Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Page 1
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
11
�� Describe & verify product behaviorDescribe & verify product behavior
�� Describe product architecturesDescribe product architectures�� Explore HW/SW design tradeoffsExplore HW/SW design tradeoffs
ss Map behavior to architectureMap behavior to architecturess Use performance simulationUse performance simulationss Perform communication refinementPerform communication refinement
�� Integrated flow to implementation Integrated flow to implementation
SystemSystemBehaviorBehavior
SystemSystemArchitectureArchitecture
MappingMapping
Flow To ImplementationFlow To Implementation
CommunicationRefinement
BehaviorBehaviorSimulationSimulation
PerformancePerformanceSimulationSimulation
21
3
4
The Essence of Polis/Felix/VCC Project: Virtual The Essence of Polis/Felix/VCC Project: Virtual Component CoComponent Co--designdesign
1988:1988:
22
TheThe next level of Abstraction …next level of Abstraction …
abst
ract
Transistor ModelCapacity Load
1970’s
cluster
abst
ract
Gate Level ModelCapacity Load
1980’s
RTL
cluster
abst
ract
SDFWire Load
1990’s
IP Blocks
cluster
abst
ract
IP Block PerformanceInter IP Communication Performance Models
RTLClusters
SWModels
Year 2000 +
Page 2
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
33
Architectural ChoicesArchitectural Choices
µP
Prog Mem
MACUnit
AddrGenµP
Prog Mem
µP
Prog Mem
Satellite
ProcessorDedicated
Logic
Satellite
Processor
Satellite
Processor
GeneralPurpose
µP
Software
DirectMapped
Hardware
HardwareReconfigurable
Processor
ProgrammableDSP
Flex
ibili
tyFl
exib
ility
1/Efficiency (power, speed)1/Efficiency (power, speed)
44
OMAP™ Block DiagramOMAP™ Block Diagram
I-MMU D-MMU
I-Cache
RISC Core
MMU
I-CacheInternal
RAM/ROM
DSP Core+
Appl Coprocessors
DMA
Memory & Traffic Controller
ProgramMemory SDRAM
PeripheralsLCD Controller, Interrupt Handlers, Timers, GPIO, UARTs, ...
uuARM9 coreARM9 core
uu16KB I16KB I --cachecache
uu8KB D8KB D--cachecache
uu22--way set way set
associativeassociative
uu150 MHz150 MHz
uuC55x DSP coreC55x DSP core
uu16KB I16KB I --cachecache
uu8KB RAM set8KB RAM set
uu22--way set way set associativeassociative
uu200 MHz200 MHz
D-Cache
Page 3
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
55 66
Hardware Platforms Not Enough!Hardware Platforms Not Enough!
uuHardware platform Hardware platform has to be abstractedhas to be abstracted
uuInterface to the application software is the Interface to the application software is the “API”“API”
uuSoftware layer performs abstraction:Software layer performs abstraction:
ss Programmable cores and memory subsystem “hidden” Programmable cores and memory subsystem “hidden” by RTOS and compilersby RTOS and compilers
ss I/O subsystem with Device DriversI/O subsystem with Device Drivers
ss Network with Network Communication SoftwareNetwork with Network Communication Software
Page 4
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
77
Software PlatformsSoftware Platforms
Output DevicesInput devicesHardware Platform
I OHardware
Software
network
Software Platform
Application SoftwarePlatform API
API
RTO
S
BIOS
Device Drivers Net
wor
kCo
mm
unic
atio
n
Compiler
88
MiddlewareJavaTV, TVPAK, OpenTV, MHP/Java, proprietary ...
Applications
NexperiaNexperia HardwareHardware
Streaming andStreaming andPlatform SoftwarePlatform Software
Ker
nel:
pSO
S, W
in-C
E, J
avaO
S
NexperiaNexperia--DVP SoftwareDVP Software
uuNexperiaNexperia™ ™ --DVP Software ArchitectureDVP Software Architecture
ss Supports multiple Supports multiple OSsOSs and and middleware softwaremiddleware software
ss Abstracts platform functionality via Abstracts platform functionality via consistent APIsconsistent APIs
uuNexperiaNexperia™™--DVP Streaming SoftwareDVP Streaming Software
ss Encapsulates implementation of Encapsulates implementation of streaming media components streaming media components (hardware and software)(hardware and software)
uuNexperiaNexperia™ Platform Software™ Platform Software
ss OS independent device drivers for onOS independent device drivers for on--chip and offchip and off --chip deviceschip devices
Page 5
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
99
HW layerHW layer
SW Platform layer(> 60% of total SW)SW Platform layerSW Platform layer(> 60% of total SW)(> 60% of total SW)
Application Platform layer(≅ 10% of total SW)
Application Platform layerApplication Platform layer((≅ ≅ 10% of total SW)10% of total SW)
µControllers Library
OSEKRTOS
OSEKCOM
I/O drivers & handlers(> 20 configurable modules)
Application Programming Interface
Boot Loader
Sys. Config.
Transport
KWP 2000
CCP
ApplicationSpecificSoftware
Speedom
eterT
achometer
Water tem
p.
Speedom
eterT
achometer
Odom
eter---------------
ApplicationLibraries
Nec78kNec78k HC12HC12HC08HC08 H8S26H8S26 MB90MB90
SW Platform Reuse> 70%
of total SW
SW SW Platform Platform ReuseReuse> 70%> 70%
of total SWof total SW
CustomerLibraries
MOSAIC SW Architecture & Components for Automotive Dashboard and Body Control
1010
PlatformsPlatforms
PlatformDesign -Space
Exploration
PlatformSpecification
Architectural Space
Application Space
Application Instance
Platform Instance
SystemPlatform
Page 6
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
1111
PlatformsPlatforms
uuA platform is, in general, an abstraction that covers a A platform is, in general, an abstraction that covers a
number of possible refinements into a lower level. number of possible refinements into a lower level. For For
every platform, there is a view that is used to map the every platform, there is a view that is used to map the
upper layers of abstraction into the platform and a view upper layers of abstraction into the platform and a view
that is used to define the class of lower level abstractions that is used to define the class of lower level abstractions
implied by the platform. implied by the platform.
1212
PlatformsPlatforms
Platform
Mapping Tools
Platform
Platform stack{
Page 7
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
1313
ApplicationApplication example:example:
uuAutomotive PowerAutomotive Power--Train Control Design: from car manufacturer Train Control Design: from car manufacturer
specs to software design to architecture selection to IC specs to software design to architecture selection to IC implementationimplementation
uuProject in collaboration with Cadence, MagnetiProject in collaboration with Cadence, Magneti--Marelli, ST Marelli, ST
Microelectronics, AccentMicroelectronics, Accent
1414
PowerPower--Train Control SystemTrain Control System
uu Electronic device controlling an internal combustion engine and Electronic device controlling an internal combustion engine and a gearboxa gearbox
uu The goalThe goal
ss offer appropriate driving performance (e.g. torque, comfort, safoffer appropriate driving performance (e.g. torque, comfort, safety)ety)
ss minimize fuel consumption and emissions minimize fuel consumption and emissions
uu Relevant characteristics Relevant characteristics
ss strictly coupled with mechanical partsstrictly coupled with mechanical parts
ss hard realhard real--time constraintstime constraints
ss complex algorithms for controlling fuel injection, spark ignitiocomplex algorithms for controlling fuel injection, spark ignition, throttle position, gear n, throttle position, gear shift … shift …
ss 135,000 lines of C code with no comments135,000 lines of C code with no comments
uu First Step was to reFirst Step was to re--design software with methodology to map into different design software with methodology to map into different
hardware platforms with little effort!hardware platforms with little effort!
Page 8
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
1515
System SpecificationsSystem Specifications
1616
Fast Negat iveForce Transient
min f(D, M f u e l)
τ < τm a x
Force Tracking
Fast Posi t iveForce Transient
S p e e d T r a c k i n g
n = n ( . ) n = n ( . )
Idle & T r a s m O n
.
G > 0 | t >τ
. .G < G
B| B = 1
.G < 0 | t > τ
. .G > G
A
G > 0
G = 0
T > 0
T=0 T=0
( G > 0 ) & ( T > 0 )
T=0
f I( n ) = 0 & G = 0
fI(n,G) > 0
.G = 0 & C = 1
.
G > 0|B = 1
T > 0
T = 0
O U T P U T :
n - E n g i n e S p e e dn - E n g i n e S p e e dFF G G - Generated ForceVV GG - Vehicle Speed - Vehicle Speed
n=n(G)
FG
= 0
Rpm Trackingn= argmin(M
fuel)
Idle
FG
= 0
VG
= VG
(.)
Stopn = 0
FG = 0
Startup
FG = 0
n= .
G > 0G = 0
( n < nm i n
) | ( K = O f f ) n > nstartup
K = S t a r t
(n < nmin ) | ( K = O f f )
FG = FG( G , T , n )FG
= FG( G , T , n )
m a x Dτ < τ
m a x
m i n τ
FG
= FG(G,T,n)
M fuel < M m a x ; D> D m i n
T>0 &G = 0
INPUTS:
G - G a s P e d a lG - G a s P e d a lT - Clutch Pedal & Gear StickT - Clutch Pedal & Gear StickB - Brake PedalB - Brake PedalC - Cruise ControlC - Cruise Control
K - K e y
D - Comfort
fI( n ) = 0
& G = 0
Page 9
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
1717
GoalGoal
uuDevelop guaranteed properties control algorithms for all Develop guaranteed properties control algorithms for all
powerpower--train modestrain modes
uuImplement control strategies on embedded controllers Implement control strategies on embedded controllers
“optimally” with respect to production cost, design time, “optimally” with respect to production cost, design time,
reliability, safetyreliability, safety
1818
Model of PowerModel of Power--traintrain
Manifold
(continuous system)
Engine sub-system
Drive-line
(continuous system with changing dynamics)
Throttle opening angle
Spark timing Torque
Manifold pressure
Clutch Insertion/Release
Gear change
Vehicle Speed
Simple?
Page 10
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
1919
CTSCTS
CTSCTS
Engine and DriveEngine and Drive--lineline
2020
Engine and DriveEngine and Drive--lineline
Page 11
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
2121
Single Cylinder Hybrid ModelSingle Cylinder Hybrid Model
2222
uuMeanMean--Value Model: accurate over a longer time Value Model: accurate over a longer time
windowwindow
ss regulation control problemsregulation control problems
ss low performance transient problemslow performance transient problems
uuHybrid Model: cycle accurateHybrid Model: cycle accurate
ss transient control problems transient control problems
ss stability of delaystability of delay--sensitive control algorithmssensitive control algorithms
ss high performance control algorithmshigh performance control algorithms
Hybrid Model vs MeanHybrid Model vs Mean--Value ModelValue Model
Page 12
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
2323
Output DevicesInput devices
Hardware Platform
I O
Hardware
network
DUAL-CORE
RTO
S
BIOS
Device Drivers
Net
wor
k C
omm
unic
atio
n
DUAL-CORE
Architectural Space (Performance)
Application Space (Features)
PlatformsPlatforms
Platform Instance
Application Instances
System Platform(no ISA)
Platform Design Space Exploration
PlatformSpecification
Platform API
Software Platform
Output DevicesInput devices
Hardware Platform
I O
Hardware
network
HITACHI
RTO
S
BIOS
Device Drivers
Net
wor
k C
omm
unic
atio
n
HITACHI
RTO
S
BIOS
Device Drivers
Net
wor
k C
omm
unic
atio
n
Output DevicesInput devices
Hardware Platform
I O
Hardware
network
ST10
RTO
S
BIOS
Device Drivers
Net
wor
k C
omm
unic
atio
n
ST10
Application Software
Application Software
2424
JanusJanus--PARADES 2000 ArchitecturePARADES 2000 Architecture
Interrupt Based Peripherals
ARM7TDMIA
ARM7TDMIB
Peripheral Bus
High Speed Peripheral Bus
ETU CAN SCI
ADC CLB
DMA Based Peripherals
RAMA
RAMB
FLASHA
FLASHB
XBAR
IO s
ub s
yste
m
IRC
Deb
ug U
nit
IRC
PA
RA
DE
S
Page 13
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
2525
The DualThe Dual--Arm ArchitectureArm Architecture
A symmetric dual processor architecture with a highA symmetric dual processor architecture with a high--bandwidth bandwidth interconnection network among processors, memory, and I/O interconnection network among processors, memory, and I/O subsub--systems systems
uu11 Million Tr., 4% more area than single processor solution but 11 Million Tr., 4% more area than single processor solution but twice the performance on applicationtwice the performance on application
uuMost performing architecture for PowerMost performing architecture for Power--train applications, train applications, designed to be redesigned to be re--used over two generations (3used over two generations (3--4 years cycles) 4 years cycles) of system products or moreof system products or more
uuEntirely designed using the methodology in less than 1 3/4 year Entirely designed using the methodology in less than 1 3/4 year from conception (March 99) to first silicon (Jan 01)from conception (March 99) to first silicon (Jan 01)
uu In production by 2002, shipments to car manufacturer 2003In production by 2002, shipments to car manufacturer 2003
2626
DesignDesign “Practice”“Practice”
Page 14
The MARCO/DARPA Gigascale SiliconResearch Center for Design & Test: Form
Executive Advisory Group Review23 September 1999
2727
Design Science: Build upon solid foundationsDesign Science: Build upon solid foundations
2828
Design of Communication Media
Design of Function Blocks
Design of Architecture Components
Metropolis Infrastructure• Model of computation• Design methodology
- Abstraction levels- Refinement
• Base tools- Design imports- Simulation
Metropolis: Synthesis/Refinement• Compile-time scheduling of concurrency
• Communication-driven hardware synthesis• Protocol interface generation• Latency insensitive protocols
Metropolis: Analysis• Static timing analysis of reactive systems• Invariant analysis of sequential programs• Refinement verification• Three-valued simulation• Delay estimation using object code
ETROPOLIS (20+ from Academia (UCB,CMU) and Industry (Intel, ST,BMW, Cadence))