PCM- and Physics-Based Statistical BJT Modeling Using HICUM and TRADICA Wolfgang Kraus Atmel Germany wolfgang.kraus@hno.atmel.com 6th HICUM Workshop, Heilbronn (Germany), June 2006 c WK Jun 12th, 2006 1 / 21
Outline c WK Jun 12th, 2006 2 / 21
roots in digital design Corner models Approach used often similar topology 2 highly correlated figures of merit : speed & power simplified : V t = I C leads to fast & slow corner models possibly good enough for slew rate but what about hysteresis of Schmidt trigger phase margin offset voltage headroom of an amplifier depends on design analog design knowhow c WK Jun 12th, 2006 3 / 21
something more flexible is needed Corner models Approach used but on solid physical base proposed approach allows wcs/bcs simulation, definable by designer on top of that simulation of local ( mismatch ) variation for selected devices simulation of global ( process ) variation sweep of process parameter c WK Jun 12th, 2006 4 / 21
Parameters 1 Parameters 2 Example 1 T : Technology parameters ( doping, layer thickness,...) S : Specific electrical parameters ( sheet res., cap. per area,...) G : Geometric parameters ( design & TDR ) E : Electrically measurable data ( E-test, PCM ) M : Model parameters T G S DUT E M c WK Jun 12th, 2006 5 / 21
why statistics on P={T,S}? Parameters 1 Parameters 2 Example 1 (P,G) only rational basis for process variations and mismatch no need to specify correlations during life cycle of technology changes occur in process flow ( P) and layout rules ( G) makes retargeting easier P = P nom + P G + P L (w, l) P G set external to models common to all instances deterministic ( corners ) or random ( process variation ) P L random ( mismatch, local variation ) evaluated per instance c WK Jun 12th, 2006 6 / 21
To get familiar with the idea, consider simplest case: Parameters 1 Parameters 2 Example 1 low ohmic poly resistor R = 2 RCS W + RSH L W +DW G : W, L P : RSH, RCS, DW E : R c WK Jun 12th, 2006 7 / 21
For the variance of the poly resistor we have : Example 1 contd. σ 2 R = ( ) 2 ( R RSH σrsh 2 + R DW ) 2 σ 2 DW + ( ) 2 R RCS σrcs 2 two ways to use : POV : calculate σ 2 R BPOV : from known variances of RSH, RCS, DW observe σ 2 R calculate sensitivities solve for variances of RSH, RCS, DW s.t. nonnegativity constraint ([5]) same approach for mismatch and process variation c WK Jun 12th, 2006 8 / 21
Use physics based nominal model at start Process to model HICUM example internal BJT external BJT We use TRADICA for model generation Shift nominal model to center if neccessary Enhanced version of TRADICA generates models incl. mapping equations For examples of mapping relations see ([3], [4]) Uses Taylor-series expansion of mapping relations Up to now sufficiently accurate c WK Jun 12th, 2006 9 / 21
Part of HICUM model for SPECTRE : Process to model HICUM example internal BJT external BJT subckt N00705B110H c b e s parameters + b e0 =0.500E-06 l e0 =0.700E-06 + r nbei l = r nbei rm nsic*r nbei sm nsic/sqrt(b e0*l e0) + r nbei = r nbei g nsic+r nbei l + a be0 l = a be0 rm nsic*a be0 sm nsic/sqrt(l e0) + a be0 = a be0 g nsic+a be0 l Q c b e s MOD model MOD bht type=npn tnom=27 updatelevel=1 + c10 = 2.364E-33-7.680E-34* r nbei + 6.674E-27* a be0 + ich = 8.956E-04 + 1.277E+03* a be0 + cjei0 = 2.023E-15 + 9.636E-16* r nbei + 2.900E-09* a be0 c WK Jun 12th, 2006 10 / 21
Process to model HICUM example internal BJT external BJT Process Parameters SGPM Parameters HICUM Parameters int. base doping Ic, BF c10, qp0, cjei0, rbi0 ( r nbei ) 1 Cje, Rb cjep0 int. base width Ic, BF qp0, tbvl, thcs ( r wb ) Rb int. collector doping Cjc, tf cjci0, tbvl, thcs ( r nci ) IKF, IT F vptci, t0 base current Ib, BF ibeis, ireis ( r jbei ) emitter size Ic, Ib, Cje c10, qp0, ich ( a be0 ) 2 tf cjei0, cjci0, t0 int. collector width IKF, IT F vptci, thcs, rci0 ( a wc ) latb, latl 1 r : relative variation 2 a : absolute variation c WK Jun 12th, 2006 11 / 21
Process to model HICUM example internal BJT external BJT Process Parameters SGPM Parameters HICUM Parameters ext. collector doping Cjc cjcx0 ( r ncx ) buried layer doping Rc rcx ( r nbl ) substrate doping Cjs cjs0, iscs ( r nsu ) Observations : external BJT parameters (mostly) uncorrelated at least 8 correlated internal BJT parameters ( single device ) now imagine you ve 10 or 100 BJT models... physics-based approach much more efficient c WK Jun 12th, 2006 12 / 21
refers to direct model parameter extraction from PCM data Open loop extraction Loop closure usual candidates : V T H, DL, DW, BF, V AF,... no guarantee your simulations are o.k. when the model contains published values from EDR reasons : many methods to extract physical parameters exist, giving different results different models for same device may need different parameters relation between electrical observables and model parameters is approximate goal of modeling : as good as possible representation of electrical behavior, not of physical parameters c WK Jun 12th, 2006 13 / 21
refers to indirect or closed-loop procedure ([1]) mimick PCM conditions as close as possible same device geometry Open loop extraction Loop closure same bias same data reduction simulate PCM test compare distributions from PCM test and simulation repeat cycle regularly, even for mature technology c WK Jun 12th, 2006 14 / 21
The following slides show selected results. All simulations have been done using HICUM/L2 with SPECTRE. Overview Cbe Vbe matching PCM correlations sim. correlations Large area BE-Capacitance of VNPN V BE -mismatch of VNPN-pair measured correlations of VPNP simulated correlations of VPNP c WK Jun 12th, 2006 15 / 21
PCM 8105 Devices Technology: UHF6S NPN E(50x50)um2 PCM Test: CBE_N50 LSL,USL=3Sigma Values Normal Q Q Plot Overview Cbe 0.0 0.2 0.4 0.6 0.8 1.0 LSL PCM USL PCM 13 14 15 16 17 CBE_N50 Studentized Order 3 2 1 0 1 2 3 2 0 2 Normal Quantiles Vbe matching PCM correlations 800 Montecarlo Simulations Normal Q Q Plot sim. correlations 0.0 0.2 0.4 0.6 0.8 1.0 LSL Simulation USL Simulation Studentized Order 3 2 1 0 1 2 3 13 14 15 16 17 3 2 1 0 1 2 3 CBE_N50sim Normal Quantiles c WK Jun 12th, 2006 16 / 21
PCM 8866 Devices Technology: UHF6S NPN E(0.5x9.7)um2 PCM Test: DVBE_NN1_085 LSL,USL=3Sigma Values Normal Q Q Plot Overview Cbe 0 500 1000 1500 2000 LSL PCM USL PCM 1e 03 5e 04 0e+00 5e 04 1e 03 dvbe Studentized Order 3 2 1 0 1 2 3 4 2 0 2 4 Normal Quantiles Vbe matching PCM correlations 2500 Montecarlo Simulations Normal Q Q Plot sim. correlations 0 500 1000 1500 2000 LSL Simulation USL Simulation Studentized Order 3 2 1 0 1 2 3 1e 03 5e 04 0e+00 5e 04 1e 03 3 2 1 0 1 2 3 dvbesim Normal Quantiles c WK Jun 12th, 2006 17 / 21
1.4e 05 2.2e 05 12 14 16 18 IC_DD1_085 0.00045 36 B_DD1_085 34 N = 8190 Bandwidth = 2.872e 07 30 32 Overview Cbe Vbe matching PCM correlations sim. correlations IB_DD1_085 N = 8190 Bandwidth = 8.617e 06 1.4e 05 1.8e 05 2.2e 05 20 0.00060 1.8e 05 0.00075 Technology: UHF6S VPNP4 PCM_Measurements:DD1 E(0.5x9.7)um2 CBEB 26 28 20 VAFI_DD1 12 14 16 18 N = 8190 Bandwidth = 0.3052 0.00045 0.00055 0.00065 0.00075 26 28 30 32 34 36 N = 8190 Bandwidth = 0.2458 c WK Jun 12th, 2006 18 / 21
Technology: UHF6S VPNP4 Process Simulations:DD1 E(0.5x9.7)um2 CBEB 2.0e 05 13 14 15 16 IC_DD1_085sim 0.00055 1.8e 05 0.00065 1.6e 05 0.00045 IB_DD1_085sim 1.9e 05 26 28 30 32 34 36 38 B_DD1_085sim N = 2500 Bandwidth = 1.482e 07 Overview Cbe Vbe matching PCM correlations sim. correlations N = 2500 Bandwidth = 6.508e 06 1.6e 05 16 VAFI_DD1_sim 13 14 15 N = 2500 Bandwidth = 0.3471 0.00045 0.00055 0.00065 26 28 30 32 34 36 38 N = 2500 Bandwidth = 0.1265 c WK Jun 12th, 2006 19 / 21
discussed statistical BJT models using HICUM ( & SGPM ) based on physics and fundamental process parameters no overhead for correlations VNPN & VPNP covered incl. mismatch limitations of corner models discussed Loop closure discussed to do extend statistics to s-parameters additional mismatch teststructures needed c WK Jun 12th, 2006 20 / 21
[1] C.C. McAndrew, Compact Device Modeling for Circuit Simulation, CICC 1997, pp. 151-158 [2] C.C. McAndrew, Statistical Modeling for Circuit Simulation, Interntl. Symp. on Quality Electronic Design, ISQED 03 [3] P.G. Drennan and C.C. McAndrew, A Comprehensive MOSFET Mismatch Model, IEDM 1999, pp. 167-170 [4] M. Schroter, H. Wittkopf, and W. Kraus, Statistical Modeling of High-Frequency Bipolar Transistors, invited paper, BCTM 05 [5] Lawson, Hanson, Solving Least Squares Problems, Prentice Hall, 1974 [6] O Riordan, Recommended Spectre Monte Carlo Modeling Methodology,, www.designers-guide.org c WK Jun 12th, 2006 21 / 21