![Hierarchical view of Energy Conservation
Software Dynamic system monitoring and intelligent control of energy
Definitions savings, work load profiling, [dvfs], profiling and partitioning
Platform Power Trees/Voltage islands, Connectivity of components & consistent
platform power modes, intelligent bus coding, dependency
Definitions discovery/optimization
Architectural Heterogeneous processing resource optimization: MCU, DSP,
Definitions accelerators, functional processing units, memory usage
optimization
Design Hardware support for voltage islands, power gating, low-
Definitions power idle modes, SRPG, AWB, DVFS, DPTC, clock gating
PROCESS node Transistor design, Vt Optimization, memory bitcell design.
Definitions Special circuits, libraries, custom and analog blocks, SOI
Thanks to Milind Padhye, Freescale Semiconductor, Austin Wireless Design Center
Slide 17](https://image.slidesharecdn.com/low-powerdesignver26mar08-130422123922-phpapp02/85/Low-Power-Design-and-Verification-16-320.jpg)
![CPF Language
● CPF is TCL-based.
● CPF Language = TCL commands + CPF objects + Design objects
Power domain
Analysis view
Delay corner
Library set
Operating condition
● Design objects: objects that already exist in the RTL/gate netlist
Module, Instance, Net, Pin, Port
● Commands – 42 commands
set_* commands [version, scope, and general commands]
define_*_cell commands [library cell description]
create_*_rule commands [design intent]
update_*_rules commands [implementation directives]
Innovation Through Collaboration
Innovation Through Collaboration – Low Power Coalition – 11 –](https://image.slidesharecdn.com/low-powerdesignver26mar08-130422123922-phpapp02/85/Low-Power-Design-and-Verification-35-320.jpg)
![Specify Power Mode Transitions
PM2
PDcore PDau PDlu PDalu PDrf
PM1 1.2v 1.2v 1.2v 1.2v 1.2v PM1 PM3
PM2 0.8v off 1.2v 1.2v 1.2v
PM3 0.8v off off off 1.2
PM4
PM4 0.8v 1.2v 1.2v 1.2v off
create_mode_transition -name PM1toPM2 –from_mode PM1 –to_mode PM2
-start_condition { pcu_inst/ctrl[0] & pcu_inst/ctrl[1] }
-clock_pin { pcu_inst/clk } –cycles 100
create_mode_transition -name PM2toPM3 –from_mode PM2 –to_mode PM3
-start_condition { pcu_inst/ctrl[0] & !pcu_inst/ctrl[1] }
-clock_pin { pcu_inst/clk } –cycles 1000
create_mode_transition -name PM3toPM4 –from_mode PM2 –to_mode PM3
-start_condition { !pcu_inst/ctrl[0] & pcu_inst/ctrl[1] }
-clock_pin { pcu_inst/clk } –cycles 1000
create_mode_transition -name PM4toPM1 –from_mode PM2 –to_mode PM3
-start_condition { !pcu_inst/ctrl[0] & !pcu_inst/ctrl[1] }
-clock_pin { pcu_inst/clk } –cycles 200
Innovation Through Collaboration
Innovation Through Collaboration – Low Power Coalition – 24 –](https://image.slidesharecdn.com/low-powerdesignver26mar08-130422123922-phpapp02/85/Low-Power-Design-and-Verification-37-320.jpg)
Low Power Design and Verification
- 1. Low Power Design And Verification Steven E. Schulz President and CEO March 26th, 2008 DVclub San Jose Si2 - Innovation Through Collaboration
- 2. Today’s Agenda Si2 – Innovation Through Collaboration 2
- 3. Today’s Agenda • Why Low-Power Now? • Design and Verification Flow Challenges / Reqts • Common Power Format Introduction / Examples • Industry / Market Adoption and Silicon Benefits • Introduction to the Low Power Coalition • 2008 Roadmap / Plans •Q&A Si2 – Innovation Through Collaboration 3
- 4. 2001 International Technology Roadmap for Semiconductors Possible kink in dynamic power Dynamic power Sub- threshold Here We are Gate Junction Junction Leakage P = ACV2f + VIleak Page 3
- 5. Power Motivation and Requirements Mobile applications trends: • Leakage is significantly increasing due to process scaling • Active power increases due to application integration (with the subsequent exponential increase in leakage). Current density is also on the increase. • Active leakage is now a significant portion of SoC active power budget. • Sleep mode techniques need to be enhanced and enabled in a consistent fashion throughout the design flow • We need a concerted effort applied to leakage minimization at the micro- architectural, system and software level. • Process variation now limits how much we can voltage scale and how we do our power accounting, and therefore new strategies need to be develop to capture these constraints, and enhance our current scaling approaches/methodologies. Page 2
- 6. Ultra-Low-Power Applications Addressing a Wide Range of Requirements Slide 18 © 2006 Virage Logic Corporation – COMPANY CONFIDENTIAL
- 7. Processor design for Power Efficiency: Different needs for different markets • Server market: – Defining property: Server processors are rarely idle. – Power goal: Increase MIPS/Watt in Power State C0 (ACPI). • Mobile market: – Defining property: Laptop processors are mostly idle. – Power goal: Reduce power in C2/C3 power states. • Techniques: – Clock gating – Multiple power domains – Multiple threshold voltages – Headers/footers – Operand Isolation (holding cell inputs stable when output is unused) – Dynamic voltage and frequency scaling – And others… 3 October 5, 2006
- 8. PowerSmart™ -- Low Power Design Methodology Integrated 1.5V OP buffer Power switch clock gated cell en 1.5V OP clk o 3.3V 1.2V Regulator 1.0V 0.8V 1.0V 1.2V Multi-output Isolation cell regulator Vdd Vdd_UPS Retention D Q flip-flop ck Multi-Vt Multi-VDD library Multi-VDD Level shifter library Level shifter Power Islands SRAM Clock Gating 0.8V Low power Multi-Vth (Fusion) IP Low Power IP Excel Your Idea to Silicon 10
- 9. Power analysis challenges: More complex than timing analysis • It is pattern-dependent. – Circuit and gate-level power analysis require good RTL-level patterns for accurate results. • It is a balancing act. (power efficiency) – Performance per watt (efficiency) is the metric, not Watts. Need to find blocks or nets that consume power without appropriate performance benefit. – Many tools sort blocks and nets by total power consumption not performance/watt. (E.g. clock nets burn a lot of power, but we already knew that) • It is an aggregate (time and space) and a user-defined constraint. – Power analysis types: average power (for budgeting & package selection), energy (for battery life), peak power (IR drop analysis), etc. – E.g. Briefly higher localized power consumption can be tolerated for package selection, unless it exceeds limits. • It requires coordination of data from physical design, gate design, RTL, and verification domains. – It requires knowledge in all these domains to cross-check results. • Must allow for accuracy to be improved over time. – Detailed circuit-level power analysis data often comes too late in the design cycle. 4 October 5, 2006
- 10. Other related issues: DFT and Timing • Are scan paths hooked up in the RTL? Are they simulated in the Verilog? How are they verified? • How do you analyze power consumption in scan mode? • Timing also needs to know about the multiple voltage domains and operating points. • Need to work on timing and power in one environment to achieve correct optimization and trade-offs. 11 October 5, 2006
- 11. Current state … • Debugging capabilities are very poor • Capacity issues • Complexity issues • Reporting weak and misleading • Functional correctness difficult to verify • Tools are mostly in Gatelevel, should be in RTL • Important is to have accuracy for RTL or otherwise it is not useful • All tools using different description for PM • PM configurations currently having thousands of statements in SoC level • No automation; It is designers responsibility to verify that all definitions are done correctly • Because updates for these definitions are done quite seldom, it is difficult to keep in mind complex configurations • There is no automation for PM definitions verification • Design hierarchy presentation varies in configuration files between tools, also between RTL and gate in same tool. Syntax is effected by scripting languages like perl and tcl 4 © NOKIA UPF Workshop / Oct 2006 /Naula
- 12. Flow and Methodology Requirements The verification flows need to enable: • a voltage aware simulation method for logic problems due to voltage island partitioning • a method for full design multi-voltage domain analysis and reporting • a vector-less rule driven analysis of architecture, RTL, and gate correctness • a method for equivalence checking (i.e. across voltage states ) • a method that captures Island ordering • a method that incorporates early detection of micro-architecture sequence errors Page 5
- 13. System Aspects require differing views Layout Algorithm Bus Architecture Implementation Power Source Temperature Refinement Security Address Space Documentation Slide 8
- 14. Address space Layout Algorithm Bus Architecture Implementation Power Source Temperature Refinement Security Address Space Documentation Slide 9
- 15. Aspect View: Bus Architectural layout Layout Algorithm Bus Architecture CPU CPU DMA DMA FLASH FLASH Implementation PLL PLL Power Processor Bus Source Temperature Refinement Bridge Bridge DMA DMA SRAM SRAM Security Peripheral Bus Address Space Documentation … … Ethernet Ethernet USB USB UART UART GPIO GPIO Slide 10
- 16. Hierarchical view of Energy Conservation Software Dynamic system monitoring and intelligent control of energy Definitions savings, work load profiling, [dvfs], profiling and partitioning Platform Power Trees/Voltage islands, Connectivity of components & consistent platform power modes, intelligent bus coding, dependency Definitions discovery/optimization Architectural Heterogeneous processing resource optimization: MCU, DSP, Definitions accelerators, functional processing units, memory usage optimization Design Hardware support for voltage islands, power gating, low- Definitions power idle modes, SRPG, AWB, DVFS, DPTC, clock gating PROCESS node Transistor design, Vt Optimization, memory bitcell design. Definitions Special circuits, libraries, custom and analog blocks, SOI Thanks to Milind Padhye, Freescale Semiconductor, Austin Wireless Design Center Slide 17
- 17. Low Power Design Needs Support Low Power Design Techniques thru the entire design flow using a single file format. Design Representation – Accurately define and capture the low power design intent, modes and constraints. Design Implementation – Floorplan and power grids. – Common constraints for all tools (Synthesis, APR, timing, DFT) – Design analysis tools with single power constraints. – Accurate power estimation and measurements Design Verification – Voltage oriented simulators – Various static power technique modeling and simulations. – Silicon validation and correlation. Slide 18
- 18. Low power implementation : What’s new ? Becoming mainstream: – For 65nm and below , Low power is crucial for low/high performance. So far: – For dynamic power • Reducing power dissipation source when not needed. • Minimize switching capacitances. – For static power • Use of multiple Vt(s) synthesis / optimization More recently: – Reducing supply reduces power, but also makes circuit slower. To meet both chip performance requirements and power goals, use voltage islands and voltage and frequency scaling. – Leakage can also be addressed by suppressing current when not needed. Island of voltages increases the difficulty on implementation techniques. Intrusive on functionality Impact across design tasks ( Design-In and Implementation ) 3 Si2-Accelera Low Power Workshop, CTO/SoCDT, Herve Menager, October 5th, 2006
- 19. Design implementation challenges New cells and their use model – Level Shifters – Retention logic – Isolation logic – Micro Switches Impacts at all levels of the design flow – Interface logic design, partitioning – Verification of power modes – Checks on interfaces between Power domains – Placement of IP in context voltage islands – Floorplanning with switches, Irdrop across switches, transient behavior. – DFT – Verification (STA, LVS, analysis) Conceptual shift : Power nets become functional signals 4 Si2-Accelera Low Power Workshop, CTO/SoCDT, Herve Menager, October 5th, 2006
- 20. Methodology and design flow impacted 6 Si2-Accelera Low Power Workshop, CTO/SoCDT, Herve Menager, October 5th, 2006
- 21. Short Term need ( 2) – Fill hole in Verification Low leakage design techniques have created a real paradigm shift. Power and ground nets are now becoming functional nets. They are not all explicitly in RTL or netlist levels. Proper connection of any other functional nets is verified by functional simulation….against the RTL or netlist. Being able to verify the power down modes , retention, recovery at power-on, etc in the context of RTL simulation is becoming mandatory. Verification tools should be power modes aware. 10 Si2-Accelera Low Power Workshop, CTO/SoCDT, Herve Menager, October 5th, 2006
- 22. For IP, context is key “Always-on” depends on Memory context Processor Buffer within CPU SoC buffer routed across CPU SoC Characterization range is important Cells, memory could be different Complex features Multiple VDD, VSS pins Multiple operating Don’t want formats limiting IP features voltages Voltage dependent behavior Closed-loop behavior (tunable voltage) 6
- 23. Canonical design to argue over……. Start with a realistic example to exercise interfaces and control Power and Ground are signals – but not as we know them…….. Power Gating, Retention, (Dynamic) Voltage Scaling, Level shifters, Memory… Isolation clamps across boundaries, a number of supply voltages e.g. a SOC with always powered logic plus: VCPU VRAM VSOC RAM RETAIN RETAIN with Core Retention PG CPU (& additional sub- power rail?) SRPG DVFS system subsystem (& LV retention?) 2
- 24. Addressing power management challenges Operational and Standby (leakage) Active power + leakage Power gating/voltage scaling On-chip – fast but with care to avoid dI/dt problems Off-chip – may add latencies as long as 100’s of microseconds Need to be able to quantify Real-time cost (e.g. interrupt latency) in “wake-up” times Energy cost functions getting into/returning from power saving states 14
- 25. Legacy Core VDD Legacy Core VSS Slide 6 © 2006 Virage Logic Corporation – COMPANY CONFIDENTIAL
- 26. Traditional Design Flow Architecture Placement & Timing Clock Design Optimization Verification Synthesis HW/SW Co- Floorplanning Optimization Design Logic Routing & RTL Design Synthesis Verification Optimization Sign-off: DRC/LVS Slide 7 © 2006 Virage Logic Corporation – COMPANY CONFIDENTIAL
- 27. Techniques Relevant To IP Right size libraries – Smaller transistors lead to smaller parasitics – Performance trade-off Multi-Vt libraries – Right Vt for the right paths at the right performance – Effectively used to control leakage – Increases the number of libraries needed to implement the design Voltage Islands – Requires updates to deal with multiple power supplies and associated conditions – Requires special level shifting components to implement Power Gating/On-Chip Regulation – Requires special power gating cells/regulation cells – Need to deal with “derived” power nets – Need to deal with POR cycle Substrate Bias – Requires dealing with multiple power supplies and possibly “negative” power supplies – Requires special level shifting components to implement Slide 9 © 2006 Virage Logic Corporation – COMPANY CONFIDENTIAL
- 28. So What Changes? … Everything … VDD VDD Voltage Island VSS Voltage Island & VDD1 Support State Retention VDD Support VSS Low Power Core On Chip VSS Regulation Support Voltage Island & VDD2 Back Bias Support VSS VSS1 Slide 10 © 2006 Virage Logic Corporation – COMPANY CONFIDENTIAL
- 29. So What Changes? … Everything … Architecture Placement & Timing Clock Design Optimization Verification Synthesis HW/SW Co- Floorplanning Optimization Design Logic Routing & RTL Design Synthesis Verification Optimization Sign-off: DRC/LVS Power Sign-off Spec Slide 12 © 2006 Virage Logic Corporation – COMPANY CONFIDENTIAL
- 30. PowerSmart™ -- Low Power Design Methodology Integrated 1.5V OP buffer Power switch clock gated cell en 1.5V OP clk o 3.3V 1.2V Regulator 1.0V 0.8V 1.0V 1.2V Multi-output Isolation cell regulator Vdd Vdd_UPS Retention D Q flip-flop ck Multi-Vt Multi-VDD library Multi-VDD Level shifter library Level shifter Power Islands SRAM Clock Gating 0.8V Low power Multi-Vth (Fusion) IP Low Power IP Excel Your Idea to Silicon 10
- 31. Low Power Design Without A Power Format Command file Specification MSV •Domains Function, timing, power SRPG PSO How do you verify ? •Level shifters •Isolation DVFS power functionality •SRPG without changing RTL? Command file Design Creation Iterate RTL Iterate Verification •Domains Constraint Coding Formal Testbench Automation •Level shifters Generation Analysis Verification Coverage Equivalence Checking Constraint Validation •Isolation •SRPG Synthesis Simulation Command file Design for Test •Domains Acceleration SVP & Emulation Command file •Domains •Level shifters Constraints Netlist •Isolation •SRPG Command file •Domains Physical Implementation •Level shifters Equivalence LVS/DRC/Ext Chip Integration •Isolation checking Prototyping Constraint Validation •SRPG Command file DFT •Domains Which one of Physical Synthesis •Modes for ATPG Analysis these is “golden”? Command file •Domains •Level shifters Routing Does the power ATPG •Isolation •SRPG shutoff really going to Sign-off work? Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition –3– GDSII
- 32. What Was the Problem? Logic is “Connected” Power is Not “Connected” Formal Formal Simulation Hardware Simulation Hardware Analysis Analysis Parser Parser Parser Parser Parser Parser Management Synthesis Management Synthesis Parser Parser Parser Parser Logic Equivalence Power Power Equivalence Parser Parser SVP Information Checking SVP Information Information Checking (Verilog) (CPF) (no consistency) Parser Parser Parser Parser Parser Parser P+R Test Test P+R IP Libraries IP Libraries Can be Automated Very Difficult to Automate Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition –4–
- 33. Si2 CPF Standardization ● Dec 4, 2006 Cadence contributed CPF v1.0 to Si2 ● January 12, 2007 LPC members unanimously voted and approved CPF v1.0 as Si2 Specification for low power standard ● January 17, 2007 Cadence contributed CPF v1.0 parser source code to Si2 ● March 5, 2007 CPF 1.0 available to everyone at no cost as a Si2 standard Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition –8–
- 34. Common Power File ASCII file to capture ● Design intent and constraints Power domain Logical: instances as domain members Physical: power/ground nets and connectivity Analysis view: timing library sets for power domains Power Logic Level Shifter Logic Isolation Logic State-Retention logic Switch Logic & Control Signals Power mode Mode definitions Mode transition definitions ● Technology information Level Shifter Cells, Isolation Cells, State-Retention Cells, Switch Cells, Always On Cells Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 10 –
- 35. CPF Language ● CPF is TCL-based. ● CPF Language = TCL commands + CPF objects + Design objects Power domain Analysis view Delay corner Library set Operating condition ● Design objects: objects that already exist in the RTL/gate netlist Module, Instance, Net, Pin, Port ● Commands – 42 commands set_* commands [version, scope, and general commands] define_*_cell commands [library cell description] create_*_rule commands [design intent] update_*_rules commands [implementation directives] Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 11 –
- 36. Minimal Command Set For Different Design Stages create_power_domain Specify power intents create_nominal_condition verification and simulation create_power_mode design exploration early power estimation create_state_retention_rule create_isolation_rule create_level_shifter_rule More implementation details define_library_set synthesis update_nominal_condition formal verification update_power_mode DFT, ATPG, gate level power estimation create_ground_nets create_power_nets Complete physical implementation update_power_domain details silicon virtual prototyping create_power_switch_rule power planning create_analysis_view physical synthesis create_operating_corner structural verification sign-off power analysis Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 12 –
- 37. Specify Power Mode Transitions PM2 PDcore PDau PDlu PDalu PDrf PM1 1.2v 1.2v 1.2v 1.2v 1.2v PM1 PM3 PM2 0.8v off 1.2v 1.2v 1.2v PM3 0.8v off off off 1.2 PM4 PM4 0.8v 1.2v 1.2v 1.2v off create_mode_transition -name PM1toPM2 –from_mode PM1 –to_mode PM2 -start_condition { pcu_inst/ctrl[0] & pcu_inst/ctrl[1] } -clock_pin { pcu_inst/clk } –cycles 100 create_mode_transition -name PM2toPM3 –from_mode PM2 –to_mode PM3 -start_condition { pcu_inst/ctrl[0] & !pcu_inst/ctrl[1] } -clock_pin { pcu_inst/clk } –cycles 1000 create_mode_transition -name PM3toPM4 –from_mode PM2 –to_mode PM3 -start_condition { !pcu_inst/ctrl[0] & pcu_inst/ctrl[1] } -clock_pin { pcu_inst/clk } –cycles 1000 create_mode_transition -name PM4toPM1 –from_mode PM2 –to_mode PM3 -start_condition { !pcu_inst/ctrl[0] & !pcu_inst/ctrl[1] } -clock_pin { pcu_inst/clk } –cycles 200 Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 24 –
- 38. Low Power Design Verification Using CPF ● No need to specify power or ground nets at RTL stage ● No need to specify implementation related constraints at this stage such as library, timing constraints etc ● Minimal set of CPF commands for front-end designers to use Simulation tools to simulation power domain on and off to simulate power mode transitions for DVFS Coverage tools to check power mode coverage to check power mode transition coverage Assertion tools to generate power domain and mode aware assertions Verification tools to check for the correctness and completeness of CPF Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 27 –
- 39. Low Power Logic Implementation and Verification Using CPF ● Still, no need to specify power or ground nets at this design stage ● Minimal set of CPF commands for designers to use Logic synthesis tools to synthesize isolation, level shifter and state retention logic to perform power domain aware logic synthesis to perform power mode aware (DVFS) synthesis Test synthesis tools to perform power domain and power mode aware DFT synthesis to generate power domain aware test control logic Formal Verification tools to check the correctness of low power structural implemented by synthesis tools to perform low power equivalency checking (RTL+CPF vs Netlist) Simulation tools to perform power aware gate level simulation to generate additional assertions for gate level simulation Analysis tools to perform power domain aware and power mode aware power analysis Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 33 –
- 40. CPF Enabled Low Power Design Flow Specification Re-use pre- Function, timing, power Quick architectural exploration Instantiate single RTL with different power profiles verified IP ? Functionally verify advanced power Design Creation Iterate RTL + CPF Iterate Verification RTL implementation b Constraint Coding Formal techniques Analysis Testbench Automation Verification Coverage Generation Equivalence Checking Constraint Validation Structural & Synthesis Funct. Checks Design for Test Simulation Acceleration SVP & Emulation Constraints CPF Netlist Hand off to drive physical implementation Golden Physical Implementation specification Equivalence LVS/DRC/Ext Chip Integration checking eliminates Prototyping Automatic partitioning of Constraint Validation DFT assumptions and power domains Physical Synthesis Analysis miscommunications Single power Routing Automatic scheduling of test modes ATPG specification used from Sign-off specification to GDSII Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 45 – GDSII
- 41. Continued Industry Wide Adoption of CPF 1Q2007 2Q2007 2H2007 Reference Flow 8.0 EnergyPro Technology Joins PFI PRIDE Flow Joins PFI Common Platform Joins PFI • CPF becomes Si2 standard Flow • > 100 customer adopting • Cadence Low Joins PFI CPF-based advanced low Power Solution power solution production PowerPro CG released V 1.0 • ~ 50 tapeouts DDR PHY Freescale, Fujitsu, NEC, NXP.. Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 46 –
- 42. Ecosystem Support for CPF Based Low Power Solution Early Adopters Foundry IP Vendor ASIC / Design Service EDA www.powerforward.org Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 47 –
- 43. TSMC 8.0 Low Power Reference Flow CPF Quality Check CPF Quality Check Conformal Low Power Conformal Low Power CPF-Enabled Functional simulation CPF-Enabled Functional simulation Incisive Design Team Simulator Incisive Design Team Simulator Incisive Design Team Manager Incisive Design Team Manager CPF-Enabled Logic Synthesis & DFT CPF-Enabled Logic Synthesis & DFT Encounter RTL Compiler Encounter RTL Compiler CPF-Enabled LEC + Power Checks CPF-Enabled LEC + Power Checks Conformal Low Power Conformal Low Power CPF-Enabled Logic simulation CPF-Enabled Logic simulation Incisive Design Team Simulator Incisive Design Team Simulator CPF CPF CPF-Enabled Physical implementation CPF-Enabled Physical implementation SoC Encounter SoC Encounter CPF-Enabled LEC + Power Checks CPF-Enabled LEC + Power Checks Conformal Low Power Conformal Low Power CPF-Enabled ATPG CPF-Enabled ATPG Encounter Test Encounter Test CPF-Enabled Timing & SI signoff CPF-Enabled Timing & SI signoff Encounter Timing System Encounter Timing System CPF-Enabled Leakage & Thermal Analysis CPF-Enabled Leakage & Thermal Analysis Encounter Timing System Encounter Timing System CPF-Enabled IR drop & Power signoff CPF-Enabled IR drop & Power signoff VoltageStorm-DG www.tsmc.com VoltageStorm-DG Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 48 – 48
- 44. ARC Proof Point Project Using CPF Based Low Power Solution ARC700 with SIMD Co-Processor I$ ARC700SCQ SCQ SCM SDM D$ SIMD I$ D$ SCQ SCQ SCM SCQ SDM Power Forward Always On low-power implementation & Functional Blocks Power Domains verification project results Clock Gating Domains ● Simulation with CPF identifies • For high bit-rate data streams, both problems that you will not otherwise identify the ARC and the SIMD run flat out ● CPF aids communication of power • For lower bit-rate data stream, the intent across team boundaries, subsystem can be run at a lower ensuring accurate implementation frequency at all flow stages ● Significant power savings results • For generic processing, the SIMD using these techniques can be inactive Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 49 – 49
- 45. Fujitsu Proof Point Project Using CPF Based Low Power Solution 90nm 940K instances 11 Power Domains 19 Power Modes DVFS ● Verified with test design PSO functional verification with simulation CPU1 peripherals CPU2 Low power structural and physical check (Shifters/Isolators/Power switches) Domain aware place and route ● Conclusion Functional verification is necessary for complex PSO design for design bugs Power Switch Power Structural check with CPF could verify LP Domains design Fujitsu will support CPF-based ASIC flow for their customers Silicon Proven September ‘07 Silicon Proven September ‘07 Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 50 – 50
- 46. NEC Proof Point Project Using CPF Based Low Power Solution 65nm 6 Power Domains NEC Electronics 5 Power Modes Corporation 2 Supply Voltage PD0: 1.2V Driver PD1:1.2V (Default, Always On) PD4:0.74V PD2:1.2V PSOcntl PD3:0.74V Validated CPF and CPF-based flow PSGcntl for major low power methodologies PD5:0.74V in NEC Electronics ISOcntl 386 checkpoints evaluated successfully CPF describe-ability Power Power Domain Mode Multi-Supply-Voltage (MSV) PD0 PD1 PD2 PD3 PD4 PD5 Power Shut Off (PSO) PM1 1.2V 1.2V 1.2V 0.74V 0.74V 0.74V State Retention Logic (SRL) PM2 1.2V PSO 1.2V 0.74V 0.74V 0.74V Variable Voltage Library (VVL) PM3 1.2V 1.2V PSO 0.74V 0.74V 0.74V Clock Tree Gating (CTG) PM4 1.2V 1.2V 1.2V PSO 0.74V 0.74V CPF based flow will be in use from Q3/2007 PM5 1.2V PSO PSO PSO 0.74V PSO Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 51 – 51
- 47. NXP Proof Point Project Using CPF Based Low Power Solution Power Forward low-power platform SoC results ● CPF-based functional • SoC consists of 11 islands verification (using simulation) • 3 major power consumers -RISC catches system level power CPU, VLIW DSP & L2 System issues early in the flow Cache are controlled using DVFS ● Use of CPF ensured what • High bandwidth expansion ports implementation built was what enable extension, with graphics was verified or cellular modem subsystems Innovation Through Collaboration Innovation Through Collaboration – Low Power Coalition – 52 – 52
- 48. Low-Power Methodology A Practical Case Study in Low-Power Methodology • Special thanks to NXP Semiconductors
- 49. Low-Power Methodology • New! Educational eBook resource for industry… no-cost download (168pp.) • Contents include: – LP Techniques - Design / Verification / Implementation w/ CPF – Real end-user chip design experiences courtesy, PowerForward.org Si2 – Innovation Through Collaboration
- 50. What Is The Low-Power Coalition? ● Flow-based solutions Standards to promote integration of open technologies into cohesive flows CPF contributed to LPC 4Q'06, approved as new Si2 standard in Mar'07 Analyze / develop semantic consistency across data exchanges ● User-centric and comprehensive Focused on user needs for faster adoption into production chip design flows Owns the industry's low-power roadmap of requirements Comprehensive: enabling software, training & educational materials, articles, books, conferences, press coverage, etc.
- 52. Recent LPC / CPF Publicity
- 53. Recent LPC / CPF Publicity
- 54. LPC Structure Full LPC Membership Technical Steering Group 3 Chief Architects Data Model Flow WG Format & API WG Req'ts WG Format Comparison WG
- 55. LPC Structure, Working Groups ● Full LPC membership (AMD, Chair) Business/policy & standards approvals ● Technical Steering Group (TSG): charters working groups, owns the low-power technology roadmap Includes 3 Chief Architects (Cadence, IBM, LSI) ● Active and completed working groups: Format Comparison WG – report on technical comparison of CPF and UPF (Done, results widely shared) Flow WG – align on low-power reference design flow and design techniques to drive clarity for enhancements Data Model and API WG – map clear semantics and data relationships in CPF, add API interface support to CPF Format Requirements WG – define priorities and detailed requirements for upcoming revision of CPF
- 56. LPC Working Groups • Flow Working Group (AMD, chair) – Definition of complete reference flow from ESL to GDSII • Target completion of ESL portion: 1Q08 – Analysis of power stimuli for SoC power estimation • Target completion date: 1H08 – Compilation of all known low power design techniques • Target completion date: 1H08
- 57. LPC Working Groups • Format Working Group (Freescale, chair) – Format extensions requirements document • DONE – Request for Technology (RFT) based on requirements document • RFT contributions for v1.1 have arrived, work begun – CPF v1.1 standardization target: mid-2008 • > 100 pp. of enhancements expected – Low-Power Roadmap through end 2009 • Coordinated among TSG, Flow, & Data Model WGs • DONE – approved for public release Feb08
- 58. Format Enhancement Requirements Immediate - Requirements for Version 1.1 (extension to 1.0) Hierarchical flow Support. Memory modeling styles and support. Gatelevel verification Flow CPF support. 1.1 Power estimation Support Clocking and related updates are required to drive power optimization. Medium Term – Requirement for CPF 1.2 Pre-Si and post_Si power modeling and budgeting. Test power definitions not represented in CPF. 1.2 Investigate Load_foreign. IO modeling and representation. Long term - Next Generation CPF needs to drive debug related to power. CPF based system level definition. 2.0 Innovation Through Collaboration – Low Power Coalition Innovation Through Collaboration –1–
- 59. Hierarchical Flow – Bottom Up Chip Level power Structure CPF Hard IP with Hard IP with Soft IP Memories RTL (Behavioural) RTL or Gates Multiple CPF Multiple CPF Multiple CPF Multiple CPF Innovation Through Collaboration – Low Power Coalition Innovation Through Collaboration –3–
- 60. Hierarchical Flow Requirements ● Retain integrity of power structures appropriately. Provide a way of overriding the IP level constraints by Chip integrators. Provide a way where IP providers can force protection. ● Merging of multiple IP level power domains. ● Integration of all power control signals for the merged domains. Switch Enable Isolation Retention Control ● Resolution of power rule conflicts when integrating IPs & merging power domains. ● Protection of variables at different level of hierarchy. ● Support parameters defined in different IP blocks ● Optionally preserve the boundary port definition of block level CPF. ● Merging of power mode definitions. Consolidation of block level power modes at top level Reference of block level power mode at top level. ● Enable power mode specific constraints. (decide which slides) Innovation Through Collaboration – Low Power Coalition Innovation Through Collaboration –6–
- 61. Summary of Low-Power Collateral 1. CPF v1.0 Standard 2. CPF v1.0 Parser software 3. CPF Pocket Reference Guide 4. CPF On-line Tutorial (2.5 hrs) 5. Low-Power Glossary of Terms 6. CPF v1.1 Requirements Document (and RFT) 7. Low-Power Industry Roadmap 8. “A Practical Guide to Low Power Design” (168pp. ebook) For downloads, go to: www.si2.org/?page=726 Si2 – Innovation Through Collaboration
- 62. Si2 – Innovation Through Collaboration
- 63. Thank You! Si2 - Innovation Through Collaboration