Transcript gsrc0203_overview-kahng

Meeting Agenda
 Introduction (9:00-9:30)
 Application driver focus of the GSRC, and implications for C.A.D. Theme
 C.A.D. Theme status and futures
 GTX (Technology Extrapolation)
 Tool status and current development (Mike Oliver) (9:30-9:45)
 Recent work (9:45-10:45)




DRAM (Michael Wang (Dai))
Interconnect modeling (Xuejue Huang (King))
Global signaling (Himanshu Kaul (Sylvester))
What can GTX do to support drivers? (10:45-11:30)
 Bookshelf (CAD-IP Reuse)
 Status and summary of recent work (Igor Markov) (11:30-noon)
 Open Access (1:00-2:00)
 What can Bookshelf do to support drivers = vertical benchmarks++ ? (2:00-2:45)
 Roadmap for C.A.D. Theme + action items (until adjourn)
1
C.A.D. Initiatives
 Specification Gap: e.g., What will be the critical design problem?
 GTX
 GTX models include canned optimizations = canned design space explorations
 Development and Delivery Gap: e.g., How to deploy DT better/faster?
 Bookshelf
 Measurement Gap: e.g., Did achievable design improve?
 Metrics
 Definition of success
 (Next up: Education? Measuring research process?)
Shared Context Is A Force Multiplier
2
Ubiquitous Node Design Specifications
 Major Constraints (depending upon the application)

Cost: < 1 $

Size: 1 mm3 … 1 cm3

Power: between 10 mW and 100 mW (depending upon ubiquitousness and
mobility)
 Hybrid

Mixed-signal (sensing, air interface, power train)

Mixed technology (passives, MEMs)
 Limited flexibility

Downloadable and adaptable application layer

Parameterizable interfaces
3
Perspective:
Single-Chip Bluetooth Radio (Alcatel, 2001)
4
PicoNode V3 Architecture
SIF = sensor interface
16kB
CODE
4kB
XDATA
256
DATA
Chip
Supervisor
DW8051
Flash
Storage
Serial
sfrbus or membus?
20MHz
Clock Source
FlashIF
SIF
ADC
MAC
Voltage
Voltage
Supply
Voltage
Supply
Supply
SIF
LocalHW
Serial
GPIO
ADC
PHY
Sensor1
Sensor2
PrgThresh0 PrgThresh1
User
Interface
OOK
Receiver
Tx0
Tx2
OOK
Transmitter
5
Challenge: Packaging
“Smart Dust” mote
Combines sensing,
computation, optical
communication, and solar
array
[K. Pister (UCB)]
6
Home Networking Driver
 System: 10 GOps/s

2-3 types of I/O ports (PCI, USB, Ethernet)

Bus speed: 100-200MHz

Bus bandwidth: 2-4 Gb/s

Memory speed: 100-400MHz
 Core Processor:

Transistors: 5-60 Million

Clock frequency: 500MHz-1 GHz

Area: 100-250mm2

Power: >= 50W
7
Specification of a Home Network 2005



System: (18 GOps/s)

4-5 types of I/O ports (PCI, USB, Ethernet, 802.11, IEEE 1394, Bluetooth)

Area: 10-50mm2 , Power: <5W

Bus speed: 400-800MHz

Bus bandwidth: 10-20 Gb/s

Memory speed: 400MHz-1GHz

Analog part ?
Core Processor:

Transistors: 5-50 Million

Clock frequency: 500MHz-1 GHz

Area: 100-200mm2

Power: < 20W
Network Processor:

Transistors: 50-100Million

Clock frequency: 200-800MHz (2-10 PE’s)

Area: 100-300 mm2

Power: < 10W
8
Home Network in 2005
laptop
Broadband modem (cable/xDSL)
Ethernet
Embedded
gateway
802.11
HDTV
IEEE1394
Bluetooth
webpad
Ethernet
Printer
pc
9
Design Challenge: What Does the Military
Want A Modern Fighter/Attack Radar To
Do?
such as:
Background*:
Contribute to…
by providing…
Attack Mission
Target acquisition
Target discrimination
Weapon delivery support
Detect, ID, locate
friend or foe
AMRAAM, JDAM, etc.
Survivability
Situational awareness
Low observability
EMCON and LPI
Electronic attack
Navigation aid
Air & surface
Low RCS
Power management
Spoofing, jamming
TF/TA
Supportability
High availability
Small logistics footprint
Long MTBF, short MTTR
Min. spares kit, test set
Affordability
Low cost of ownership
No DMS problems
Low impact on the aircraft Weight, cooling
* Source: Mike Lucas, Northrop Grumman
such as:
These broad
objectives were set
forth by Mike
Lucas (Northrop
Grumman) in his
DARPA
presentation last
December
10
The Future Design Challenge:
A Digital ESA Radar
Digital ESA
A/D
A/D
Aircraft power
Power
Supply
Array
Driver
A/D
Digital
Beam
Former
A/D
BSC
.
.
.
Exciter
Processor/
Controller
to Mission
Computer
* Source: Mike Lucas
11
Digital Radar Technology Directions
 AESA/Receiver

More Channels: 1000

Higher A/D Sampling Rates: 1Gsps and Above

Higher Dynamic Range: 14-15 bits
 Beam Forming

Higher Signal Processing Throughputs: 100 of TFLOPS

Continued Power Constraints: Needs 100 GFLOPS/Watt

Optimized Mission Specific Processing, Low Cost ASICs

High AESA to Beam Former Bandwidth: Multi Tbps
 Signal Processing

Parallel Processing Architectures with High Bisection Bandwidth

Increased Use of COTS and Standards

Increased Software Reuse
* Source: Mike Lucas
12
Drivers, GSRC, and C.A.D. Theme
 GSRC is now managed by DARPA  design drivers are key


Require quantified proofs of impact
Also, quarterly progress reports, etc.
 C.A.D. Theme status






+ “Living Roadmap”: high perceived impact and value
+ Research not centrally managed, aligned  freedom to do as we please
+ Funding this year was stable for everyone
- Bookshelf fairly dormant, external participation only for $
- Metrics dormant (but, progressing in Cadence)
- Integration with other Themes, FRCs is minimal


Fabrics, Power/Energy, System-Level integrations should be deep/active
Integrations with Interconnect, MSD, C2S2 focus centers should be deep/active
 Theme processes will have to change



Roadmap, concrete plans, quarterly progress report roll-up
Alignment with theme work
Conference calls, … (other mechanisms)
13
Living Roadmap (of Application Drivers)
 Network, telecommunications, embedded computing systems



Synchronous buses  1Gbps, differential signaling  10Gbps
Network, optical interfaces have multipliers of 10x, 4x (faster than device density,speed)
Train wrecks: chip-to-package and system-level interconnects (materials, signaling
standards, implementation costs), power, design TAT, cost
 Appropriate metrics are “non-traditional”: density, cost, performance, power, and
RAS (reliability, availability, and serviceability)




Density: connections and bandwidth per cm(2,3) , watts/m3
Performance: How many interconnect/cm(2,3)? How long are traces? What types of
signals, and what voltage levels, will meet signaling rate needs?
Cost: decompositions (mother, switch/routing, control, port interface, application), and
dimensions (per (gE, FC, DWDM, …) port, Gbps, MIPS, $ …)
RAS: unintentionally / intentionally (for func) dropped bits/packets dropped, failure rates
 Many models to build and integrate: SOC integration (what is integratable, at what
cost), analog circuits/DT (how badly do these fail to scale), design quality and cost,
power (circuits, multi-Vdd/Vt/tox / biasing, GALS/GSLA, …), manufacturing interface
(variability, NRE, layout densities, …)
 GTX within DT: What are the key design technology needs?



Application roadmap (= ITRS System Drivers Chapter = complement to ITRS ORTCs)
Application product ROI = value/cost (= attributes not yet well-defined/-measured)
Impacts of Design Technology (== Metrics initiative)
14
Bookshelf
 Goal is to produce component-based, application-specific design
methodologies and flows


How will the methodology space be explored, and flows prototyped?
Where are the reusable components?
 Open-source (understandable, reusable), malleable DT components
 Centered on back end, completely missing AMS capabilities, …
 Common data model and access mechanism (and repository?)
 OpenAccess source code release
 Design Drivers very close to vertical benchmarks (= existing Bookshelf
slot)


Recent overtures from IBM, LSI w.r.t. OpenAccess, working vertical benchmarks
Potential work with Fabrics on snap-on flows, etc.
KEY: Common DT Infrastructure
 Other: synergy with education in VLSI design, design technology
15
Metrics
 Goal: measure and improve
 Systems
 Processes
 Relevant system attributes / metrics
 System value
 System cost (design, production)
 From system ROI, have a platform from which to evaluate technology ROI
 Technology cost (research, advanced research, development, …)
 Supporting technologies / infrastructures (data mining, parameter
identification, model fitting)
 Other: Research process
 What is the impact of FCRP ? (# newspaper articles? # papers? Coauthorship
statistics? Survey results? Scientific health of (Design/Test, Interconnect, etc.)
communities?) == part of original “Measure and Improve” goals
16
Meeting Agenda
 Introduction (9:00-9:30)
 Application driver focus of the GSRC, and implications for C.A.D. Theme
 C.A.D. Theme status and futures
 GTX (Technology Extrapolation)
 Tool status and current development (Mike Oliver) (9:30-9:45)
 Recent work (9:45-10:45)




DRAM (Michael Wang (Dai))
Interconnect modeling (Xuejue Huang (King))
Global signaling (Himanshu Kaul (Sylvester))
What can GTX do to support drivers? (10:45-11:30)
 Bookshelf (CAD-IP Reuse)
 Status and summary of recent work (Igor Markov) (11:30-noon)
 Open Access (1:00-2:00)
 What can Bookshelf do to support drivers = vertical benchmarks++ ? (2:00-2:45)
 Roadmap for C.A.D. Theme + action items (until adjourn)
17