA Tool-kitImmediate Need: EDA for
scaled CMOS + product ready tech
Immediate Need: EDA applied to near fields – automotive,
robotics, and energy
Electronics: Hybrid CMOS with Emerging
Technologies
New Markets:Cyber-physical, Cyber-secure, and Bio-medical Technologies
Traditional EDA Tool-kitImmediate Need: EDA for
scaled CMOS + product ready tech
Immediate Need: EDA applied to near fields – automotive,
robotics, and energy
EDA Approaches on Big Data
Electronics: Hybrid CMOS with Emerging
Technologies
New Markets:Cyber-physical, Cyber-secure, and Bio-medical Technologies
Traditional EDA Tool-kitImmediate Need: EDA for
scaled CMOS + product ready tech
Immediate Need: EDA applied to near fields – automotive,
robotics, and energy
EDA Approaches on Big Data
Transformative: Big data research–system level design and verification
SiGdaspecial interest group on
design automation
ORGANIZERS
Alex Jones – Pim Iris Bahar -‐ Brown Srinivas Katkoori -‐ USF
Patrick Madden – Binghamton Diana Marculescu -‐ CMU Igor Markov -‐ Michigan
SiGdaspecial interest group on
design automation
PARTICIPANTS AND CONTRIBUTORS R. Iris Bahar, Sankar Basu, Sanjukta Bhanja, Randy Bryant, Paul Bunyk, Krish Chakrabarty, Yiran Chen, Derek Chiou, Bob Colwell, Andre DeHon, Sujit Dey, Alex Doboli, Nik Dutt, Dale Edwards, Jim Faeder, Richard Goering, Patrick Groeneveld, Ian Harris, Mark Johnson, Alex Jones, Bill Joyner, Ramesh Karri, Srinivas Katkoori, Selcuk Kose, Steve Levitan, Hai Li, Xin Li, Patrick Madden, Diana Marculescu, Radu Marculescu, Igor L. Markov, Pinaki Mazumder, Mac McNamara, Noel Menezes, Prabhat Mishra, Natasa Miskov-Zivanov, Kartik Mohanram, Vijay Narayanan, Sani Nassif, John Nestor, David Pan, Mandy Pant, Sudeep Pasricha, Rob Rutenbar, Sachin Sapatnekar, Lou Scheffer, Carl Sechen, Don Thomas, Josep Torrellas, Jacob White, Mehmet C. Yildiz.
SiGdaspecial interest group on
design automation
FINAL REPORT AVAILABLE
Workshops on Extreme Scale Design Automation (ESDA) Challenges and Opportunities for 2025 and Beyond
R. Iris Bahar, Alex K. Jones, Srinivas Katkoori, Patrick H. Madden, Diana Marculescu, and Igor L. Markov
AbstractIntegrated circuits and electronic systems, as well as design technologies, are evolving at a great rate—both quantitatively and qualitatively. Major developments include new interconnects and switching devices with atomic-scale uncertainty, the depth and scale of on-chip integration, electronic system-level integration, the increasing significance of software, as well as more effective means of design entry, compilation, algorithmic optimization, numerical simulation, pre- and post-silicon design validation, and chip test. Application targets and key markets are also shifting substantially from desktop CPUs to mobile platforms to an Internet-of-Things infrastructure. In light of these changes in electronic design contexts and given EDA’s significant dependence on such context, the EDA community must adapt to these changes and focus on the opportunities for research and commercial success. The CCC workshop series on Extreme-Scale Design Automation, organized with the support of ACM SIGDA, studied challenges faced by the EDA community as well as new and exciting opportunities currently available. This document represents a summary of the findings from these meetings.
Workshop ParticipantsR. Iris Bahar, Sankar Basu, Sanjukta Bhanja, Randy Bryant, Paul Bunyk, Krish Chakrabarty, Yiran Chen, Derek Chiou, Bob Colwell, Andre DeHon, Sujit Dey, Alex Doboli, Nik Dutt, Dale Edwards, Jim Faeder, Richard Goering, Patrick Groeneveld, Ian Harris, Mark Johnson, Alex Jones, Bill Joyner, Ramesh Karri, Srinivas Katkoori, Selcuk Kose, Steve Levitan, Hai Li, Xin Li, Patrick Madden, Diana Marculescu, Radu Marculescu, Igor L. Markov, Pinaki Mazumder, Mac McNamara, Noel Menezes, Prabhat Mishra, Natasa Miskov-Zivanov, Kartik Mohanram, Vijaykrishnan Narayanan, Sani Nassif, John Nestor, David Pan, Mandy Pant, Sudeep Pasricha, Rob Rutenbar, Sachin Sapatnekar, Lou Scheffer, Carl Sechen, Don Thomas, Josep Torrellas, Jacob White, Mehmet C. Yildiz, Hao Zheng.
WORKSHOPS ON EXTREME SCALE DESIGN AUTOMATION (ESDA) CHALLENGES AND OPPORTUNITIES FOR 2025 AND BEYOND
SiGdaspecial interest group on
design automation
Workshops on Extreme Scale Design Automation (ESDA) Challenges and Opportunities for 2025 and Beyond
R. Iris Bahar, Alex K. Jones, Srinivas Katkoori, Patrick H. Madden, Diana Marculescu, and Igor L. Markov
AbstractIntegrated circuits and electronic systems, as well as design technologies, are evolving at a great rate—both quantitatively and qualitatively. Major developments include new interconnects and switching devices with atomic-scale uncertainty, the depth and scale of on-chip integration, electronic system-level integration, the increasing significance of software, as well as more effective means of design entry, compilation, algorithmic optimization, numerical simulation, pre- and post-silicon design validation, and chip test. Application targets and key markets are also shifting substantially from desktop CPUs to mobile platforms to an Internet-of-Things infrastructure. In light of these changes in electronic design contexts and given EDA’s significant dependence on such context, the EDA community must adapt to these changes and focus on the opportunities for research and commercial success. The CCC workshop series on Extreme-Scale Design Automation, organized with the support of ACM SIGDA, studied challenges faced by the EDA community as well as new and exciting opportunities currently available. This document represents a summary of the findings from these meetings.
Workshop ParticipantsR. Iris Bahar, Sankar Basu, Sanjukta Bhanja, Randy Bryant, Paul Bunyk, Krish Chakrabarty, Yiran Chen, Derek Chiou, Bob Colwell, Andre DeHon, Sujit Dey, Alex Doboli, Nik Dutt, Dale Edwards, Jim Faeder, Richard Goering, Patrick Groeneveld, Ian Harris, Mark Johnson, Alex Jones, Bill Joyner, Ramesh Karri, Srinivas Katkoori, Selcuk Kose, Steve Levitan, Hai Li, Xin Li, Patrick Madden, Diana Marculescu, Radu Marculescu, Igor L. Markov, Pinaki Mazumder, Mac McNamara, Noel Menezes, Prabhat Mishra, Natasa Miskov-Zivanov, Kartik Mohanram, Vijaykrishnan Narayanan, Sani Nassif, John Nestor, David Pan, Mandy Pant, Sudeep Pasricha, Rob Rutenbar, Sachin Sapatnekar, Lou Scheffer, Carl Sechen, Don Thomas, Josep Torrellas, Jacob White, Mehmet C. Yildiz, Hao Zheng.
WORKSHOPS ON EXTREME SCALE DESIGN AUTOMATION (ESDA) CHALLENGES AND OPPORTUNITIES FOR 2025 AND BEYOND
For more informaDon hZp://www.cra.org/ccc/visioning/visioning-‐ac8vi8es/esda
THANK YOU! Alex K. Jones
University of Pittsburgh [email protected]
SiGdaspecial interest group on
design automation