INTERNATIONAL JOURNAL OF ELECTRICAL, ELECTRONICS AND COMPUTER SYSTEMS (IJEECS), VOLUME 1, ISSUE 1 MARCH 2011 WWW.IJEECS.ORG ISSN: 2221-7258(PRINT) ISSN:2221-7266(ONLINE)

Accurate power measurement of high power GaN devices for W-CDMA base station application Armin Liero, Roland Gesche Ferdinand-Braun-Institut für Höchstfrequenztechnik (FBH), Gustav-Kirchhoff-Str. 4, D-12489 Berlin, Germany.

Abstract A method of accurate power measurement is presented here based on a load and source pull measurement system. The measurement was carried out for a packaged high power GaN powerbar fabricated and packaged in FBH, Berlin. Commercial source and/or load pull measurement systems for packaged transistors use a standard test fixture where the device is placed with the tuners located outside the periphery which introduce in many cases a serious problem of inaccurate results due to high impedance mismatch at the input and at the output of the device. The method presented here is a simple pre-matching on the test fixture. The prematching network is designed on basis of small signal S-parameter measurement. Load and sourcepull measurement is then done with the combination of prematching network and tuner. Final transformation networks were built based on large signal input and output impedances. Keywords: Mobile communication, W-CDMA, Power bar, Power measurement, Loadpull, Sourcepull, Device characterization etc.

whole device. The second reason is that the bond wires and the package itself may present a strong transformation at input and output. The low impedance presents a tuning problem for the load and source-pull measurement due to the fact that the impedances are either outside the range of the tuner, or where the tuner accuracy degrades [3]. The tuner losses increase with the increasing reflection coefficient seen from the input and output of the transistor toward the tuners. As a result its difficult to estimate the accurate power delivered from the device. It was observed that after measuring the whole packaged device in a standard commercial test fixture, expected power was not available. This paper presents a method of investigation and improvement of power measurement. Small signal S-parameter measurement was done for the packaged device on an accurate calibrated fixture. This is used to estimate input and output impedance of the packaged device. Then S-parameter data was used to design a pre-matching network which improve the measured output power drastically to reach expected power level.

I. INTRODUCTION Modern mobile telecommunication systems like WCDMA or UMTS standard demands low cost high power broadband devices. Due to very unique characteristic of GaN, its possible to make very high power devices more compact than ever. These high power devices must be characterized before they can be used for base station amplifiers. Wide bandgap GaN offers a very high break down field and relatively lower intrinsic carrier generation than GaAs or Si [1]. With these characteristic GaN devices can operate under high drain voltage and therefore high output impedance level [2]. To deliver very high output power from a single chip, so called power bars are designed. A power bar consists of multiple single transistors fabricated in a column and then bonded together to form a single chip. Normally the single devices of the power bar are characterized on-wafer individually before packaging. But after packaging, it is very important to characterize it as a single chip specially due to two reasons. Firstly as a power-bar consists of many devices bonded together in parallel, these parallel connected devices results in a very low impedance both at output and input of the

II. TUNER CHARECHTERIZATION & LOSS CALCULATION The investigated GaN power-bar consists of eleven single cells bonded on a standard commercial package. Before packaging on-wafer load-pull measurements were done for each of the cells. 50Ω Tuner

[SR]

Pq

Line2

Line1

[S(2)]

[S(1)]

Tuner

[ST]

DUT P0 Γ0

P1 Γ1

Pg Γg

Pin Γin

PT ΓT

Pout ΓL

Fig. 1 Schematic diagram of the load and source pull measurement system with tuners.

From this standard on-wafer measurement the whole power bar was estimated to deliver about 40 watts of power. The packaged device was measured with a load and source pull on a commercial test-fixture. The testfixture is a 50 Ohm input and output impedance system.

Tuners are connected after the test fixture as shown in Fig. 1. The packaged power bar was operated with Vds = 20 Volt at and Ids=2.3 Amps. It was only possible to avail a little better than half of the expected power level with the combination of all possible tuner positions. But from this measurement it was noticeable that the tuners present the highest reflection coefficients at their best i.e, above 0.8. Tuners were characterized to estimate the loss. Fig. 2 shows the tuner loss in relation to reflection coefficient of the tuner Γ. It can be seen that the tuner loss is very high above the reflection coefficient 0.8 and has a very steep shape. The figure shows that loss at Γ=0.9 is about 0.87dB and at Γ=0.92 its about 1.3dB. As the tuners are not so precisely tunable its almost impossible to determine the accurate loss.

used. The test fixture was calibrated accurately using TRL method. It is important to do a TRL or LRL calibration due to the fact that the S-parameter measurement inaccuracy or uncertainty remains reasonable for high reflection coefficient even about 0.9 where the other calibration methods such as SOLT or sliding load shows a very high uncertainty [3]-[4]. As our device expected to show very low input and output impedance a TRL calibration is done. Input and output impedance of the device based on S-parameter measurement is shown in fig. 3. It clearly shows the load and source pull target at the region where the tuners are very sensitive and even may not be able to transform properly.

1.6 1.4

Loss [dB]

1.2

S11

1

S22

0.8

Source Prematch

0.6

Load Prematch

0.4 0.2

(a)

0 0

0.2

0.4

0.6

0.8

1

Input reflection coefficient

Fig. 2 Tuner loss for different input reflection coefficients.

The total loss of the input and output network that is the test fixture and tuner can be calculated using the following formula. Total loss in the input network:

S

(1) 2 21

S

 1−  

21

Γ

1−

(R) 2

Γ

2 0

  

2 g

1− S Γ

2

R

22

(b)

Fig. 3 Pre-matching of (a) the input impedance from 4.37 Ohm ∠67.953 to 50 Ohms and (b) the output impedance from 6.232Ohm∠50.995 to 50 Ohms

Quarter wavelength transformers were designed as prematching networks. This means a lower impedance transmission line directly at the device which matches input or output impedance of the transistor to 50 Ohm. The impedance of the quarter wavelength transformer can be calculated by square root of the multiplied two impedances which are to be matched. Fig. 4 shows the on fixture pre-matching network.

(1)

1

And the loss in the output network: S

T 11

 1 −  

2

S

.

Γ

(2)

2 in

2

21

  

(2)

III. DESIGN OF PRETRANSFORMATION BOARD As measurement inaccuracy was predicted from the tuner loss calculation, an impedance change was planned to do stepwise. To do that properly it is necessary to know the approximate input and output impedance of the device. To estimate the impedance, Sparameter of the device was measured in a 50 ohm environment. To do this measurement the same commercial test fixture for load-pull measurement was

Fig. 4. Pre-matching board designed to replace 50 Ohm line.

Having a transmission line which is much closer to the impedance level of the transistor, the amount of loss decrease and the tuner is relaxed considerably to tune the device further more accurately. Using pre-matching

board and tuner connected out side the fixture, the measured power was increased drastically. Actual power delivered by the device and gain of the device can be calculated including the losses of the input and output network as shown in Eq. 1 and Eq. 2. In the following Eq. 3 and Eq. 4 the actual power delivered by the device and the gain of the device is shown.

Output Power, PAE [%]

Gain =

P P

m q

.

= Pm .

Power

 1−  

S

T

Γ

2

11

.

2 in

S

  1−    

(2) 2 21

.

S

Γ

.

11

2 0

(4) 2 21

2

T

S

Γ

.

  

S

2 in

S

  

(3)

(2) 2 21

1− S Γ

2

R

22

(R) 2 21

 1−  

1

Γ

2 g

  

50

15

40

13

30

11

20

9

10

7

0

(4)

Gain [dB]

Output

  1− 

5 0

10

20

30

40

Input Power [dBm]

Fig. 5 Measured output power (×) in dBm, (Ο) in Watt, (∆)PAE and gain in dB ( ) of the GaN powerbar.

IV. FINAL BOARD DESIGN & MEASUREMENT The tuner positions were recorded for the highest power delivered. The S-parameter of the whole transforming input and output network (the pre-matching board and the tuner) was measured separately. An equivalent network of this whole transformation network was designed to present appropriate transformation to the device. In this case, usually tuners are no more necessary except a fine tuning is wished. The tuner loss is avoided by excluding tuner or at least minimized (in the case of fine tuning) which results a very accurate power measurement. After designing the final board the power bar was measured to have the final result. Fig. 5 shows the output power, gain and PAE of the measured device. The measurement was done at 2 GHZ operating frequency. The measurement shows an maximum output

power about 40 watts with a gain about 14 dB and maximum PAE above 40%. IV. CONCLUSIONS An accurate power measurement was presented step by step for very high power low transition impedance devices. Although a single cell represent high impedance in case of a GaN device, parallel connection of many cells results a low impedance in case of powerbar. Therefore its very difficult to measure accurate output power with conventional methods. Tuners were characterized to determine the tuner loss at different tuner positions and it was shown that they are inaccurate in the required load and source pull target for such large power-bars. Therefore its necessary to relax the impedance level presented to such packaged devices. Finally a measurement based test fixture was designed to even avoid the use of tuners to have optimum output power. A clear improvement in output power was shown by using presented method.

REFERENCES [1] I. Daumiller, C. Kirchner, M. Kamp, K. J. Ebeling, and E. Kohn, “ Evaluation of the temparature stability of AlGaN/GaN heterostructure FET’s”, IEEE Electron device lett., Vol. 20, September 1999, pp. 448-450. [2] Jong-Wook Lee, Lester F. Eastman, “ A GalliumNitride push-pull microwave power amplifier”, IEEE transaction on microwave theory and techniques, Vol. 51, No. 11, November 2003, pp. 2243-2248. [3] Product Note 52, “Prematching Tuners for very high VSWR and Power Load Pull Measurement”, Focus Microwaves, 1999. [4] Engen, G.F.; Hoer, C.A, “Thru-Reflect-Line: An Improved Technique for Calibrating the Dual SixPort Automatic Network Analyzer”, IEEE Transactions on Microwave Theory and Techniques, Volume 27, Issue 12, Dec 1979 pp. 987 - 993. [5] Palmour, J.W.; Sheppard, S.T.; Smith, R.P.; Allen, S.T.; Pribble, W.L.; Smith, T.J.; Ring, Z.; Sumakeris, J.J.; Saxler, A.W.; Milligan, J.W, “Wide bandgap semiconductor devices and MMICs for RF power applications”, Electron Devices Meeting,, 2001. IEDM Technical Digest. International ,2-5 Dec. 2001 pp. 17.4.1 - 17.4.4.