IEEE BCTM 1.4

MEMS-Based Reconfigurable Multi-band BiCMOS Power Amplifier A.J.M. de Graauw, P.G. Steeneken*, C. Chanlo, J. Dijkhuis, S. Pramm, A. van Bezooijen, H.K.J. ten Dolle, F. van Straten, P. Lok. Philips Semiconductors, ICRF, Gerstweg 2, 6534AE Nijmegen, The Netherlands *Philips Research Laboratories, Prof. Holstlaan 4, 5656AA Eindhoven, The Netherlands Abstract — This paper presents a small dual-band 0.9GHz /1.8GHz inverse class F power amplifier with load-switch functionality using a single BiCMOS amplifier line-up with a MEMS based reconfigurable matching network. The realized prototype measures 40mm2, offers 31dBm with 40% efficiency at 0.9GHz and 30dBm with 34% at 1.8GHz. The load-switch provides up to 10% efficiency improvement at 0.9GHz for reduced power levels. Index Terms — RF power amplifier, Reconfigurable Matching Networks, BiCMOS, MEMS.

I. INTRODUCTION Future mobile handsets are expected to offer increased functionality by providing access to multiple cellular-, connectivity- and broadcastwireless networks. The diversity of used frequency bands and modes will result in a dramatic increase in the complexity of the RF functionality of the phone. A critical function in the RF section is the power amplifier (PA). The PA should be efficient as it directly affects the talk-time of the phone. In addition it has to be sufficiently linear to meet the spectral requirements of the modulation standards. Finally, it has to be small in order to fit in the continuously shrinking size of the RF part of the phone. The conventional concept for a multi-band/mode PA is the use of a dedicated PA for each service which is optimized with respect to the band and mode under consideration. A disadvantage of this approach is that operation in n-frequency bands requires namplifier line-ups resulting in a cost and size of the PA function that increases linear with the number of bands. The concept presented here uses a single PA line-up in combination with a reconfigurable output matching network to cover multiple frequency bands/modes. This approach results in size and cost advantages compared to the conventional approach due to the re-use of circuits as shown in Fig. 1. The use of PA’s with reconfigurable RF circuits to meet the multi-band/mode challenges is well known in the wireless industry and is currently an important research and development topic. Recent work has demonstrated the use of reconfigurable matching networks in PA’s for frequency-band switching [1,2] as well as load-line adaptation [3].

1-4244-0459-2/06/$20.00 ©2006 IEEE.

:

F1 F1

F2

Fn

Fn (a)

(b)

Fig. 1. (a) Conventional multi-band PA, (b) Single LineUp reconfigurable multi-band PA .

The RF performance is however limited in those concepts to optimization of the PA load impedance at the fundamental frequency. High efficiency PA operation classes like (inverse) class D,E and F require in addition optimum loading at higher harmonics and proper suppression of these harmonics at the output to meet the spectral requirements. This work presents a concept that offers optimization of efficiency and linearity for each frequency band by reconfiguration of the output matching network using switchable RF-MEMS capacitors. The new output matching network topology offers highly efficient class F-1 amplifier operation in two frequency bands by optimizing both the fundamental and harmonic impedance levels in combination with strong suppression of those harmonics at the output of the amplifier. In addition, the network also offers the possibility to optimize the fundamental load impedance as function of the modulation type and power level. The concept was demonstrated by a 40mm2 prototype which offers 31dBm with 40% efficiency at 0.9GHz and 30dBm with 34% efficiency at 1.8GHz. A build-in loadswitch provides up to 10% efficiency improvement at 0.9GHz for reduced power levels. II. RECONFIGURABLE CLASS F-1 AMPLIFIER The optimum load impedance for the output stage for class F-1 operation is given by [4]:

(π .Vc ) 2 (8 . Psat ) Zl ⎯ ⎯→ ∞ Zl ⎯ ⎯→ 0

Zl =

for

F = Fo

for all even harmonics for all odd harmonics

(1)

IEEE BCTM 1.4

where Zl is the collector load impedance, Vc the DC collector voltage and Psat the maximum RF output power. The conditions stated in (1) result in a square wave collector current waveform and a half sinusoidal collector voltage waveform without overlap in the time domain corresponding to no power dissipation in the transistor. The theoretical peak efficiency of class F-1 operation of 100% is considerable better than the theoretical value of 78.5% for class B operation but requires a more complex network. An efficiency of 90% can however be achieved by providing the optimum impedance only at the second and third harmonics, fig. 2 shows the wave-shapes for that case. U [V]

I [A]

III. RECONFIGURABLE OUTPUT MATCHING NETWORKS The requirements on the load impedance which has to be realized by the reconfigurable output matching network for dual-band class F-1 operation can be summarized as: a. Real valued load impedance at the fundamental frequency which depends upon the RF power. b. Open circuit at second harmonic- and short circuit at third harmonic-frequency band. Fig. 4 shows a basic network topology in which a single switchable capacitor is used to switch the fundamental load impedance between two real values. Zlow Zhigh

Zl C C

t [ns] Fig.2. Simulated Class F-1 collector voltage and current wave-shape with optimum H2, H3 termination (ηc=90%).

The highly reflective terminations at the second and third harmonic frequency bands result in a low level of the transfer of harmonic power to the antenna terminal and is desired for a clean output spectrum. The collector efficiency of a class F-1 stage is a function of the output power level and is given by:

ηC =

Pload Psat

(2)

(a)

(b)

Fig.4. (a) Basic load-switch network (b) Simulated Load-impedance (Zo=50Ω).

topology.

Fig. 5 shows a basic circuit topology for the realization of the desired harmonic terminations. Band switching is obtained by a single switched capacitor which tunes the harmonic trap L,C between the second and third harmonic of the low frequency band. ZH3_LB&HB

where Pload is the actual power and Psat the maximum available power, both with Zl given by (1). (2) shows that the efficiency drops with the square root of the actual load power. Figure 3 shows the relation between efficiency, power and load resistance for a typical 0.9GHz GSM PA which is designed to offer a maximum power of 35.5dBm at Vcc=3.0V using a nominal load-line of 3Ohms. The efficiency at reduced power levels can be restored to the peak value by adjusting the load impedance such that it doubles for every 3dB reduction in power. ηc [%] 12Ω

6Ω

3Ω

Pl [dBm] Fig.3. Simulated relation between collector efficiency and power level as function of load impedance level.

ZH2_LB&HB

Zl C L H2_HB H2/3_LB (a)

Zfund_LB Zfund_HB (b)

Fig.5. (a) Basic network topology for dual-band harmonic loading. (b) Simulated Load-impedance (Zo=10Ω).

A combination of the topologies shown in fig. 4 and 5 can be used to meet both requirements a. and b. mentioned above in a dual-band 0.9GHz/1.8GHz matching network for class F-1 operation including 0.9GHz load-switch functionality. Figure 6 shows the overall network topology with 3 MEMS used for band switching and 1 for loadline switching.

IEEE BCTM 1.4

Vbat

load_switch

band_switch

Fig.6. Dual-band reconfigurable output matching network with optimized second and third harmonic loading and loadswitch functionality.

Table 1 shows the three operating states of the network: low-band maximum power, low-band reduced power and high-band. State LB, 33dBm LB LS, 30dBm HB, 30dBm

Zl 3Ω 6Ω 5Ω

load_switch high low low

band_switch high high low

Table 1. Operating states of reconfigurable output matching network.

IV. IMPLEMENTATION The amplifier was implemented as a System in a Package (SiP) to allow for an optimum partitioning of the functionality with respect to cost, size and performance. The SiP measures 40mm2 and consists of a stack of an organic laminate and a high-Ohmic silicon substrate (PASSI) with flipped BiCMOS (QUBiC) and MEMS on top as shown in Figure 7. MEMS

LAMINATE PASSI ABCD QUBiC

The high Ohmic Silicon crystal serves as a carrier for the flipped crystals on top and as a heat spreader /interposer between the flipped dies and the laminate converting the fine pitch (90um) of the dies to the coarse pitches (>300um) used on laminate. The silicon is also used for integration of high Q MIM capacitors (C=145pF/mm2), a high density MIS capacitor (C=25nF/mm2) and Poly-silicon resistors (R=10Ohm/square). The RF LineUp is realized in a 0.25um BiCMOS technology with special High Voltage NPN device for PA applications (Ft=26GHz, BVCBO=18V). The MEMS switched capacitors are made in a low cost thin film passive integration technology on high Ohmic Silicon [5]. This technology offers capacitors with a high switching ratio (Con/Coff >10) in combination with low parasitic losses (ESR < 0.4 Ω). The charge pump used for actuation of the MEMS is realized in a 1.2um SOI process (ABCD) with a High Voltage (BV> 60V) DMOSFET device. V. MEASUREMENTS The output match was evaluated separately as well as combined with the BiCMOS active line-up. Fig. 8 shows the input impedance of the matching network for the network states defined in Table 1. The lowband results show that the load-impedance in the pass-band is doubled from 4Ω to 8Ω if the network is switched from the nominal to the load-switch state. The impedance in both low-band states is high at the second and low at the third harmonic frequency band. The high-band results show a nominal pass-band impedance of 4Ω again in combination with the desired second- and third-harmonic termination for class F-1 amplifier operation.

MEMS2 PA

(a)

m1=0.95GHz m2=1.90GHz m3=2.70GHz

(c)

m7=1.6GHz m8=3.6GHz m9=5.4GHz

(b)

m4=0.95GHz m5=1.90GHz m6=2.70GHz

Fig.7. RF-SiP using stack of laminate, high-Ohmic Silicon and flipped BiCMOS and MEMS dies.

The laminate serves as a module carrier and as a heat spreader/interposer between the Passive die and the phone-board. It offers the interconnect to the phone board by means of an Land Grid Array (LGA) on the backside. The laminate is also used for integration of high Q inductors and coarse interconnects using two 25um thick Copper layers.

Fig.8. Measured input impedance of output matching network (Zo=10Ω): (a) LB, 33dBm, (b) LB LS, 30dBm, (c) HB, 30dBm.

IEEE BCTM 1.4

The pass-band losses and harmonic filtering properties of the matching network were evaluated by measuring the frequency transfer using a source impedance equal to the conjugated input impedance. The results are shown in Fig. 9. The low-band results shows about 1.5dB pass-band loss with over 40dB attenuation in the second and third harmonic frequency band. The high-band pass-band loss is about 1.5dB with over 30dB second and third harmonic attenuation.

efficiency improvement at 0.9GHz for reduced power levels. η [%]

60

40 30

HB

20 10 0 24

S21 [dB]

LB LS LB

50

26

28

30

32

Pl [dBm]

HB

Fig 11. Measured power efficiency versus output power as function of network state.

LB

VI. CONCLUSION

Freq [GHz] Fig.9. Measured pass-band loss and harmonic rejection of matching network in LB- and HB-state with matched source at input terminal.

The linearity of the matching network has been evaluated by exciting the network with an 8PSK (EDGE) signal from a conjugated matched source and measuring the EVM and ACPR of the output signal. The low-band results are shown in Fig. 10. Both EVM and ACPR are well below the typical PA requirements of 4% and -57dBc at power levels up to 33dBm. This demonstrates the good linearity of the applied MEMS-based switched capacitors. EVM [%]

ACPR [dBc]

This work demonstrates the feasibility of the use of a MEMS-capacitor based reconfigurable output matching network for efficient dual-band operation of a single BiCMOS amplifier line-up with EDGE compliant linearity. In particular it was demonstrated that a new reconfigurable matching network topology offers proper wideband termination for efficient class F-1 amplifier operation in both bands in combination with low-band load-switch functionality. The demonstrated amplifier concept offers cost, size and performance advantages for multi-band/mode handset PA’s, a advantage that increases with the number of bands/modes covered.

ACKNOWLEDGEMENT The authors wish to acknowledge the support of Philips Research and the ICRF Concept and Technology development teams of Philips Semiconductors. REFERENCES

Pl [dBm] Fig.10. Measured EVM and ACPR (400kHz) of matching network in LB-state when excited at 0.9GHz with matched 8PSK (EDGE) source at the input terminal.

The full amplifier consisting of the BiCMOS active line-up connected to the MEMS-based reconfigurable matching network has been evaluated with respect to output power and efficiency. Fig. 11 shows the overall efficiency versus output power for the three states defined in Table 1. The 0.9GHz results shows a maximum power level of 31dBm with 40% efficiency while at 1.8GHz a maximum power of 30dBm has been achieved with 34% efficiency. The build-in load-switch provides up to 10%

[1] A. Fukada, “Novel 900MHz/1.9GHz Dual-Mode Power Amplifier Employing MEMS Switches for Optimum Matching”, IEEE Microwave and Wireless components Letters, Vol 14, No 3, 3 March 2004 pp.121-123. [2] Y. Lu, “A MEMS Reconfigurable Matching Network for Class AB Amplifier”, IEEE Microwave and Wireless components Letters, Vol 13, no 10, Oct 2003 pp.437-439. [3] X. Liu, Y. Lin, W.C.E.Neo, L.C.N. de Vreede et al, “Improved Power Amplifier Efficiency Using Varactor-Based Tunable Matching Networks”, Proceedings BCTM 2006. [4] F.H. Raab, “Maximum Efficiency and Output of class F amplifiers”, IEEE Transactions on Microwave Theory and Techniques, Vol. 49, no 6, June 2001 pp.1162- 1166. [5] J.T.M. van Beek et al, “High-Q Integrated RF Passives and micro-mechanical Capacitors on Silicon”, Proceedings BCTM 2003, pp.147-150.