Transcript Slides for Richard - FCRP Review, Washington DC

Calibrating Achievable Design
Roundtable Discussion
June 9, 2002
Wayne Dai, Andrew Kahng, Tsu-Jae King, Wojciech
Maly, Igor Markov, Herman Schmit, Dennis Sylvester
Facilitator: Bill Joyner, IBM/SRC
Recent Progress in Technology
Extrapolation and GTX
•
•
•
•
Transition-aware and active-shielding global signaling
approaches, RLC interconnect performance models
Taxonomy and correlated models of process variability,
and framework for examining impact on achievable design
New DRAM access time model in support of eDRAM vs.
multi-die cost/performance model
Release of ITRS-2001 ORTCs in GTX
•
•
•
Updated GTX version in Sematech ITRS-2001 website
Enhancements in June02 release
Ongoing studies, modeling of ITRS “shared red bricks”
•
Low-k benefit, variability tolerance of design, design-limiting
process rules, …
GSRC Annual Review, 020609
2
Recent Progress in CAD-IP Reuse
•
•
•
•
Prototype release of Bookshelf.exe
• New infrastructure that hosts executable Bookshelf
content and allows remote execution
• Web interface, distributed computation platform, and
scripting capabilities
New tilable (i.e., scalable) circuit benchmarks
• Network tile
• Processing elements (floating-point adder, floatingpoint multiplier, programmable FIR filter)
Initial discussions of OpenAccess API and data model
Benchmark scaling approaches
GSRC Annual Review, 020609
3
Recent Progress in METRICS
•
•
•
•
Expanded METRICS system to front-end acceptance and
clock tree synthesis domains
Work with STRJ-WG1 (Japan Roadmap group) to obtain
design quality / value survey questions that will be used to
survey U.S. companies
Started to propagate open-source METRICS system
available to university user group communities
• Exploring potential links to Bookshelf.exe
• METRICS Birds-of-Feather meeting at DAC-2002
Developing and applying machine learning techniques to
improve existing data mining and regression infrastructure
GSRC Annual Review, 020609
4
Summary: Roles of the C.A.D. Theme
 Identify best opportunities for “sharing of red bricks” between EDA
and other semiconductor supplier industries

GTX studies + Manufacturing Calibration: Is low-k worth the development cost?
What is the best interconnect process architecture for 65nm? What FEOL and
BEOL variabilities can designers tolerate? What is the most cost-effective
memory-logic integration?
 Repository of best known methods,
models, metrics


C.A.D. Theme
ITRS
Living ITRS
Semi
Industry
Connects applications and drivers (e.g.,
Models and Calibrations
Design
ITRS MPU, SOC) to technology roadmap
What is the design problem? Houses
Connects algorithm (e.g., Fabrics) and
design technology (e.g., Power-Energy)
How should Design help
within GSRC and to designs, interconnects,
solve ITRS red bricks?
devices, and materials (= other FRCs)
VLSI Design Education
 Agent of culture change
 Measurement & characterization of EDA
 Open-source CAD-IP; vertical benchmarking
 New vectors: e.g., reusable curriculum IP
for VLSI, VLSI design, VLSI design
technology education
GSRC Annual Review, 020609
Other
FRCs
Manufacturing Calibration
Open-Source CAD-IP
METRICS & Design Process Opt
EDA
Industry &
Academia
Other
GSRC
Themes
5
C.A.D. Theme Draft Roadmap
Interactions
o NEMI, C2S2 FRC – app roadmaps
o Fabrics– “SIP design tech backplane”
o Fabrics Theme – Bookshelf, flows research
o ITRS community o Power-Energy Theme – capture/guide
o MSD FRC – mfg calibration, manufacturability shared red bricks
o Self-Test Theme – capture/guide
o Interconnect FRC – model/calib, mfg/var shared red bricks
o NSF (education)
1/99
1/01
Today
1/03
1/04
1/05
1/99
1/01
Today
1/03
1/04
1/05
o Natl VLSI design curriculum initiative
o ITRS “shared red brick” modeling, analysis, roadmapping
Initiatives and Milestones
o Vertical benchmarks: cell libraries, DSP, netwk
o 30% of ITRS in GTX
o Bookshelf @ 25 slots, first papers
o METRICS  broader academic release
o GTX model dirs = mfg cost/cal, global sig, mem-logic, app domains
o “SIP design tech backplane”
o BX.exe v.1
o METRICS v.1 + start industry integration
is METRICized
o GTX v.2 (with ITRS-2001)
o GTX v.1 + ITRS support
o BK 25% PD adopt, 50% slot interop
o Fabrics  C.A.D. + Constructive Fabrics Themes
o Open-source RTL-GDSII flow
o Start Fabrics Thrust, initial data model / flow discussions
GSRC Annual Review, 020609
6
Foci of Industry Feedback and Interaction
•
•
•
•
OpenAccess (common data model) – what is GSRC doing?
CAD-IP reuse – Bookshelf Slots, codes, functionality
Benchmarks – “industry-strength validation of research”
Education initiative?
•
•
ITRS needs (= how to best share “red bricks”)
•
•
•
•
•
•
National-scale VLSI education infrastructure
Incremental benefits of lower k
“Optimal interconnect stack”? (balancing power delivery,
routing density, delay and signal integrity performance, …)
“What variability can designers tolerate?”
ESD protection impact on off-chip signaling bandwidth
Understanding how ground rules impact layout density
What else?
GSRC Annual Review, 020609
7
Roundtable Discussion Questions
•
•
•
•
For our industrial partners: in your designs, what is the largest
problem that you face ?
• And how does the research work in this Theme help?
What is being done well in this Theme?
What could be improved?
On one or two slides (max), each industrial co-facilitator please
summarize key conclusions from this discussion.
Questions and issues for discussion
not restricted to this set.
GSRC Annual Review, 020609
8