Pulsed-IV Pulsed-RF Measurements using a Large Signal Network ...

Pulsed-IV Pulsed-RF Measurements using a Large Signal Network Analyzer

S. J. Doo, P. Roblin, S. Lee, D. Chaillot, and M. V. Bossche

ARFTG 65th Conference, CA, June 2005

Pulsed-IV Pulsed-RF Measurements Using a Large Signal Network Analyzer Seok Joo Doo*, Patrick Roblin*#, Sunyoung Lee*, Dominique Chaillot*+ and Marc Vanden Bossche+ *The Ohio State University, *+on leave from CEA, +NMDG, # [email protected] Abstract—A new pulsed-IV pulsed-RF measurement system using a large signal network analyzer (LSNA) is proposed to address the problem of desensitization afflicting conventional pulsed-RF measurement systems. Several extraction methods using the entire spectrum measured by the system are presented to extract non-desensitized pulsed-RF S-parameters of a transistor. The comparison of the calculated S-parameters using the least-square fitting in time domain with those using only the fundamental tone reveals the significant increase in dynamic range achieved by the proposed measurement scheme. Index Terms—Pulsed measurement, large signal network analyzer (LSNA), desensitization.

S

I. INTRODUCTION

LOW memory effects are low frequency dispersions originating from various physical processes in transistors such as self- heating, parasitic bipolar transistor in FETs, and traps[1]. It is essential for successful device modeling to account for these slow memory effects in the acquisition of the DC and RF device characteristics. To account for these slow memory effects, pulsed-IV pulsed-RF measurement techniques have been developed to reproduce the conditions under which a transistor operates for large RF signals [2]. Low-frequency memory effects have indeed a slow time response and the fast pulsed-IV pulsed-RF measurements maintain a constant temperature, body BJT voltage or trap state in FETs like when there are excited by large RF signals. By bypassing slow memory effects, pulsed-IV pulsed-RF measurements enable us to obtain realistic isothermal RF characteristics of the FETs. Most pulsed-RF measurement systems using conventional network analyzers can only acquire the center tone of the pulse spectrum and this results in a significant loss of dynamic range due to the resulting desensitization of 20log(duty_rate). Typically a 1µs duration pulse with 1% duty rate is used to avoid low-frequency memory effects [3], and the dynamic range of the network analyzer for the pulse decreases by 40dB. The reduction in dynamic range is an important problem which affects the accuracy of the pulsed measurement data obtained, especially for low duty rate pulse. Since the power provided by the center-tone does not reflect the actual signal RF power applied during the pulse, it is desirable for more tones in the RF pulse to be included into the characterization of the DUT. Fortunately, it is possible to get a large portion of the spectrum of the pulse signal with a large

signal network analyzer (LSNA). Therefore the LSNA can be expected to improve the dynamic range in the pulsed-IV pulsed-RF measurement system [4]. The goal of this paper is to introduce the new pulsed-IV pulsed-RF measurement system developed using the LSNA, and present the analysis used to recover the S-parameters from the data measured.

II. PULSED-IV PULSED-RF MEASUREMENT SYSTEM A. System Realization In general in pulsed-IV systems both Vgs and Vds pulsed biases are applied at the gate and the drain of a transistor. In the experiment reported here, a pulse is only applied at the drain and a constant DC voltage is used at the gate. It is assumed that slow memory effects are dominated by self-heating and the contribution of traps in the gate is neglected. For this entire study, we have used the Infineon CLY 5 GaAs FET as a DUT. Fig. 1 shows the pulsed-IV pulsed-RF measurement system implemented using the LSNA. Port 1 is used for the gate, and port 2 for the drain. The constant Vgs is applied to the gate using the LSNA internal bias tee. An external bias tee is used for the drain. A current sensor consisting of a resistor is used to measure the drain current with an oscilloscope. A FPGA digital testbed synchronized to the LSNA [5] was initially used to synthesize a band-limited pulsed-RF signal using an IQ modulator. This permitted to address the fact that the modulation bandwidth of the LSNA is presently limited to 20 MHz. However the 20 MHz bandwidth limitation is not a major concern for a 0.33µs duration pulse. The final approach adopted is to use the pulsed RF signal provided by a RF signal

Large Signal Network Analyzer RF Signal Generator

PORT1

RF

(RF)

Pulse

Vgs Power Supply

Trigger

PORT2

DUT

RF+DC

Bias Tee

Vds Pulse Generator

Pulse

Current Sensor

Fig. 1. Pulsed-IV pulsed-RF measurement system.

Osilloscope

stable. Note that 104 pulses are acquired by the LSNA when using 95 Hz resolution. Fig. 2 shows a generated RF pulse with a duty rate of 0.33% and 0.33µs pulse duration. In addition to generate a RF pulse, the RF signal generator is also triggering the Vds pulse generator to overlay the RF pulse upon the drain pulse signal at the proper moment. More details on the timing for the pulsed-IV pulsed-RF signals are provided in Fig. 3. The drain pulse has a duty rate of 1% and 1µs duration which is short enough to achieve the targeted isothermal condition.

(a) Sinc function in frequency domain (magnitude: white, phase: red). The main lobe consists of 601 tones with frequency spacing of 9.918KHz.

B. Measurement Bandwidth To remove the problem of desensitization in conventional pulsed-RF system, it would be theoretically desirable to use the entire spectrum of the pulse to obtain higher measurement accuracy. In practice, however, it is not possible to acquire the complete spectrum due to the 20MHz IF bandwidth limitation of the LSNA. Nonetheless by acquiring the spectrum in a wide enough range we can enhance the measurement accuracy. According to Parseval’s theorem, the total average power of the signal x(t ) is the sum of the average power in each harmonic component: 1 T

T

∫

∞

2

x(t ) dt =

∑C

2 n

∞

∑C

=2

n = −∞

0

2 n

.

(1)

n =0

where C n is Fourier coefficients, and usually has the form of a (b) RF pulse in time domain (pulse width: 0.33µs, duty rate: 0.33%) Fig. 2. LSNA pulse measurements at modulation frequency of 9.918KHz and resolution BW of 95.367Hz. With this setup the LSNA captures 104 pulse periods for each measurement.

sinc function if x(t ) is a pulse signal. Thus the power included in the bandwidth range m is m

power (%) =

∑C n =1

2 n

∞

∑C

2 n

× 100

(2)

n =1

Fig. 4 shows that the main lobe of a sinc function includes about 90.3% of the total average power, providing enough power to essentially suppress the desensitization experienced in the conventional pulsed-RF system. It is also preferable to focus only on the main lobe because of the noise floor level of

Fig. 3. Timing for the pulsed-IV pulsed-RF system. Signals (a) and (b) indicate pulses between the input and output of the current resistor sensor for the drain (Fig.1), with a pulse width of 1µs and a duty rate of 1%. The RF signal (c) which is applied to the gate at the measurement time from 0.6µs to 0.93µs, has a 0.33% pulse duty rate.

generator (Anritsu MG3692A) as shown in Fig. 1. The pulse-modulated signal from this signal generator is very

Fig. 4. As the number of single-sideband (SSB) tones in a sinc function increases, the included power also goes up. With only main lobe (m=300), it is possible to obtain the 90.3% of total average power (duty rate: 0.33%, duration: 0.33µs).

the LSNA. In other words, with resolution bandwidth of 12KHz the noise floor level of the LSNA is typically -70dBm up to 20GHz when the IF frequency is 10MHz [6], and deep nulls of the sinc function are usually expected to be below the noise floor (see Fig.2 (a)).

III. S-PARAMETER ANALYSIS In this work the system is used to acquire the pulsed-IV pulsed-RF S-parameters of a RF transistor. For each of the 601 tones ( i ) in the pulsed RF spectrum as shown in Fig. 2 (a), an S-parameter can be defined as

Fig. 7. Amplitude and phase of S11(ωi) varying tone index. These data are fitted using the least-square algorithm to recover S11(ω0).

S kl (ωi ) = bk (ωi ) al (ωi ) | (a m≠l = 0) .

(3)

Typical S-parameters resulting for all ω i in the center lobe are shown (green dots) in Fig. 5.

Fig. 5. Distribution of S-parameters (green dots) for frequencies between 0.97GHz to 1.03GHz for a single pulsed RF measurement at 1 GHz. The blue ‘+’s indicate the S-parameters of the center tone (at 1GHz) in the pulse, while the red ‘o’s represent the estimated S-parameters recovered by the below leastsquare fitting algorithm in time domain using all tones in the 601 frequency range specified.

Note that S11 (ω 0 ) and

S 21 (ω 0 ) which are obtained by using only the fundamental center tone (i=0) will naturally be noisier due to the desensitization. In Fig. 6, it is shown that the dynamic range of the center tone decreases by about 50dB. To estimate the pulse-RF S-parameters based on all the 601 measured tones in the spectrum, three different approaches are considered.

A. Least-square fit in frequency domain Because the raw S-parameters S kl (ω i ) are usually noisy, an estimation of the trend of the S-parameters is needed. It is reasonable to assume that the S-parameters are basically linear functions of the frequency within the bandwidth of the main lobe. Least-square fitting in frequency domain is then used to find the best fitting straight line. The straight lines fitting the amplitude and phase of the S-parameter S kl (ω i ) are then used to predict S kl (ω 0 ) as shown in Fig. 7. However, as we move away from the center tone in Fig. 7, Sparameters are getting noisier due to the sinc nature of the spectrum of |a1|. To reduce the impact of the increased noise on the edge, the least-square algorithm is applied only to a reduced bandwidth corresponding to 401 tones. B. Envelope convolution in time domain

Fig. 6. 5-consecutive incident waves for a continuous wave and a pulse signal. The RF input power is 5dBm. In the case of the pulse signal (duty rate: 0.33%), the center tone at 600MHz has only -45dBm power, and indicates the 50dB desensitization.

The signal envelope E[ ] of an arbitrary signal x(t ) can be recovered from the amplitudes and phases in the spectrum using E[ x(t )] = I 2 (t ) + Q 2 (t ) ,

(4)

Fig. 8 shows the recovered envelopes and their convolutions. The times at which the convolutions E[b1 (t )] * E[a1 (t )] and E[b2 (t )] * E[a1 (t )] reach their peak gives the group delay

τ S11 = d∠S11 dω i and τ S21 = d∠S 21 dω i of the envelopes of b1 (t ) and b2 (t ) relative to the envelope of a1 (t ) . In addition

the ratios of E[b1 (t )] * E[a1 (t )] and

E[b2 (t )] * E[a1 (t )]

relative to E[a1 (t )] * E[a1 (t )] give the magnitudes of S11 and S 21 , respectively.

Fig. 8. The top figure shows the envelope signals E[a1(t)], E[b1(t)] and E[b2(t)] of the pulsed RF-signals a1(t), b1(t) and b2(t) which are reconstructed from 601 tones spectrum measured. The bottom figure shows the convolution of E[a1(t)], E[b1(t)] and E[b2(t)] with E[a1(t)].

where I(t) and Q(t) are in-phase/quadrature-phase components of the signal x(t ) . After getting the envelopes E[a1 (t )] , E[b1 (t )] and E[b2 (t )] , subsequent convolutions between these signals can be used to extract the device group-delay (τ) and the amplitude of Sparameter using the following equations.

τ S kl = d∠S kl dω i

(5)

S kl (ω 0 ) = E[bk (t )] * E[al (t )] E[al (t )] * E[al (t )]

(6)

50dB desensitized signal

Fig. 9. Reconstructed RF signals a1(t), b1(t) and b2(t) for times corresponding to the center of the pulse. The fundamental tones which are desensitized due to the pulsing, yield weak signals (dashed lines) with relatively negligible amplitudes. However the reconstructed signals using all the 601 tones in the main lobe yields much stronger signals less sensitive to noise.

C. Least-square fit in time domain The RF signals in time domain can be reconstructed from the amplitudes and phases of the RF pulse spectrum. Fig. 9 shows the reconstructed RF signals a1 (t ) , b1 (t ) and b2 (t ) using Eq. (7). It is obvious that the reconstructed signals (plain line) in the center of the pulse are much stronger than those of the desensitized center-tone signal (dashed line). This results from the fact that the reconstructed signals using 601 tones are keeping about 90.3% of the original average power. Note that the decreased dynamic range for a pulse with a duty rate of 0.33% is of about 50dB compared to a CW tone. A least-square fitting into cosine and sine functions in the time domain of a1 (t ) , b1 (t ) and b2 (t ) is used to extract the I and Q components of the signal x(t ) : x(t ) = I cos(ω 0 t ) − Q sin(ω 0 t )

(7)

This permit us to calculate in turns the amplitude and phase of S kl (ω 0 ) . As a result of the larger signal amplitude, a reduced noise is expected in the S-parameters. In addition to considering the above three analyses, additional processing steps are required for extracting S-parameters from LSNA measurements. In distinction to a conventional network analyzer, the LSNA calibration is not used for transforming the termination at port 1 and 2 into perfect match loads. It results that the small reflections from the coupler and terminations introduce reproducible oscillation features in the obtained S-parameters if port 1 and 2 are assumed to be perfectly matched. To remove these unwanted features one can alternately send an excitation at port 1 and 2. Obviously these two consecutive measurements must be performed for the same pulsing conditions. Then solving the four simultaneous equations obtained from Eq. (8) for both measurements, gives the error-corrected S-parameters. ⎛ b1, p = S11a1, p + S12 a 2, p ⎞ ⎟ ⎜ ⎜ b2, p = S 21a1, p + S 22 a 2, p ⎟ ⎠ ⎝

(8)

where a and b are incident / reflected components that are reconstructed using the least-square fitting in time domain, and p indicates the primary port (1 or 2) excited .

IV. EXPERIMENTAL RESULTS A. Pulsed-IV characteristics Fig. 10 shows the measured pulsed-IV and DC IV characteristics for the GaAs FET. It is clear that due to self-heating the DC IV characteristics include the negative drain conductance at high bias. Typically the FET used has a maximum thermal resistance ( Rth ) of 35°K/W, and the observed maximum temperature ( Tdev ) is 93°C for the gate-source voltage of 0V. But, on the other hand, the pulsed-IV characteristics bypass low-frequency dispersions as a result of traps and thermal effects. Based on the pulsed-IV characteristics, a DC quiescent point with Vds = 3V and I ds = 350mA is used for S-parameter measurements. (a) S11 and S21

Fig. 10. Comparison of pulsed-IV and DC IV characteristics for Vgs from -2V to 0V in steps of 0.25V. DC IV curves are obtained by a Keithley 4200 semiconductor characterization system, and the developed pulsed-IV system uses a pulse signal with a duty rate of 1.0%, a duration of 1µs for the drain with a constant substrate temperature of 25°C. The load-lines for different RF power levels at 600MHz are also shown.

B. Comparison of S-parameters Fig. 11 compares the S-parameters obtained using four different methods: 1) Using only the center tone in the pulses. 2) Using the main lobe in the pulses, and by - applying a least-square fitting in frequency domain. - applying a least-square fitting in time domain. 3) Using a continuous wave for a single tone signal. These data are all averaged over 5 consecutive measurements. As expected, the S-parameters obtained using center tones show bigger distortion due to the desensitization. The S-parameters given by the other methods are smooth functions of frequency because both for the continuous wave and for the pulse wave recreated from main spectrum lobe, the measurements use 100% and 90.3% of the average CW and peak pulse power, respectively. However, note that the DC IV

(b) S12 and S22 Fig. 11. Comparison of S-parameters for the RF input power of -10dBm in a frequency range from 600MHz to 2.6GHz with a 10MHz step frequency.

(a) S11 and S21

(b) S12 and S22 Fig. 12. S-parameters with the increased RF input power

characteristics in Fig. 10 are submitted to strong low-frequency dispersion due to self-heating. To approximately compensate for this self-heating effect in the continuous wave measurement of the S-parameters, the gate voltage was arbitrarily adjusted (by about 0.2 V) to obtain the same DC drain current as in the pulsed-IV measurement. Fig. 11 also shows that the least-square fitting in time domain is better than the fitting in frequency domain. In addition, note that the small loops observed in S22 in Fig. 11(b) are reproduced in 5 subsequent pulsed RF measurements. This reproducible feature indicates that these loops most likely arise either from the bias-tee deembedding or a residual calibration error. To get satisfactory results from center tone extraction, one may need to increase the output power of the RF signal generator to compensate for the desensitized power at the risk of introducing non-linear effects (see self-biasing shift of load-lines in Fig. 10). Fig. 12 shows the S-parameters for two different RF input powers. As the RF input power increases, the S-parameters obtained from the center tone data are notably stabilized. But even the stabilized S-parameters for the RF power of 5dBm show more distortion than those extracted by least-square fitting in time-domain from the main lobe for an RF power of -10dBm. It is verified in Fig. 13 that the S-parameters obtained by least-square fitting in time-domain show good consistency for a wide range of RF input power. This is largely due to the fact that the family of tones in the main lobe of the sinc spectrum provide enough energy to calculate accurate S-parameters. This measurement benefits also from the fact that the LSNA makes it possible to measure very small signals with power around -70dBm. V. CONCLUSION

(a) S11 and S21

A pulsed-IV pulsed-RF measurement system using an LSNA is implemented to overcome the problem of desensitization. It is confirmed that by using a large number of tones in the spectrum one can reduce the noise compared to using the center tone only, yielding a more reliable small-signal measurement of the DUT. This measurement setup could also be easily modified for the acquisition of pulsed-RF S-parameters for pulsed-gate voltage to study trap effects in GaN devices. Finally this LSNA-based pulsed-RF system makes possible the acquisition of the pulsed-RF harmonic (nω0) response of the device under various load terminations. In that regard such non-linear applications would follow in the foot steps of the work done at IRCOM [7] except that the measurement scheme used in our approach does not require any modification of the LSNA hardware. ACKNOWLEDGMENT

(b) S12 and S22 Fig. 13. S-parameters obtained by least-square fitting in time domain for three different RF input powers

We would like to thank Jean Pierre Teyssier of IRCOM for fruitful discussions on pulsed RF measurements. We are grateful to Dr. Gerald Witt of AFOSR and the State of Ohio for the support they provided to the development of the Non-Linear RF Laboratory at OSU.

REFERENCES [1] [2]

[3] [4] [5]

[6] [7]

P. Roblin and H. Rohdin, High-Speed Heterostructure Devices-From Device Concepts to Circuit Modeling. Cambridge University Press, 2002, ch. 13. J.-P. Teyssier, P. Bouysse, Z. Ouarch, D. Barataud, T. Peyretaillade, and R. Quere, “40-GHz/150-ns Versatile Pulsed Measurement System for Microwave Transistor Isothermal Characterization," IEEE Transactions on Microwave Theory and Techniques, Vol. 46, No. 12, pp. 2043-2052, Dec. 1998. J. Scott, J. G. Rathmell, and A. Parker, “Pulsed device measurements and applications,” IEEE Transactions on Microwave Theory and Techniques, Vol. 44, pp. 2718-2723, Dec. 1996. P. Vael, Y. Rolain, C. Gaquiere and H. Gerard, “Calibrated linear and nonlinear pulsed RF measurements on an amplifier,” 2001 MTT Digest, pp. 2187-2190, 2001. S.K. Myoung, X. cui, D. Chaillot, P. Roblin, R. Verbeyst. M.V. Bossche, S. J. Doo, and W. Dai, “Large signal network analyzer with trigger for baseband & RF system characterization with application to K-modeling & output baseband modulation linearization,” ARFTG 64 Conference Digest, Orlando, Dec. 2004 Large-Signal Network Analyzer System Manual, NMDG Engineering, 2004 J.-P. Teyssier, S. Augaudy, D. Barataud, J-M Nébus, R. Quéré, “Large-signal time domain characterization of microwave transistors under RF pulsed conditions, ” 57th ARFTG, Phoenix, Arizona, May 2001.