Low energy and HighLow energy and High-performance ...

Low energy and HighLow energy and High performanceperformanceLow energy and HighLow energy and High--performance performance Embedded Systems Design and Embedded Systems Design and R fi bl A hiR fi bl A hiReconfigurable ArchitecturesReconfigurable Architectures

Ass. Professor Dimitrios SoudrisAss. Professor Dimitrios Soudris

Dept. of Electrical and Computer Eng., Dept. of Electrical and Computer Eng., Democritus Univ. of Thrace, GreeceDemocritus Univ. of Thrace, Greece

11

Democritus Univ. of Thrace, GreeceDemocritus Univ. of Thrace, [email protected], [email protected]@ee.duth.gr, [email protected]

ENIAC –The first electronic computer (1946)The first electronic computer (1946)

22

Historical Driving ForcesHistorical Driving Forces

33

Cell Processor for Playstation3

44

Embedded systems definitions Embedded systems definitions HiPEAC Roadmap http:HiPEAC Roadmap http: www hipeac netwww hipeac netHiPEAC Roadmap http: HiPEAC Roadmap http: www.hipeac.netwww.hipeac.net

Embedded systems (ES) = informationEmbedded systems (ES) information processing systems embedded into a larger producta larger product

55

Embedded Systems ApplicationsEmbedded Systems Applications

66

Embedded Systems:Embedded Systems:Memory managementMemory managementMemory managementMemory management

S

PROC

D

SRAM

SR

AM

EXTERNALMEMORY

EXTERNALMEMORY

DP

EmbeddedDRAMMMU

MEMORYMEMORY

P(External Access) = 30 x P(Arithmetic Operations)

77P(Internal Memory) = typ. 40 % - 60 % P(Chip)( ) ( p )

Embedded Systems: Embedded Systems: Dynamic ApplicationsDynamic Applicationsy a c pp cat o sy a c pp cat o s

Graphical ProcessingVirtual Games Platforms

Current multimedia and wireless network applications:Current multimedia and wireless network applications:-- Very complex => designed with highVery complex => designed with high--level languages level languages

Graphical ProcessingVirtual Games Platforms

y p g gy p g g g gg g(e.g. C++) (e.g. C++)

-- Dynamic Memory (DM) required => several sources of Dynamic Memory (DM) required => several sources of unpredictability: user movements? image features?unpredictability: user movements? image features?gg

-- Multiple subMultiple sub--algorithms => different DM access patternsalgorithms => different DM access patternsFinal platforms:Final platforms:-- Portables: limited Resources (e g memory cpu power)Portables: limited Resources (e g memory cpu power)

88

-- Portables: limited Resources (e.g. memory, cpu, power)Portables: limited Resources (e.g. memory, cpu, power)-- Low power and high performanceLow power and high performance

Methodology for Dynamic Data Type Methodology for Dynamic Data Type Refinement ManagementRefinement ManagementRefinement ManagementRefinement Management

Multimedia and Multimedia and NetworkNetworkNetwork Network ApplicationsApplications

Significant Significant improvementsimprovementsimprovements improvements in energy in energy consumption consumption >80%>80%

TradeTrade--offs of offs of memory memory footprint, footprint, p ,p ,performance, performance, energy energy consumed in consumed in DDT DDT

99

Refinement Refinement are possibleare possible

Customized Dynamic Memory Manager: Customized Dynamic Memory Manager: Reduced Power ConsumptionReduced Power ConsumptionReduced Power ConsumptionReduced Power Consumption

SingleSingle--functionedfunctionedSingleSingle--functionedfunctioned–– Executes a single program, repeatedlyExecutes a single program, repeatedly

Ti htlTi htl t i dt i dTightlyTightly--constrainedconstrained–– Low cost, low power, small, fast, etc.Low cost, low power, small, fast, etc.

Reactive and realReactive and real--timetime–– Continually reacts to changes in the Continually reacts to changes in the 84% more

thansystem’s environmentsystem’s environment

–– Must compute certain results in realMust compute certain results in real--time time

AMDREL25% more

than

1010

without delaywithout delay AMDREL

Results for Infenion EasyPortResults for Infenion EasyPortResults for Infenion EasyPortResults for Infenion EasyPort

Lower fragmentation level than Lea 2.7.2Lower fragmentation level than Lea 2.7.2Higher performance than the WinXPHigher performance than the WinXP

1111

Higher performance than the WinXP Higher performance than the WinXP allocator modelallocator model

AMDREL: Architecture of a fullAMDREL: Architecture of a full--custom designed FPGA Hardwarecustom designed FPGA Hardwarecustom designed FPGA Hardwarecustom designed FPGA Hardware

• The “island style”The island style

• Uniform routing channel architecture

• The I/O pads are evenly• The I/O pads are evenly distributed around the perimeter

1212

FPGA layout (8X8 array)FPGA layout (8X8 array)y ( y)y ( y)

Fine-grain Specs8X8-8X8 array

-Area: 5.8 X 6.1mm^2-Up to 330MHz(register-to-register delay)0 18 um CMOS STM-0.18 um CMOS STM

-6 metal layers-Core Power Supply: 1.8V-LSE configuration time: 42nsFull Configuration time: 2 9us-Full Configuration time: 2.9us

-20 tracks on routing channels

I/O PINS96 Data I/Os-96 Data I/Os

-1 Global Reset -1 Global Clock-12-bit Configuration Address Bus

1313

Bus-16-bit Configuration Data Bus- Power/Ground pinsFull-custom design by AUTH,

Subcontractor in AMDREL

Design 3D FPGA architecturesDesign 3D FPGA architecturesDesign 3D FPGA architecturesDesign 3D FPGA architectures

A hit t l lA hit t l lArchitecture level Architecture level exploration of FPGAsexploration of FPGAs

–– Design an interconnectionDesign an interconnection–– Design an interconnection Design an interconnection network based on the network based on the connectivity demandsconnectivity demands

–– Temperature/Power Temperature/Power ––Aware Placement and Aware Placement and RoutingRouting 3D stack:Routing Routing

–– Alternative interconnection Alternative interconnection schemes for 3D viasschemes for 3D vias

3D stack:Reduced system sizeShort interconnectsReduced packaging

1414

Reduced packaging costLower power

Energy RequirementsEnergy RequirementsEnergy RequirementsEnergy Requirements

0,82

0,76

0,78

0,80

lized

en

ergy Energy

by 37%

Energy Savings by 37%

0,70

0,72

0,74

Nor

mal

0,64

0,66

0,68

0,62

,

100% 90% 80% 70% 60% 50% 40% 30% 20% 10%

2 Layers 3 Layers 4 Layers 5 Layers 6 Layers

% fabricated vias

1515

15

2 Layers 3 Layers 4 Layers 5 Layers 6 Layers

7 Layers 8 Layers 9 Layers 10 Layers

CAD Tools for 2D and 3D reconfigurable CAD Tools for 2D and 3D reconfigurable architectures:architectures:MEANDER Design FrameworkMEANDER Design Framework

Application description in HDL

Synthesis

Technology Mapping2D Flow 3D Flow

2D or 3DArchitecture?

EX-VPR

2D P&RPowerModel

2D architecture library

3DPRO

3DPower

3D architecture library

3D P&R

Bitstream generation

2D P&R 3D P&R

1616

Bitstream generation

Available for on-line execution at http://vlsi.ee.duth.gr/amdrel

Current and Future research Current and Future research trendstrends New marketsNew marketstrends trends -- New marketsNew markets

Dynamic multimedia and network Dynamic multimedia and network applicationsapplicationsapplicationsapplications

System Level modeling, exploration and System Level modeling, exploration and ti i titi i tioptimizationoptimization

RTOS RTOS –– realreal--time operating systems. Runtime operating systems. Run--time managementtime management

MultiMulti--core architecturecore architectureNetworkNetwork--onon--chip interconnectionschip interconnectionsNetworkNetwork onon chip interconnectionschip interconnections

⇒⇒ 3D 3D –– integrated circuitsintegrated circuits

1717

VLSI Design and Testing Center ActivitiesActivities

Personnel: 1 Professor, 2 PhDs, 4 G d d 4 M S dGraduate students, 4 M.Sc. studentsIt consists of: (a) Low Power Design Group and (b) Embedded SystemsGroup and (b) Embedded Systems GroupCooperation with leading universities, Cooperation with leading universities, institutes and industryinstitutes and industryinstitutes and industryinstitutes and industry

Funding from National and EUFunding from National and EUFunding from National and EU Funding from National and EU projects. More 20 projects (>1.2 mil. projects. More 20 projects (>1.2 mil. euros, the last 8 years)euros, the last 8 years)New projects in 7New projects in 7thth IST frameworkIST framework

1818

⇒⇒New projects in 7New projects in 7thth IST frameworkIST framework

VLSI Design and Testing Center Activities: AWARDSActivities: AWARDS

Award of 2,000$ VLSI 2005sponsored by Intel and

IBM, Arizona, USAVLSI 2005

1919

4-th Position for AMDREL project

2020

Thank you very much!!!

2121