Modeling of the SiGe power HBT IM Distortion

Transcription

1 Modeling of the SiGe power HBT IM Distortion P.Sakalas %,$, M.Schröter %, L.Kornau &, W.Kraus & % Dresden University of Technology, Mommsenstrasse 13, Dresden, Germany & Atmel Germany GmbH, Theresienstrasse 2, Heilbronn, Germany $ Semiconductor Physics Institute, FRL, Goštauto 11, Vilnius 2600, Lithuania HICUM WORKSHOP, DRESDEN June

2 Outline: Introduction SiGe power HBTs device characteristics and model comparison IM Distortion and Power Spectrum measurement set-up Power and IMD Spectrum model comparison Summary 2

3 Introduction A result of the distortion of HBT or nonlinear circuit is that it generate frequency components in the output signal that are not present in the input. Bandpass filtering can eliminate much of the effects of harmonic distortion, while intermodulation distortion is difficult to filter out because the IMD components may be very close to the carrier frequency. Thus investigation of the common figure of merit, two-tone intermodulation distortion is of high importance for power amplifiers. Nowadays Si based power amplifiers enable system on a chip technology and low-cost lightweight communication systems. This attracts space enterprises, such as NASA, which rely on SiGe power Heterojunction Bipolar Transistors and passive circuit elements on silicon substrates. Thus intermodulation analysis of circuit basic elements, HBTs, with the help of advanced compact models, such as HICUM is of high scientific and commercial importance. 3

4 SiGe Power HBT basic cell E/G E/G CEBEC 7,2 7,2 High voltage (power) device fabricated with Atmel s SIGE1 process: f T =20 GHz and f max =45 GHz 7,2 7,2 B 7,2 7,2 C R EX emitter stripe window area is A E0 = 2x49.7x1.3µm 2 E/G E/G 7,2 7,2 emitter balast resistances R EX (for thermal stability) included in layout 4

5 e-embedding of Pad parasitic Network open short 1 short 2 thru Z Thru 7,27,2 7,2 7,2 G3 Three step de-embedding Assumptions: [(1/G 3 )+Z 1 ] >> Z 3 => (f< 50 GHz) Z Thru <<Z 1 or Z 2 G1 Z1 DUT Z3 Z2 G2 G1 is gate pad-source admittance, G2 is drain pad-source admittance, G3 is gate pad-drain pad admittance, Z1, Z2 are metal interconnection series impedances between port 1 and port 2, Z3 is series impedance of ground lead to DUT 5

6 e-embedded cold HBT y-parameters Re(y11)[S], Im(y11) HICUM red lines Cold HBT Im(y11) Re(y11) fre q, GHz Re(y12)[S], Im(y12) Cold HBT Re(y12) fre q, GHz Im(y12) Re(y22)[S], Im(y22) Cold HBT Im(y22) Re(y22) h21 [db] E8 Vce=0V Vbe =0V Good indication of Cjei and Cjci values. 1E9 High Re influence 1E10 3E10 fre q, GHz Frequency (Hz) 6

7 Re(y11)[S] f=1ghz Bias dependent y11, y I C /A E0 [ma/µm 2 ] HICUM red lines VCE=0.400 VCE=0.800 VCE=1.500 VCE=2.000 Im(y11) f=1ghz VCE=0.400 VCE= I C /A E0 [ma/µm 2 ] VCE=1.500 VCE= VCE=1.500 VCE=2.000 VCE=0.800 Re(y12)[S] f=1ghz I C /A E0 [ma/µm 2 ] Im(y12) f=1ghz VCE= I C /A E0 [ma/µm 2 ] 7

8 Bias dependent y21, y22 HICUM red lines Re(y21)[S] f=1ghz VCE=0.400 VCE=1.500 VCE=2.000 VCE=0.800 Im(y21) f=1ghz VCE=0.400 VCE=1.500 VCE=2.000 VCE= I C /A E0 [ma/µm 2 ] I C /A E0 [ma/µm 2 ] -6 Re(y22)[S] f=1ghz I C /A E0 [ma/µm 2 ] VCE=0.400 VCE=0.800 VCE=1.500 VCE=2.000 Im(y12) f=1ghz I C /A E0 [ma/µm 2 ] VCE=0.400 VCE=0.800 VCE=1.500 VCE=

9 Gummel forward characteristic HICUM is red lines. High R E reduce I C. 1 1 I C /A E0 [ma/µm 2 ] 1E-1 1E-2 1E-3 1E-4 1E-5 Ic /Ae o Ib/Ae o R E =11 Ω I C /A E0 [ma/µm 2 ] 1E-1 1E-2 1E-3 1E-4 1E-5 R E =1.2 Ω V BE [V] V BE [V] Gummel plots for higher and lower R E values. 9

10 C E Output characteristic, measured HICUM m2 versus simulated Avalanche multiplic. m1 IBB=360.u IBB=330.u IBB=300.u IBB=270.u IBB=240.u IBB=210.u IBB=180.u IBB=150.u IBB=120.u IBB=90.0u IBB=60.0u IBB=30.0u I C /A E0 [ma/µm 2 ] Gummel Poon m2 Thermal lattice heating. m V CE [V] IBB= V CE [V] HICUM yields good agreement with measured data, except at V CE >5 I B >90 µa, where discrepance is due to simplified expression of avalanche multiplication current (for saving CPU time!). Gummel Poon model overestimates I C due to missing internal R CI bias dependence. At V CE > 3 V selfheating causes decrease of I C ; both effects are not accounted for by the model. 10

11 Cut-off frequency V CE = 0.4 V V CE = 0.8 V V CE = 2 V V CE =2.0V V CE =0.8V f T [GHz] V CE =0.4V I C /A E0 [ma/µm 2 ] 11

12 Beta Beta U CE = 0.4 V U CE = 0.8 V U CE = 1.5 V U CE = 2 V V CE =1.5V V CE =0.8V V CE =2V V CE =0.4V I C /A E [ma / µm 2 ] I C /A E0 [ma/µm 2 ] 12

13 Two-tone IM Distortion and Power Spectrum measurement set-up Two-tone measurement setup Power spectrum measurement setup Splitter/coupler Dc power supply Bias tees Frequency synthesizer High resolution in-house filters Spectrum Analyser HP83650B HP4145 FILTERS BT BT DUT PC CASCADE WILTRON 562 Frequency synthesizers Cascade Probestation DUT Cascade Probestation Scalar Network Analyser 13

14 Power Spectrum P OUT [dbm] Vce=3V, Pinp=0dBm HICUM 1E-4 1E-3 1E-2 1E-1 4E-1 I C /A E0 [ma/µm 2 ] 0.5 GHz 1 GHz 1.5 GHz Lines are HICUM Circle is meas ured 0.5 GHz Square is measured 1.0 GHz Triangle is meas ured 1.5 GHz P OUT [dbm] 20 Vc e =3V Lines are Gummel Poon Circle is meas. 0.5 GHz Square is meas. 1.0 GHz Triangle is meas.1.5 GHz Gummel Poon 0.5 GHz 1.5 GHz 1 GHz 1E-4 1E-3 1E-2 1E-1 7E- I C /A E0 [ma/µm 2 ] ICUM gives better agreement compared to Gummel Poon model at low and high I C /A E0. ummel Poon harmonic Balance simulation convergence is faster (simulator dependent though) at the ost of accuracy. 14

15 Single tone 1 db compression point 1 db compression point P OUT [dbm] P IN [dbm] 15

16 P fund, 3 rd IMD [dbm] f1=5.00 GHz f2=5.01 GHz HICUM Two-tone IM Distortion HICUM Third order intercept point (IP3). PFund1 IMD Pinp [dbm] 3 rd IMD~P inp 3 Vce=2.5V, Vbe=0.88V, Ib=214µA, Ic=25 ma P fund, 3 rd IMD [dbm] -100 f1=5.00 GHz f2=5.01 GHz -120 Vce=2.5V, Vbe=0.88V, -140 Ib=214µA, Ic=25 ma Output power at f 1, f 2 and IMD3 vs two-tone input power. V CE =2.5 V, I C /A E0 =0.38 ma/µm 2. Frequency spacing f=10 MHz between input signals => f 2 =5.01 GHz. The input power of both signals (P in,1 =P in,2 ) was swept from to dbm. HICUM yields good agreement over a large input power range, while beyond Pinp=-10dBm Gummel Poon starts to slightly underestimate measured IMD3. If intercept point is known, IMD level related to output signal level can be calculated from: IMD dbc =2(P OUT,dBm -IP3,dBm ) Gummel Poon Gummel Poon PFund1 IMD3 Pinp [dbm] 16

17 Two-tone Fine Spectrum 50 Pin=-1.86 dbm HICUM Spectrum [dbm] 0-50 IMD3 PFund1 PFund2 IMD3 Desired for power amplifiers IMD3 power level Frequency Two-tone measured and simulated output fine spectrum at V CE =2.5 V, I C /A E0 =0.192 ma/µm 2 Since IMD3 increases with increasing signal levels it can be used for to establish dynamic range of the amplifier: DR={2*IP3-2[10log(k*T EQV *B)+NF+G]}/3. 17

18 Transducer Power Gain Transducer power Gain [db] HICUM Two-tone input signal. f 1 =5.0 GHz, f 2 =5.01 GHz Pinp [dbm] Transducer Power Gain versus two-tone input power. V CE =2.5 V, I C =25 ma, I B =214 µa. G T is flat at 10.5 db until about P in =-15 dbm, and then decreases rapidly down to 5 db at 1dBm. Analytical calculations using de-embedded measured y-parameters yield G T =10.6 db, which agrees fairly well with both HICUM and measurements. 18

19 Summary IV, S-parameters, single tone power spectrum 1-dB compression, intermodulation distortion, resulting from two-tone input signals with fundamental frequencies up to 5 GHz were measured on a real Si/SiGe HBT, fabricated in a production process. Geometry scalable HICUM parameters were extracted. Good agreement between model and measurements was obtained, demonstrating the suitability of HICUM for critical high-frequency applications. 19

20 Acknowledgments German Research Society (DFG): financial support within the SFB 358 project. Suss Microtech and Agilent: hardware and software donations. Iltcho Angelov from Chalmers university: for power spectrum measurement support. 20