1 / 28

Vishwani D. Agrawal James J. Danaher Professor Department of Electrical and Computer Engineering

ELEC 5970-003/6970-003 (Fall 2004) Advanced Topics in Electrical Engineering Designing VLSI for Low-Power and Self-Test Estimating Power Consumption. Vishwani D. Agrawal James J. Danaher Professor Department of Electrical and Computer Engineering Auburn University

kaipo
Télécharger la présentation

Vishwani D. Agrawal James J. Danaher Professor Department of Electrical and Computer Engineering

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. ELEC 5970-003/6970-003 (Fall 2004)Advanced Topics in Electrical EngineeringDesigning VLSI for Low-Power and Self-TestEstimating Power Consumption Vishwani D. Agrawal James J. Danaher Professor Department of Electrical and Computer Engineering Auburn University http://www.eng.auburn.edu/~vagrawal vagrawal@eng.auburn.edu ELEC 5970-003/6970-003

  2. Power Estimation Techniques • Logic simulation • Circuit-level simulation • Probabilistic estimation • Peak power estimation • Power estimation for a high-level design ELEC 5970-003/6970-003

  3. Ca Logic Model of a CMOS Circuit VDD pMOS FETs a Da c Dc a b Db c Cc b Daand Dbare interconnect or propagation delays Dcis inertial delay of gate Cb nMOS FETs Ca , Cb and Cc are parasitic capacitances ELEC 5970-003/6970-003

  4. Options for Inertial Delay(simulation of a NAND gate) Transient region a Inputs b c (CMOS) c (zero delay) c (unit delay) Logic simulation X rise=5, fall=5 c (multiple delay) Unknown (X) c (minmax delay) min =2, max =5 Time units 5 0 ELEC 5970-003/6970-003

  5. Signal States • Two-states (0, 1) can be used for purely combinational logic with zero-delay. • Three-states (0, 1, X) are essential for timing hazards and for sequential logic initialization. • Four-states (0, 1, X, Z) are essential for MOS devices. See example below. • Analog signals are used for exact timing of digital logic and for analog circuits. Z (hold previous value) 0 0 ELEC 5970-003/6970-003

  6. True-Value Simulation Algorithm • Event-driven simulation • Only gates or modules with input events are evaluated (event means a signal change) • Gate and interconnect delays are used to determine the transients at gate outputs • Per-vector complexity of computation is linear in number of gates × total input to output time delay units ELEC 5970-003/6970-003

  7. Event-Driven Algorithm Example Scheduled events c = 0 d = 1, e = 0 g = 0 f = 1 g = 1 Activity list d, e f, g g a =1 e =1 t = 0 1 2 3 4 5 6 7 8 2 c =1→0 g =1 2 2 d = 0 4 f =0 b =1 Time stack g 8 4 0 Time, t ELEC 5970-003/6970-003

  8. Time Wheel (Circular Stack) max Current time pointer t=0 Event link-list 1 2 3 4 5 6 7 ELEC 5970-003/6970-003

  9. Power Estimation • For every vector (changes at primary input): • At every signal node (gate output): • Count number of transitions • Compute #transitions × node capacitance × VDD2/2 • If node capacitances are not known, use fanout approximation – often used for relative power comparison between circuits • Add pre-estimated leakage power for vector period ELEC 5970-003/6970-003

  10. PowerMill: A Power Estimator • Event-driven simulator • Switch-level simulation with delays • 2-3 orders of magnitude faster than Spice • Estimates power for given vectors, due to • Instantaneous, average and rms currents • Steady-state transitions and glitches • Short-circuit and leakage currents ELEC 5970-003/6970-003

  11. PowerMill (Cont.) • Determines current density and voltage drop in the power net • Points to potential problem areas on a VLSI chip for EM failures, ground bounce, excessive voltage drop, heating • Reference: C. Deng, “Power Analysis for CMOS/BiCMOS Circuits,” Proc. International Workshop on Low Power Design, April 1994, pp. 3-8. ELEC 5970-003/6970-003

  12. Switch-Level Simulation 1 Channel-connected components 1 0 ? 1 1 R. E. Bryant, “A Survey of Switch-Level Algorithms,” IEEE Design & Test of Computers, vol. 4, no. 4, pp. 26-40, August 1987. 1 ELEC 5970-003/6970-003

  13. Entice-Aspen: Gate-level Tool • Gate-level circuit is partitioned into cells. • Cells are simulated in Spice for power dissipation for all possible input events. • Logic simulation determines the events at cell inputs, adding the corresponding power dissipated by each cell. • B. J. George, D. Gossain, S. C. Tyler, M. G. Wloka, and G. K. H. Yeap, “Power Analysis and Characterization for Semi-Custom Design,” Proc. International Workshop on Low Power Design, April 1994, pp. 215-218. ELEC 5970-003/6970-003

  14. RTL Power Estimation • Two step procedure: • Behavioral simulation to collect the input statistics for all modules in RTL description • Develop power macro-model for each module and sum the power • Q. Wu, C.-S. Ding, C.-T. Hsieh, and M. Pedram, “Statistical Design of Macro-Models for RT-Level Power Estimation,” Proc. Second Asia-Pacific Design Automation Conference, Jan. 1997. ELEC 5970-003/6970-003

  15. Behavioral Activity Simulation • Module description is modified to collect input statistics. • Example: 16-bit multiplier module c = a*b; r1 = a^a’; r2 = b^b’; for (i=0; i <16; i++ ) { sw_a[ i ] += r1 & 1; sw_b[ i ] += r2 & 1; r1 = r1 >> 1; r2 = r2 >> 1; } a’ = a; b’ = b; ELEC 5970-003/6970-003

  16. Power Macro-Model • Develop analytic models for estimating the switched capacitance as a function of circuit complexity and technology/library parameters. OR • Synthesize the circuit and then estimate power dissipation by simulation with random vectors. • Both methods determine effective switched capacitance per input transition. ELEC 5970-003/6970-003

  17. Effective Switched Capacitance Power dissipation of a module = 0.5 V2f C E • where • V is supply voltage • f is vector frequency • C is effective switched capacitance/input transition • E is input activity (bit changes) per vector ELEC 5970-003/6970-003

  18. Probabilistic Methods • Signal probability: Expected value of a binary signal, s T E(s) = lim (1/T)∫ s(t) dt = Prob(s=1) = p(s) =1.p(s) + 0.(1-p(s)) t=0 ELEC 5970-003/6970-003

  19. Switching Probability psw(s(t)) = p(s(t-ε))(1-p(s(t))) + (1-p(s(t-ε)))p(s(t)) = p(s(t-ε)) + p(s(t)) – 2p(s(t-ε))p(s(t)) If p(s(t-ε)) = p(s(t)) = p(s), then psw(s(t)) = 2p(s)(1-p(s)) Dynamic power = 0.5 CVDD2psw(s(t)) f ELEC 5970-003/6970-003

  20. Uncorrelated Signals • NOT: c = a, p(c) = 1 – p(a) • AND: c = ab,p(c) = p(a)p(b) • OR: c = a+b, p(c) = 1 - (1-p(a))(1-p(b)) = p(a) + p(b) – p(a)p(b) ELEC 5970-003/6970-003

  21. Correlated Signal Example p(a) = 0.5 0.25 p(c) = 0.5 0.25+0.25-0.0625 = 0.4375 output 0.5 Switching probability psw(output) = 2×0.4375×(1-0.4375) = 0.4921875 0.25 p(b) = 0.5 ELEC 5970-003/6970-003

  22. Correlated Signals (Corrected) p(output) = 0.5 psw(output) = 2×0.5×0.5 = 0.5 ELEC 5970-003/6970-003

  23. Symbolic Analysis p(a) = 0.5 p(a)p(c) = 0.25 0.25+0.25-0.0625 = 0.4375 p(c) = 0.5 output 1-p(c) = 0.5 p(a)p(c) + p(b)(1-p(c)) - p(a)p(b)p(c)(1-p(c)) =p(a)p(c)+p(b)-p(b)p(c) = 0.5 p(b)(1-p(c)) = 0.25 p(b) = 0.5 K. P. Parker and E. J. McCluskey, Probabilistic Treatment of General Combinational Networks,” IEEE Trans. Computers, vol. C-24, no. 6, pp. 668-670, June 1975. ELEC 5970-003/6970-003

  24. a c output b Supergate Supergate of a signal is the smallest circuit partition including all fanout stems dominated by that signal. Supergate(output) S. C. Seth and V. D. Agrawal, “A New Method for Computation of Probabilistic Testability in Combinational Circuits,” Integration, the VLSI Journal, vol. 7, no. 1, pp. 49-75, April 1989. ELEC 5970-003/6970-003

  25. PREDICT Algorithm • To calculate p(output), enumerate signal states at fanout signal(s) – c in this example. • For each case i, compute pi(output) separately. • Compute p(output) = p(c)p1(output) + (1-p(c))p0(output) ELEC 5970-003/6970-003

  26. Supergate Example p(a)=0.5 p(a)=0.5 0.5 0.0 0.5 c=1 0.5 c=0 0 1 output output 0.0 0.5 p (b)=0.5 p (b)=0.5 P(output) = p(c) 0.5 + (1-p(c)) 0.5 = 0.5 ELEC 5970-003/6970-003

  27. Other Signal Probability Methods • Weighted averaging • B. Krishnamurthy and I. G. Tollis, “Improved Techniques for Estimating Signal Probabilities,” IEEE Trans. Computers, vol. C-38, no. 7, pp. 1245-1251, July 1989. • Cutting algorithm • J. Savir, G. Ditlow and P. Bardell, “Random Pattern Testability,” IEEE Trans. Computers, vol. C-33, no. 1, pp. 79-90, Jan. 1989. • OBDD • R. E. Bryant, “Graph-Based Algorithms for Boolean Function Manipulation,” IEEE Trans. Computers, vol. C-35, no. 8, pp. 677-691, Aug. 1989. • Transition density • F. N. Najm, “Transition Density: A New Measure of Activity in Digital Circuits,” IEEE Trans. CAD, vol. 12, no. 2, pp. 310-323, Feb. 1993. ELEC 5970-003/6970-003

  28. Working with Delays • Signal probability methods do not take delays into account. Hence, glitch power is not included. • Timed symbolic simulation • A Ghosh, S. Devadas, K. Keutzer and J. White, “Estimation of Average Switching Activity in Combinational and Sequential Circuits,” Proc. 29th Design Automation Conf., June 1992, pp. 253-259. • Probability waveform simulation • C.-S. Ding, C.-Y. Tsui and M. Pedram, “Gate-Level Power Estimation Using Tagged Probabilistic Simulation,” IEEE Trans. CAD, vol. 17, no. 11, pp. 1099-1107, Nov. 1998. ELEC 5970-003/6970-003

More Related