Large-signal PHEMT and HBT modeling for power amplifier applications. Ce-Jun Wei Skyworks Inc. Sept

Transcription

1 Large-signal PHEMT and HBT modeling for power amplifier applications Ce-Jun Wei Skyworks Inc. Sept

2 Agenda Introduction Phemt modeling issues Empirical model vs table-based model; Charge model vs no-charge model Class-inverse-F operation of power PHEMTs Models of III-V-based HBTs Modofied GP model VBIC model Modified VBIC model Thermal coupling model Small-signal and large-signal HBT model verification HBT modeling application to power amplifier

3 Challenges of modeling for power amplifier designs PHEMT/HBTs feature higher efficiency, high frequency and good linearity and are being widely used in power amplifiers for wireless communications Commercial models are difficult to predict consistent small-signal and large-signal power performance including linearity. The requirements for a good model are: Must be capable of reproducing three-terminal dc IV curves over wide range and possible IV collapses Must be capable of fitting measured S-parameters over a wide bias range Must accurately predict power, efficiency and linearity Must be able to predict load-pull behavior Must be scaleable to large-size used in power amplifiers Good convergence

4 PHEMT modeling issue: Self-heating Positive RF Gds but Negative DC Gdso at Higher Power Dissipation Region Id A) Vds (V)

5 PHEMT modeling issue: I-V dispersion Id A) - DC-IV Does Not Mean Equal to RF-IV -RF IVs That Fit RF Gm and RF Gds Differ From Each Other DC IV RF IV That Fits RF Gm RF IV That Fits RF Gds Vds (V) Igm (A) Vds (V) Igds (A) Vds (V)

6 PHEMT modeling issue: Charge Conservation? 2d-charge Qg Can Be Integrated From Extracted (Based on Measurement Data) Cgs(vgs,vgd) and Cgd(vgs,vgd) and Should Be Path-independent Charge Conservation or Path Independence Rule Requires: Q g = ƒ(c gs dv gs + C gd dv gd ) C gs / V gd = C gd / V gs For Small Size Devices and No Significant Dispersion, Path Independence Does Hold. In general, it does not hold, because of improper equivalent circuit

7 PHEMT modeling issue: consistence and others A derived small-signal model from the large-signal model must be consistent with small-signal models over a wide range of biases 2D QV Functions in Large-signal Model Introduce Additional Trans-capacitances that do not exist in small-signal models Be continuous up to at least third derivatives of IV and QV curves Accurate gate current model including leakage and breakdown Qgd(Vgs,Vgd) Qgs(Vgs,Vgd)

8 Empirical model verses Table-based model Both models use simple -shaped intrinsic equivalent circuit Both models use IV and QV characteristics and assume path-independence Both models use simple linear or nonlinear RC-type circuit on drain side to account for low-frequency dispersion Empirical models have advantages of approximate mapping onto device physical structure, large-dynamic range independent of measurement range. Their disadvantage is accuracy. Table-based models have advantages of least-parameter-extraction, technology-independence, accuracy but the disadvantages are: slower convergence, limited validity in its measurement range in extraction.

9 Dispersion model of PHEMTs Instead of Using RC Branch in Drain Port, Alpha Model Uses a Feed-back and Feedforward Circuit to Modify the RF Gds and RF Gm. Self-heating Effects Are Modeled by a Sub-thermal-circuit and a Coefficient of Id Modification Rg Lg = Id (1-exp(-ct Vth)) Qgd Ld Rd Id(Vgt,Vds) Qgs Id Cds VCVS Vgs Crf Cfb Vrf Feedforward Vgt Vrf Self-heating Induced Current Rrf Gm=+GmoCfb Gds=+GdsoCbk Feedback VCVS S Ls Rs Vth Thermal Sub-circuit Ith=Vds Id Cth Rth

10 No-Charge model Use Capacitive Current Sources to Replace Charge Sources Create a Virtual Node (Voltages dv_dt) That Are Proportional to Timederivative of Vgs or Vgd. The Capacitance Current, C(Vs)*dV_dt, Is the Nonlinear Function of Vgs,vgd and dv_dt V C dv/dt + - Cgd dvgd/dt igs igd ids CSVS (1/C) Cgs dvgs/dt

11 Charge model verses Non-charge model No extra trans-capacitances are involved Complete and one-by-one-correspondence consistence with smallsignal models over all bias-points measured Care must be taken to avoid average component of capacitive currents. Use CR broke circuit for each current Charge model is still better in convergence. Both models can be table-based or empirical.

12 Application: 2-tone Load-pull Simulation The Results Are Verified by Comparing the Measured at Several Points p p u rs_ o n t o _ c P del u rs_ o n t E _co A P m2 m1 Gain Gain Unchanged but TMD Improved by 8 db p _ u rs n t o p u rs_ o n t D _co D _co M M O rdi O rdi f th d T hi r F i TMD & FMD m3 ThirdO level=impeda indep(pae_contours_p) (0.000 to 6.000) indep(pdel_contours_p) (0.000 to ) m1 m1 indep(m1)=6 PAE_contours_p=0.728 / level= , number=1 impedance = Z0 * ( j0.338) indep(thirdordimd_contours_p) (0.000 to ) indep(fifthordimd_contours_p) (0.000 to ) Minimum 3rd-Order IMD, dbc Minimum 5th-Order IMD, dbc

13 Application: Waveform at Inverse-F and Class-F Operation Vds (V) Class Inverse-F (PAE 80%) Class F (PAE 69%) 2 nd : open; 3 rd : short 2 nd : short; 3 rd : open Time (ps) Id (ma) Phase (rad) at 0.938GHz Symbol: Measured Line: Modeled Vds = 3.2 V Vg = V (Inverse F), Vg = -1.1 V (Class-F), Total Wg = 2 mm Id (ma) Vds (V)

14 Application: load-line of ideal Inverse-F and Class-F Operation High PAE Requirement: -Id 0 when Vd swings, Vd minimized, when Id swings - fast transit for Id*Vd 0 (broken line) Class F: visit more time on resistive loss area than clss inverse-f Ids (A) Class F Bias point Class IF Vds (V)

15 HBT modeling Most hand-set PA s are using HBTs The advantages over PHEMTs: unipolar DC supply, uniformity and high yield, linearity. Caution must be taken on thermal management Commercial and non-conmercial models - Commercial models: GP, VBIC, Mextram, Hicum - Non-commercial models: Modified-GP, Modified-VBIC or others

16 VBIC model and features Self-heating Separation of the transfer current and base current External BE diode Parasitic PNP Early effect on Tf Quasi-saturation Comprehensive Temperature-dependent parameters

17 Modified GP model and features major Self-heating effects including nonlinear terms Separation of the transfer current and base current Comprehensive Temperature-dependent parameters Additional terminal for thermal coupling simulation C Rcx Dbce Cjbc1 Dbc Cjbc Cdbc Icc-Iee B Rb CSRC2 Rbi Dbei Dbep Cjbe Cdbe Re dtj SRC1 R1 C1 CSRC1 E

18 Tf and Cbc characteristics that commercial models can not fit Ft as function of Vcb & Ic Vcb=-0.8, & -0.5 to 4 V step 0.5 V Curoff frequency (Hz) 6.00E E E E E E E+00 Vcb Collector current (A) Base-collector capacitance (F) Cbc as finction of Vbc & Ic Vbc=0.5 to 3.5 V step 0.5 V 1.0E E E E Collector current (A) measured

19 Modified VBIC model and features Self-heating accurate Tf model to account for ft drop at higher current (Kirk Effect) 1PF Idvbcdt Vbc & Ic dependent Cbc due to mobile-charge modulation and Kirk Effect VCVS Var Eqn RCI VAR1 Var Eqn _VAR2 Implemented with SDD in ADS LC C B RB LB RBI RCX Tj E LE RE Rth MOD_VBIC

20 Ic-Vc and Vb-Vc curves at constant Ib modeled vs measured Ae=56um^2 Ic-Vc Vb-Vc Ic_meas10.i Ic_meas9.i Ic_meas8.i Ic_meas7.i Ic_meas6.i Ic_meas5.i Ic_meas4.i Ic_meas3.i Ic_meas2.i Ic_meas.i IC.i datasetname..it70 datasetname..it60 datasetname..it50 datasetname..it40 datasetname..it30 datasetname..it20 datasetname..it10 datasetname..it0 var("datasetname..it-10") var("datasetname..it-20") var("datasetname..it-30") I_Ic.i, A Vc VDS VC Symbol:Modeled, Solid line:measured

21 Modified VBIC fits ft at higher current Ft as function of Vcb & Ic Vcb=0V Solid line: Modified VBIC Broken line: VBIC model 7.00E+10 cutoff frequency (Hz) 6.00E E E E E E E Ic (ma) at Vcb=0 V ft-model ft_meas ft-vbic

22 IV collapse modeled vs measured Ae=960 um2 Ib=0.4mA to 4.4 ma step 0.4mA 500 simulated 0.5 measured IDS.i, ma Ic-meas (A) VDS Vc (V)

23 Power performance Modeled vs Measured V27 H um^2 Vc=3.2V Ic=7.54mA 25 m1 Pin= Gain= m1 50 var("psweep..pout") var("psweep..gain") Pout Gain Psweep..Efficiency PAE Pin 0

24 Harmonics performance Modeled vs Measured V27 H um^2 Vc=3.5V Ic=7mA var("datasetname..3f1") var("datasetname..2f1") datasetname..pout_meas LoadPull_VBIC..f3 LoadPull_VBIC..f2 LoadPull_VBIC..f1 f3 f2 f Pin Pin_meas Vc=3.5 V Vb=1.357 V Ic=6.98 ma Rbext=600ohm Source: f1: F2: F3: Load: f F2: F3: Symbol:measured, solid line:new model, broken lin:vbic

25 IM3 & IM5 performance Modeled vs Measured V27 H um^2 Vc=3.5V Ic=7mA Vc=3.5 V Vb=1.357 V Ic=6.98 ma Rbext=600ohm Source: f1: F2: F3: Pout IM3H IM3L IM5H IM5L twotone_mtfq..im3h twotone_mtfq..im3l twotone_mtfq..im5h twotone_mtfq..im5l twotone_mtfq..pout MAXPAE..IM3H MAXPAE..IM3L MAXPAE..IM5H MAXPAE..IM5L MAXPAE..Po1 Load: f Pin F2: F3: Symbol:measured, solid line:new model, broken lin:vbic

26 Linearity improves for punch-through structure Punch- through above

27 Pout load-pull, Modeled vs Measured V27 H um^2 Vc=3.2V Ic=37mA Pin=0dBm Gamm(2)=0.52<-117 range14 range13 range12 range11 range10 range9 range8 Pdel_contours_p indep(pdel_contours_p) (0.000 to ) (0.000 to ) Maximum Power Delivered, dbm Pout_step=1 Range8 : >max-1 Range9 : max-2 Range10 : max-3 Range11 : max-5 Range12 : max-7 Range13 :max-10 Range13 :max-15 Max=21.4

28 PAE load-pull, Modeled vs Measured V27 H um^2 Vc=3.2V Ic=37mA Pin=0dBm Gamm(2)=0.52<-117 Maximum Power-Added Efficiency, % range7 range6 range5 range4 range3 range2 range1 PAE_contours_p PAE_step=5 range1 : >max-6 range2 : max-15 range3 : max-25 range4 : max-35 range5 : max-45 range6 :max-55 indep(pae_contours_p) (0.000 to ) (0.000 to ) Max=61.6

29 IM3 load-pull, Modeled vs Measured V27 H um^2 Vc=3.2V Ic=37mA Pin=0dBm Gamm(2)=0.52<-117 range4 range6 range5 range3 range2 range1 ThirdOrdIMD_contours_p m3 m3 indep(m3)=4 ThirdOrdIMD_contours_p=0.679 / level= , number=2 impedance = Z0 * ( j0.395) IM3_modeled_step=2 range1 : range2: range3 : Range4 +: Range5 x: Ranger6 :>-8 indep(thirdordimd_contours_p) (0.000 to ) (0.000 to )

30 PAE load-pull for 2 nd harmonic, Modeled vs Measured V27 H um^2 Vc=3.2V Ic=37mA Pin=3dBm Gamm(1)=0.558<111 Maximum Cal. PAE, % range7 range6 range5 range4 range3 range2 range1 PAE_contours_p No obvious difference of class F and inverse-f! PAE_modeled_step=2.5 range1 : max-3 max range2 : max-6 max-3 range3 : max-10 max-6 range4 : max-15 max-10 range5 : max-20 max-15 ranger6 :<max-20 indep(pae_contours_p) (0.000 to ) (0.000 to ) Max=56.7

31 Pout load-pull for 2 nd harmonic, Modeled vs Measured V27 H um^2 Vc=3.2V Ic=37mA Pin=3dBm Gamm(1)=0.558<111 Maximum Cal. Pout, dbm range14 range13 range12 range11 range10 range9 range8 Pdel_contours_p For Pout inverse-f is better than class F! Pout_modeled_step=0.5 Range8 : range9 : range10 : range11 : range12: range13: ranger13: <10.9 indep(pdel_contours_p) (0.000 to ) (0.000 to )

32 IM3 load-pull for 2 nd harmonic, Modeled vs Measured V27 H um^2 Vc=3.2V Ic=37mA Pin=3dBm Gamm(1)=0.558<111 Minimum 3rd-Order IMD, dbc range4 range6 range5 range3 range2 range1 ThirdOrdIMD_contours_p m1 indep(m1)=3 ThirdOrdIMD_contours_p=0.736 / level= , number=1 impedance = Z0 * ( j3.197) m1 For IM3 inverse-f is also better than class F! IM3_modeled_step=0.5 range1 : range2 : range3 : Range4 +: Range5 x: ranger6:>-8 indep(thirdordimd_contours_p) (0.000 to ) (0.000 to )

33 Conclusion The problems with conventional large-signal PHEMT models are addressed that include: dispersion, non-charge-conservation originated from use of simple equivalent circuit, etc Dispersion and no-charge models are presented that overcome the difficulties The issues in HBT modeling in terms of mobile charge-modulation and Kirk effects are addressed and modified MP and VBIC models are presented The models are verified with comprehensive load-pull results Class inverse-f with 2 nd harmonic tuned at high impedance is recommended for PHEMT PA design due to its higher PAE over class-f and is likely useful due to its better linearity for HBT power amplifiers.

34 Acknowledgement Dr. Gene Tkachenko, Dr. Ding Dai for his support Significant contribution by Dr. G. Tkachenko, A. Klimashow, J, Gering and D. Bartle