cows, cowbirds, and the conundrum of eda research … 2002 2003 2004 2005 2006 2007 2008 2009 2010...

55
Cows, Cowbirds, and the Conundrum of EDA Research Funding DAC-2017 Technical Panel, Session 30 Moderator: Andrew B. Kahng, UCSD This panel is 4:30pm – 6:00pm (some versions of the program are incorrect)

Upload: dinhthien

Post on 13-May-2018

215 views

Category:

Documents


1 download

TRANSCRIPT

Cows, Cowbirds, and the Conundrum of EDA Research Funding

DAC-2017 Technical Panel, Session 30

Moderator: Andrew B. Kahng, UCSD

This panel is 4:30pm – 6:00pm (some

versions of the program are incorrect)

Cows? Cowbirds?

• “If you get free milk, why buy the cow?”

• Cowbird: A (parasitic) species that lays its eggs in other species’ nests, so that the burden of raising its young is borne by the other species.

Why This Panel ?

• More burden on EDA as scaling, cost levers run out of steam

• EDA research ecosystem today

– Students: Academic Research � EDA Companies– Problems: Semi Companies � EDA Companies � Academic Research– $$$: {Semi Companies, Government} � Academic Research

Flows

Academia

EDA Companies Semi

Companies

Government

Why This Panel ?

• More burden on EDA as scaling, cost levers run out of steam

• EDA research ecosystem today

– Students: Academic Research � EDA Companies– Problems: Semi Companies � EDA Companies � Academic Research– $$$: {Semi Companies, Government} � Academic Research

• Consequences have been accumulating

– “Design capability gap” (2013 ITRS): Design metrics vs. Tech metrics

– Graying of the EDA industry

– Action is shifting (geographies, conference topics, industry sectors…)

Yesterday’s SKY Talk: Prof. Shaojun Wei

Why This Panel ?

• More burden on EDA as scaling, cost levers run out of steam

• EDA research ecosystem today

– Students: Academic Research � EDA Companies– Problems: Semi Companies � EDA Companies � Academic Research– $$$: {Semi Companies, Government} � Academic Research

• Consequences have been accumulating

– “Design capability gap” (2013 ITRS): Design metrics vs. Tech metrics

– Graying of the EDA industry

– Action is shifting (geographies, conference topics, industry sectors…)

• This panel

– Is there an EDA research funding problem? “Natural?” “Crisis?”

– What must be done, by whom?

The Panelists

• Mamta Bansal, Qualcomm

• Leon Stok, IBM

• Tim Cheng, HKUST

• Shishpal Rawat, IEEE CEDA

• Andreas Olofsson, DARPA

Questions to Panel

• Is there an EDA research funding problem?– How does this impact the IC/EDA industries?

• Why doesn’t academic EDA research attract more investment?– Wrong attitudes or problems? Not useful?– Root causes?

• How can we achieve a virtuous cycle of research $, results, and industry benefits?– New areas, frameworks, enablements …

The Panelists

• Mamta Bansal, Qualcomm

• Leon Stok, IBM

• Tim Cheng, HKUST

• Shishpal Rawat, IEEE CEDA

• Andreas Olofsson, DARPA

11

Cows, Cowbirds, and the Conundrum of EDA Research Funding

Mamta Bansal

Transforming World

13

DA Ready ?

HealthcareSmart homes DronesDatacenter

WearablesAutomotive NetworkingSmart cities

14

Process Technology

Implementation

Thermal, EM

Analog- Digital co-design

Data Management

Machine Learning

Security Encryption

The Disruptions

15

A Work Model

Industry Academia EDA

Problem Definition

Access

Research

Prototype

End2End

Productize

16

Summary

� Widening gap between semiconductor industry roadmap and DA

� Industry and academia work closely to define the issues better,

prototype solutions

� EDA plays a big role in providing comprehensive solution

The Panelists

• Mamta Bansal, Qualcomm

• Leon Stok, IBM

• Tim Cheng, HKUST

• Shishpal Rawat, IEEE CEDA

• Andreas Olofsson, DARPA

18 © 2017 IBM Corporation

Cows, Cowbirds and the Conundrum of EDA Research Funding

Leon Stok, VP, IBM Systems Group

19 © 2017 IBM Corporation

What are we looking for from universities?

20 © 2017 IBM Corporation

Is the next wave of innovation in EDA required ?

21 © 2017 IBM Corporation

Popularity of databases

[https://db-engines.com/en/ranking_categories]

22 © 2017 IBM Corporation

� Using Neo4j we now have an unprecedented capability to store detailed timing information in a highly efficient format, enabling very powerful query analysis.

� Example

–nearly 100 million nodes and associated timing properties in Neo4j.

–Using an IBM Power8 Linux server, we are able to import data within 11 minutes.

–The resulting graph database is indexed on several key query parameters, such instance names and slack values, within a matter of a few minutes.

–Once indexed, we have demonstrated the ability to perform lookups, for example iterating through all timing end points in the top-level chip (Power9) run, and returning the worst N sorted by slack within 2 seconds.

Some stats on running static timing analysis on a graph database

23 © 2017 IBM Corporation

Popularity of databases by database model

[https://db-engines.com/en/ranking_categories]

24 © 2017 IBM Corporation

Is there an EDA funding problem ?

� CADathlon winners [http://www.sigda.org/programs/cadathlon/winners]

–Last time USA 1st and 2nd place: 2010

–Number USA of Top 2 finishers in last 5 years: 1

� Back to Basics

–Stability

–Scalability

The Panelists

• Mamta Bansal, Qualcomm

• Leon Stok, IBM

• Tim Cheng, HKUST

• Shishpal Rawat, IEEE CEDA

• Andreas Olofsson, DARPA

26 © 2017 IBM Corporation

Tim ChengHong Kong Univ. of Science and Technology

EDA Funding is Very Dry

• Fewer research universities hired new faculty in EDA recently

• Some faculty used to work on EDA left the field and some

diverted fraction of research efforts to other fields

• Fewer PhD produced, fewer new talents entering the field

• Average employee age of EDA companies is higher than those

in other hi-tech sectors

Widening gap for a critically important field

to the semiconductor industry!

Growing the EDA Pie

• Expanding EDA research to XDA research

– X for photonics, flexible electronics, etc. and

– their heterogeneous integration with silicon electronics

• DA research playing critical roles in several large-scale,

industry-driven, solution-/application-driven projects

• Extending DA for broader technology platforms, for

heterogeneous and system integration

$75 million DoD funding over 5 years + $96

million cost sharing from non-federal sources

$110 million DoD funding over 5 years + $500

million cost sharing from non-federal sources

Two of US Manufacturing Innovation Institutes -Integrated Photonics and Flexible Hybrid Electronics

American Institute for Manufacturing Integrated Photonics

30

• EPDA: Electronic and Photonic Design Automation

• MPWA: Multi-Project Wafer and Assembly

• ICT: Inline Controls and Test

• TAP: Test, Assembly, and Optical PackagingD

esi

gn

an

d M

an

ufa

ctu

rin

g

Photonics Technology Challenges

Trends of monolithic integrationAIM Matrix Operation Model

Electronic & Photonic Design Automation (EPDA)

Goals: Develop/enhance a set of design tools for

photonics and combined electronic-photonic

components

• Compact models for photonic devices

• Integrated electronic-photonic design environment

• PDK for silicon, InP, and hybrid photonic devices

• Process variation models and design-for-

manufacturing tools

NEXTFLEX – America’s Flexible Hybrid Electronics Manufacturing Institute

32

•Funded by US DoD NextFlex

•Partnership of HPE, UCSB, Stanford,

Georgia Tech, Boeing, Cadence,

ANYSYS

DA Tools for Flexible Hybrid Electronics

International DA Research Consortium?

• Global semicon industry landscape continues to change

– E.g. China stepping up efforts to become a major player in IC design and manufacturing

– “Made in China 2025” Initiative to improving self-sufficiency rate for ICs in China

– Many new initiatives in design and manufacturing – but few solely devoted to EDA

• EDA industry landscape does not change as much

• Opportunity to form an international DA research consortium?

The Panelists

• Mamta Bansal, Qualcomm

• Leon Stok, IBM

• Tim Cheng, HKUST

• Shishpal Rawat, IEEE CEDA

• Andreas Olofsson, DARPA

Shishpal RawatPresident

36

DAC Panel 2017Cows, Cowbirds and the Conundrum of EDA Research Funding

Background

� Council on EDA: All aspects of enablement & productivity improvement in

design and development of electronic systems and embedded systems.

� Historically CAD topics evolved to suit designs of increasing complexity

– Device Modeling and Circuit Analysis tools

– Layout Automation; Tapeout Tooling

– High Level Synthesis, Logic Synthesis and Physical Synthesis

– Abstract Modeling; Virtual Machines & Platforms for HW/SW co-development

� As designs became bigger we got cross Domain tools & IP modules

– Power Estimation; Signal integrity; Clocking; 3D IC’s

– IP modules for imaging, graphics, security & privacy etc. boosted designer productivity.

6/27/201737

� Majority of the designs are produced with current EDA tooling

� A minority of the designs (bigger chips, new nodes, new packaging etc) need consistent

redevelopment and optimization.

– EDA industry works closely with lead customers

– EDA industry invests as much as a third of its revenue into these new developments

� New Designs are also being developed to address specific domains (Cyber-physical, IoT,

SmartCities, SmartEnergy, bio-inspired computation and learning

– Academia and Industry developing system level design tools

– Academia Support (NSF/SRC) ~$10M/yr + Special Projects (Multicore Design and Architecture; Failure

Resistant Systems; Energy Efficient Computing: Devices to Architectures)

– EDA companies contribute tools & support to academia

– EDA companies also provide some support to startups in various incubators. Start ups in turn

engage academia for some focused CAD research

6/27/201738

Approx 12,000 new designs each year

Design Productivity Gains Possible via EDA tools

� EDA has continuously raised the level of abstraction for major portions of

design starting with polygon editing

� EDA has enabled five orders of magnitude productivity since 1985 (Wally

Rhines, DVCon 2016)

� And in turn growing the EDA industry to $7.8B USD (2015)

� New Productivity platforms are needed to revolutionize next generation

EDA offerings.

© 2006 IBM Corporation

EDA Industry Evolution (~40 years)

1. Research & AdhocDevelopment – early 70’sSpice (1973)

2. Research & Large in-houseCAD teams (mid 80s –Intel, IBM, NEC, TI etc.)

3. Equity Investments & newbusiness models (mid 90s –Standardization, new biz models)

4. Supplier Consolidation(ongoing mergers/acquisitions,Higher barrier to enter markets)

Academia and Industry will Drive SDA

� Academia is focused on Emerging computing paradigms

– Data Centric computing with new devices (or conventional devices); Neuromorphic, AI

inspired chips, Cryogenic, Approximate, Adiabatic etc.

– Each computing paradigms is driving their own CAD development to support the build-out

of their systems.

– Researchers should also focus on validation and debug of such systems in "reasonable"

amount of time � Innovative validation/verification techniques.

– Successful implementation of some of these systems will eventually drive a new class of

EDA commercialization in the future (similar to cycles in TV, lighting).

� New Players in Industry

– Apple, Google, Microsoft, Facebook, Amazon etc. developing domain specific chips

6/27/201741

© 2006 IBM Corporation

The Evolution will Continue ……..

4 phases for EDA 4 phases for IP

4 phases for SDA (?)We are at in-housedevelopment stage

2017

2017

2017

Summary

� Is there an EDA research funding problem? - NO

– How does this impact the IC/EDA industries? – N/A

� Why doesn’t academic EDA research attract more investment? – EDA must be

viewed as an “integral” part of design; EDA is inter-disciplinary. It is an

attractive research topic (NSF Centers; DARPA funded STARnet/JUMP ….)

– (wrong attitudes? wrong problems? research outputs not useful? tech transfer hurdles? …)

– (root causes across academia, industry, consortia, government …)

� How can we achieve a virtuous cycle of research $, results, and industry

benefits? – New entrants in the market will drive this – Neuromorphic

Computing; Quantum Computing etc. Research focus must be augmented to

ensure that these systems TTM can be met (debugged & validated).

– New areas, frameworks, enablements …

6/27/201743

The Panelists

• Mamta Bansal, Qualcomm

• Leon Stok, IBM

• Tim Cheng, HKUST

• Shishpal Rawat, IEEE CEDA

• Andreas Olofsson, DARPA

Panel: Cows, Cowbirds, and the Conundrum of EDA Research Funding

Andreas Olofsson

20 June, 2017

Distribution Statement “A” (Approved for Public Release, Distribution Unlimited)

Distribution Statement “A” (Approved for Public Release, Distribution Unlimited) 46

Industry background

Fab Circuit Design

EDA/CAD

Logic Design

Semiconductor Test

• 12 years at TI and Analog Devices

• 9 years as CEO/founder at Adapteva

• Majority of career spent in the semiconductor trenches

Architecture

47

Why I came to DARPA

1

10

100

1,000

10,000

100,000

1,000,000

10,000,000

100,000,000

1,000,000,000

10,000,000,000

100,000,000,000

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

1960 1970 1980 1990 2000 2010 2020

Automation Transistors

Auto

mation

Tra

nsi

stors

8088

4004

486

Pentium4

Westmere

Distribution Statement “A” (Approved for Public Release, Distribution Unlimited)

48

Historical Chip Trend in Mil/Aero Market

$-

$100

$200

$300

$400

$500

$600

$700

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

ASIC & ASSP FPGA

Mil

lio

ns

Global Military/Aeronautics Shipments

Source: Multiple industry market trackers & DMEA internal data from FPGA manufacturers

Distribution Statement “A” (Approved for Public Release, Distribution Unlimited)

49

The EDA R&D Challenge

$M

Research Year

0

20

40

60

80

100

120

140

1 1.5 2 2.5 3 3.5 4

Industory Total Cost

DoD Total Cost

Industry Production Cost

DoD Production Cost

Status Quo

100% Automation2X silicon Area

100% Automation1X silicon Area

Distribution Statement “A” (Approved for Public Release, Distribution Unlimited)

50

Can EDA learn from Open Source Trends?

Facebook Market Cap: $433B!

LINUX

MemCache MySQL

Apache PhPThrift

Cassandra Jenkins

Yoga

Facebook Code

User Content

$15BFOSS

Codebase

Distribution Statement “A” (Approved for Public Release, Distribution Unlimited)

51

My DARPA dream…

…when I leave DARPA…

48 Hours

Code EDA

Distribution Statement “A” (Approved for Public Release, Distribution Unlimited)

www.darpa.mil

Distribution Statement “A” (Approved for Public Release, Distribution Unlimited) 52

The Panelists

• Mamta Bansal, Qualcomm

• Leon Stok, IBM

• Tim Cheng, HKUST

• Shishpal Rawat, IEEE CEDA

• Andreas Olofsson, DARPA

� IEEE CEDA President: EDA research reasonably well-funded!

� Global context: Trajectories of geographies (esp. China)

� New frameworks (Qualcomm, DARPA)

� “Beyond E”: XDA

What We Just Heard

� $10M/year (NSF, SRC) spreads thinly: 30 new Ph.D.s graduates/year – is this enough?� Will “XDA” provide enough $ for “fundamental DA research” ?

� Design is ad hoc – will design motivate foundational DA?� EDA and Open Source = A Marriage Made in …. ?

� Internal CAD: Would you open-source your script/toolware?� Yosys (Clifford W.): “service” vs. “novelty” and research $ ?� Would funding sources pay for “production-quality” research product?

� Ulf S.: Tim’s suggested international research consortium – how would this work?� (Difficult with government funding) � Any showstoppers seen by the panel? (cf. 3X redundancy of research; 10X less people � grows

capacity ………)� David Y.: “Big CAD follows Big Design” � Solve problems of Design community, not “CAD for CAD”…� Farinaz K.: CEDA DA Futures (“taking the E out of EDA”): � with XDA, which community? [core,culture]� Ramesh K.: Looking around, why are we painting rosy pictures?

� � This panel has CHALLENGED academia (to turn this around)� Regional trajectories – “let a thousand flowers bloom”, or problematic in some way?� How should problems flow?

Semi � Academia � EDA (Qualcomm, today)

OR: Semi � EDA � Academia ? (traditional)

Questions Posed to Panelists During Q&A