Introduction to Statistical Simulation

Transcription

1 Itroductio to Statistical Simulatio Aalyze yield by observig the effects of Elemet parameter variatio Model parameter variatio.measure.alter Statistical Methods Simple Sweeps Worst Case Mote Carlo 1

2 MEASURE.MEASURE (Vol. 1, p ) Prits user-defied electrical specificatios of a circuit.measure is a post processor Used extesively for optimizatio ad curve fittig Seve fudametal measuremet modes: Rise, Fall, Delay Average, RMS, Mi, Max, ad Peak-to-Peak Fid-Whe Equatio Evaluatio Derivative Evaluatio Itegral Evaluatio Relative Error 2

3 MEASURE: Rise/Fall Sytax.MEASURE <DC TRAN AC> result TRIG TARG <optimizatio optios> result - ame give the measured value i the HSPICE output. TRIG trig_var VAL=trig_val <TD=timedelay> <CROSS=#of> <RISE=#of> + <FALL=#of> TRIG AT=value TARG targ_var VAL=targ_val <TD=timedelay> <CROSS=#of LAST><RISE=#of LAST> + <FALLS=#of LAST> Example (Vol. 1, pg 3-14).MEAS TRAN TDELAY TRIG V(1) VAL=2.5 TD=10s RISE=2 + TARG V(2) VAL=2.5 FALL=2 Delay 10s TDLAY V(1) 2.5v V(2) 2.5v 3

4 MEASURE: Avg, RMS,Sytax.MEASURE <DC TRAN AC> result fuc out_var <FROM=val> <TO=val> + <optimizatio optios> fuc: AVG, RMS, MIN, MAX, PP result: ame give the measured value i the HSPICE output out_var: ame of the output variable to be measured. Examples.MEAS TRAN avgval AVG V(10) From=10s To=55s Prit out average odal voltage of ode 10 durig tra time 10 to 55s. Prit as avgval..meas TRAN maxval MAX V(1,2) From=15s To=100s Fid the maximum voltage differece betwee odes 1 ad 2 from time 15s to 100s. Prit as maxval. 4

5 MEASURE: FIND-WHEN Sytax FIND-WHEN (Vol. 1, p. 3-18) Allows ay idepedet variables (time, freq, parameter), by usig WHEN sytax, or ay depedet variables (voltage, curret, etc), by usig FIND-WHEN sytax, to be measured whe some specific evet occurs..measure <DC TRAN AC> result WHEN out_var=val <TD=val> + <RISE=#of> LAST> <FALL=#of LAST> <CROSS=#of LAST> <optimizatio optios> result - ame give the measured value i the HSPICE output file. Example - WHEN.MEAS TRAN fifth WHEN V(osc_out)=2.5v RISE=5 measure the time of the 5th rise of ode osc_out at 2.5v. Report as fifth i listig. Example - FIND-WHEN.MEAS TRAN result FIND v(out) WHEN v(i)=40m measure v(out) whe v(i)=40m - store i variable result 5

6 MEASURE: Equatio Evaluatio Sytax Equatio Evaluatio (Vol. 1, p. 3-19) Use this statemet to evaluate a equatio that ca be a fuctio of the results of previous.measure statemets. The equatio MUST NOT be a fuctio of ode voltages or brach currets..measure <DC TRAN AC> result PARAM= equatio <optimizatio optios> Example result - ame give the measured value i the HSPICE output file..meas TRAN Tmid PARAM= (T_from+T_to)/2 6

7 MEASURE: Derivative Fuctio Derivative Fuctio (Vol. 1, p. 3-19) Provides the derivative of a output variable at a give time or frequecy or for ay sweep variable. Also provides the derivative of a specified output variable whe some specific evet occurs..measure <DC TRAN AC> result DERIVATIVE out_var WHEN +var2=val <TD=val> <RISE=#of> LAST> <FALL=#of LAST> +<CROSS=#of LAST> <optimizatio optios> Example result - ame give the measured value i the HSPICE output file..meas TRAN slewrate DERIV v(1) WHEN V(osc_out)=2.5v RISE=5 7

8 MEASURE: Itegral Fuctio Itegral Fuctio (Vol. 1, p. 3-21) Provides the itegral of a output variable over a specified period..measure <DC TRAN AC> result INTEGRAL out_var <FROM=val1> + <TO=val2> <optimizatio optios> Example result - ame give the measured value i the HSPICE output file..meas TRAN charge INTEG I(cload) FROM=10s TO=100s 8

9 LAB 6 9

10 .ALTER: Descriptio.ALTER (Vol. 1, pp. 2-22) Reru a simulatio several times with differet Circuit topology Models Library compoets Elemets Parameter values.alter Sequece for Worst Case Corer Aalysis.DELETE LIB Removes previous library selectio.lib Add ew library case Optios Source stimulus Aalysis variables Prit/plot commads (must be parameterized) 1st Ru - Star-Hspice reads iput etlist file up to the first.alter Subsequet - Reads iput etlist to ext.alter, etc.optio alt Ca use up to 9999 alters i oe ru. With.ALTER statemet oly, default is

11 .ALTER: Limitatios Limitatios: CAN iclude Elemet Statemets (except source).data,.lib,.del LIB,.INCLUDE,.MODEL statemets.ic,.nodeset statemets.op,.options,.param,.temp,.tf,.tran,.dc,.ac CANNOT iclude Mai Program.ALTER.ALTER.END $Alterative Mai Program 1.END Mai Program 2.END etc.print,.plot,.graph, or ay other i/o statemets (CAN parameterize) AVOID addig aalysis statemets uder each.alter block. ( Will cause huge pealty i simulatio time ad cofusio i result outputtig!) 11

12 .ALTER: Example (Vol. 1, pp 2-23) ALTER Test - CMOS Iverter.OPTIONS ACCT LIST.OP.DC VIN PRINT DC v(3) v(2).print tra v(3).param VDD=5.LIB MOS.LIB NORMAL VDD 1 0 VDD VIN 2 0 M P 6U 15U M N 6U 15U.ALTER Chage Supply.PARAM VDD=10.ALTER Fast Process.DEL LIB MOS.LIB NORMAL.LIB MOS.LIB FAST.ALTER Add Trasiet.TRAN.1 2.END.OP DC Sweep chage VDD to 10v.OP DC Sweep alter chage to fast library (MUST delete first) (VDD=10v still i effect).op DC Sweep alter to add trasiet (VDD=10v still i effect) (.lib mos.lib FAST still i effect).op DC Sweep Trasiet 12

13 .ALTER: Example Example - Parameterizig Source Statemets.PARAM A=4s B=5s V1 VA GND PULSE (0v 5v 0s A B 46.5s 100s) V2 VB GND PULSE (0v 5v 0s A B 96s 200s) V3 VC GND PULSE (0v 5v 0s A B 196.5s 400s).ALTER.PARAM A=5s B=6s.ALTER.PARAM A=6s B=7s.END 13

14 Statistical Methods Aalyze yield by observig the effects of: elemet parameter variatio model parameter variatio Perform multiple aalyses to modify DC operatig poits, sweeps, AC, ad trasiet aalyses. Types of Aalysis Simple sweep aalysis (Vol. 1, p. 9-19) Automated over a fixed set of values. Worst case corers aalysis (Vol. 1, p. 9-21) Compoet value limits are kow, measured, ad implemeted as skew parameters. Mote Carlo Aalysis (Vol. 1, p. 9-25) Statistical stadard deviatios of compoet values are kow. Automated over a radom rage of values. 14

15 Statistical Methods - Worst Case Simple Worst Case Aalysis (Vol. 1, P. 9-19) Switch to differet process corer libraries ad repeat simulatio..alter,.del LIB,.LIB (you MUST delete a lib before callig a ew lib) Worst Case Sweep Aalysis.LIB CMOS.lib SLOW.Tra.ALTER Ru Fast Corer Library.DEL LIB CMOS.lib SLOW.LIB CMOS.lib FAST HSPICE models ca iclude physically measurable model skew parameters. e.g. poly CD, active CD, tox, diffusio resistivity, threshold voltage variatio, etc. Icorporated ito models as skew variables. e.g. +/- 3 sigma variatio to TOX..PARAM sigma=0 toxcd=200 tox= toxcd - sigma * 10.TRAN 20p 1 SWEEP sigma model tox=tox 15

16 Statistical Methods - Mote Carlo Statistical deviatios for elemet ad model parameters are kow. Radom umber geeratio creates variatios withi: Gaussia distributio. Uiform distributio. Limit distributio. Setup (Vol. 1, 9-26).PARAM statemet. (Set elemet/model parameter to fuctio distributio). Set.DC,.AC,.TRAN to MONTE.MEASURE to calculate mea, variace, sigma, std. deviatio..param tox = agauss ( 200, 10, 1 ).TRAN 20p 1 sweep MONTE=20.model tox=tox 16

17 Statistical Methods - Mote Carlo Settig up Mote Carlo (Vol. 1, P. 9-27) Use a.param to set a elemet or model parameter to: GAUSSIAN.PARAM ver=gauss(om_val,rel_variatio,sigma, mult).param ver=agauss(om_val,abs_variatio,sigma, mult) UNIFORM PARAM ver=uif(om_val,rel_variatio, mult).param ver=auif(om_val,abs_variatio, mult) LIMIT PARAM ver=limit(om_val,abs_variatio) WARNING - if you misspell GAUSS (or uiform or limit) there is o warig, but o distributio will occur. 17

18 Statistical Methods - Mote Carlo Settig up Mote Carlo Aalysis must be.ac,.dc or.tran to eable MONTE.DC vi sweep mote=val (val=# of radomizatios).ac dec meg sweep mote=val.tran 1 10 sweep mote=val A.MEASURE statemet ca be used to calculate the mea, variace, ad sigma of ay output desired. R1 1 0 Res R2 2 0 Res 2.tra 2 s 100s sweep mote=30.param Res=gauss(10k, 1k, 1) + Res2=gauss ( Res+1, 1k, 1).measure tra ame fid v(3) at 50s 18

19 LAB 7 19