Presentation is loading. Please wait.

Presentation is loading. Please wait.

Tunable Sensors for Process-Aware Voltage Scaling Tuck-Boon Chan ‡ and Andrew B. Kahng †‡ CSE † and ECE ‡ Departments, UCSD

Similar presentations


Presentation on theme: "Tunable Sensors for Process-Aware Voltage Scaling Tuck-Boon Chan ‡ and Andrew B. Kahng †‡ CSE † and ECE ‡ Departments, UCSD"— Presentation transcript:

1 Tunable Sensors for Process-Aware Voltage Scaling Tuck-Boon Chan ‡ and Andrew B. Kahng †‡ CSE † and ECE ‡ Departments, UCSD 1

2 Outline Intro: Adaptive Voltage Scaling (AVS) Overview of Proposed Method Voltage Scaling Properties Designing the Circuit Results 2

3 Adaptive Voltage Scaling Circuits are designed to guardband for performance variation There is margin for typical chips Adaptive voltage scaling (AVS) adjusts voltage to reduce power 3 Voltage a typical chip worst-case scenario (e.g, due to process variation) Maximum frequency margin reduce voltage  meet performance with less power V nominal

4 Taxonomy of AVS Techniques Open-Loop AVS Closed- Loop AVS Power Freq. & V dd LUT Post-silicon characterization AVS Pre-characterize LUT [Martin02] Process-aware AVS Post-silicon characterization [Tschanz03] Generic monitor Design dependent replica In-situ monitor Process and temperature-aware AVS Generic on-chip monitor [Burd00] Design-dependent monitor [Elgebaly07, Drake08, Chan12] In-situ performance monitor Measure actual critical paths [Hartman06, Fick10] Error Detection System Error detection and correction system V dd scaling until error occurs [Das06,Tschanz10] Error Tolerance AVS approachesAVS classes 4

5 Motivation for Closed-Loop AVS Closed-loop AVS saves up to 62% dynamic power [Hartman06] 5

6 Classes of Closed-Loop AVS Critical path may be difficult to identify (IP from 3rd party) Calibrating monitors at multiple modes/voltages requires long test time Closed- Loop AVS Design-dependent replica In-situ monitor Generic monitor Does not capture design-specific performance variation 6 This work: Tunable monitor for closed-loop AVS Can be applied as a generic monitor Or tuned to capture design-specific performance

7 Outline Intro: Adaptive Voltage Scaling (AVS) Overview of Proposed Method Voltage Scaling Properties Designing the Circuit Results 7

8 Voltage Scaling Key Concepts Process distance: process-induced frequency shift relative to target frequency Scaling rate: frequency shift (  f) per unit voltage difference (  V) V min = Minimum V dd to meet target frequency Calculated from process distance and scaling rate Voltage SS k Process distance Max. freq. Scaling rate = 8

9 Monitor Design Concept Use V min of ring-oscillator (RO) as a reference Design ROs with worst-case voltage scaling properties  an arbitrary circuit will meet target frequency at V min_ro V min of ROsMax. V min of paths 9 > V RO Critical paths Freq. Process corner A RO V Critical paths Freq. Process corner B Max.

10 Proposed Method: Tunable Monitor Our focus is on voltage scaling property  analyze worst-case voltage scaling Store config. Scenario 2: With chips at process corners Extract F max and V min of chips Tune voltage scaling properties of ROs so that V min_ro > V min_chip Recover margin with one calibration Scenario 2: With chips at process corners Extract F max and V min of chips Tune voltage scaling properties of ROs so that V min_ro > V min_chip Recover margin with one calibration Scenario 1: Without circuit information Configure RO for worst-case V min Guardband for arbitrary circuits Scenario 1: Without circuit information Configure RO for worst-case V min Guardband for arbitrary circuits 10

11 Problems Goal: V min_ro > V min_path Questions:  Given a process technology, what is the range of the V min that is defined by process distance and scaling rate for arbitrary critical paths?  What circuit techniques can “tune” V min ? 11 V V min of arbitrary critical paths freq. V min Path BPath A = ? Also, V min changes at different process corners Path C

12 Outline Intro: Adaptive Voltage Scaling (AVS) Overview of Proposed Method Voltage Scaling Properties Designing the Circuit Results 12

13 V min Analytical Derivation 13 f path = inverse of average delays of NMOS & PMOS (1) (2) Calculate delays with Elmore delay model Effective currents of transistors (3) Process distance Scaling rate

14 V min Sensitivity V min is not very sensitive to fanout, interconnect load, etc. Empirically, bounds on V min determined by NMOS and PMOS 14 V min for PMOS only V min for NMOS only

15 Effects of Fanout and Series Resistance Fanout has little effect on V min 15  High series resistance reduces V min  But, need long wires

16 Effects of Cell Type 16 Cell type affects V min Maximum V min at different corners are determined by different cell types Stacking causes cell delay biased to PMOS or NMOS  changes device characteristics and V min

17 Effects of Cell Strength 17 V min does not increase from X1 to X3 But increases from X0 to X1 X1 to X3  {1,2,3} fingers, same device characteristic X0 to X1  Both 1 finger but different diffusion area Cell layout changes device characteristics and V min

18 Outline Intro: Adaptive Voltage Scaling (AVS) Overview of Proposed Method Voltage Scaling Properties Designing the Circuit Results 18

19 Design of RO with Tunable V min Identified two circuit knobs to tune V min Series resistance Cell types (INV, NAND, NOR) Proposed circuit ROs with different cell types (worst-case V min are determined by different cells at different process corners) Tune V min  a configurable series resistance at each stage 1 bit Control pins High resistance Low resistance 19

20 Tunability V min decreases linearly with % high-resistance passgates ROs with different gate types have similar trend INVX3 20

21 Outline Intro: Adaptive Voltage Scaling (AVS) Overview of Proposed Method Voltage Scaling Properties Designing the Circuit Results 21

22 Experiment Methodology 22 Goal: Validate PVS ROs in simulation Check V min of ROs vs. V min of paths with arbitrary circuits and process variation Experiment setup: 65nm industrial technology Implement 3 testcases (arbitrary circuits) Implement 3 tunable ROs (INV, NAND, NOR)

23 Process Variation Setup Simulate critical paths and ROs with SPICE  200 Monte Carlo samples (global variation) 4 variation sources, Gaussian distributions Difference between slow and fast corners define +/- 3 sigma values of variation sources 23

24 V min Extraction and Comparison Define f target of chip and ROs at  “slow-slow” process corner  nominal voltage = 1.0V V min_chip = max. V min of critical paths of a testcase V min_est = max. V min of 3 ROs For each testcase, calculate V min_est - V min_chip of every Monte Carlo sample A chip is safe when V min_est - V min_chip > 0 24

25 Scenario 1: Guardband for Arbitrary Circuit V min_est - V min_chip > 0 under process variation Similar results for different testcases Small difference between normal and tunable ROs  due to series passgates 25 FPU testcase MUL testcase TLU testcase

26 Scenario 2: Tune ROs for Margin Reduction Extract V min_chip at different process corners Configure % high-resistance passgates Ensures V min_est guided by ROs is always safe 26 min. : s.t. :

27 Experiment Result on Tunability 27 Aggressive config.  V min_est < V min_chip  Some chips will fail Optimized config. Increase % high resistance passgates V min_est ≈ V min_chip Default config. Low resistance passgates Guardband for worst-case V min_est > V min_chip 13mV margin

28 Experiment Result on Tunability 28 Aggressive config.  V min_est < V min_chip  Some chips will fail Default config. Low resistance passgates Guardband for worst-case V min_est > V min_chip 13mV margin Optimized config. Increase % high resistance passgates V min_est ≈ V min_chip Benefits of tunability Recover voltage margin Compensate for difference between SPICE model vs. silicon Recover margin when chip performance variation is reduced due to improvements in chip manufacturing

29 Summary Monitor design based on voltage scaling properties Estimate the worst-case voltage scaling property across different process corners  Does not require information about critical paths  Can be used as an IP for arbitrary circuits Recover margin if f max of sample silicon is available Future works  Proof of concept silicon  Account for performance variation due to layout context 29

30 30 Thank you!

31 Backup Slides 31

32 Effects of Pass Gates Pass gate is equivalent to large resistance V min decreases with fewer parallel pass gates 32 V min decreases

33 Effects of Cell Type and Strength Key observations: V min is affected by cell types  Use NAND, NOR type ROs Cell strength changes V min  Use cells with large V min 33


Download ppt "Tunable Sensors for Process-Aware Voltage Scaling Tuck-Boon Chan ‡ and Andrew B. Kahng †‡ CSE † and ECE ‡ Departments, UCSD"

Similar presentations


Ads by Google