156 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 40, NO. 1, JANUARY 2005 SiGe Bipolar Transceiver Circuits Operating at 60 GHz Brian A. Floyd, Member, IEEE, Scott K. Reynolds, Ullrich R. Pfeiffer, Thomas Zwick, Member, IEEE, Troy Beukema, and Brian Gaucher Abstract—A low-noise amplifier, direct-conversion quadrature mixer, power amplifier, and voltage-controlled oscillators have -GHz MAX been implemented in a 0.12- m, 200-GHz T SiGe bipolar technology for operation at 60 GHz. At 61.5 GHz, the two-stage LNA achieves 4.5-dB NF, 15-dB gain, consuming 6 mA from 1.8 V. This is the first known demonstration of a silicon LNA at V-band. The downconverter consists of a preamplifier, I/Q double-balanced mixers, a frequency tripler, and a quadrature generator, and is again the first known demonstration of silicon active mixers at V-band. At 60 GHz, the downconverter gain is 18.6 dB and the NF is 13.3 dB, and the circuit consumes 55 mA from 2.7 V, while the output buffers consume an additional 52 mA. The balanced class-AB PA provides 10.8-dB gain, 11.2-dBm 1-dB compression point, 4.3% maximum PAE, and 16-dBm saturated output power. Finally, fully differential Colpitts VCOs have been implemented at 22 and 67 GHz. The 67-GHz VCO has a phase noise better than 98 dBc/Hz at 1-MHz offset, and provides a 3.1% tuning range for 8-mA current consumption from a 3-V supply. 290 + Index Terms—Direct-conversion receiver, low-noise amplifier (LNA), millimeter-wave bipolar integrated circuits, mixer, power amplifier, SiGe, V-band, voltage-controlled oscillator (VCO), 60 GHz. I. INTRODUCTION M ILLIMETER-WAVE (MMW) integrated circuits are traditionally implemented using compound semiconductors such as gallium arsenide or indium phosphide [1]–[9]. Silicon-germanium (SiGe) technology may well begin to intrude on this territory, though, since the transition and maxof the most advanced imum oscillation frequencies SiGe bipolar transistors now exceed 200 GHz [10]–[14]. In the literature, one can now find examples of SiGe voltage-controlled oscillators (VCOs) operating between 60 and 100 GHz [15]–[20], SiGe low-noise amplifiers (LNAs), and mixers for 24 GHz [21]–[23] and 40 GHz [24], and a SiGe power amplifier for 77 GHz [20]. As ever, silicon holds the promise of high levels of integration, which should reduce cost and power dissipations. This may, in turn, enable new MMW applications and market opportunities. One promising millimeter-wave application is wireless communications in the 60-GHz industrial, scientific, medical (ISM) band. The 60-GHz ISM band features a large available bandwidth, with at least a 3-GHz overlap (59–62 GHz) for Europe, Japan, and the U.S. This enables high data-rate communications for wireless personal-area network (PAN) [25] or point-to-point applications [1], with possible data rates of at least 150 to 1000 Mb/s. Note that although this band undergoes attenuation due to oxygen absorption of 10–15 dB/km [25], this particular absorption is insignificant for short-range links. The long-range limitation imposed by path loss and wall attenuation [26] can instead be beneficial, since it improves frequency reuse and minimizes interference with other systems. An example of the performance already achieved at 60 GHz can be found in [1], where a 1.25-Gb/s 60-GHz directional link over 10 meters was achieved with simple ASK modulation. Antennas with 20–27 dBi gain were used with a transceiver assembly composed of five GaAs chips and two filters placed together in multi-chip modules. This paper describes key front-end 60-GHz SiGe transceiver circuits [27] designed and implemented individually as a first step toward realizing a fully integrated wireless PAN transceiver. The circuit blocks include a 60-GHz LNA; a 60-GHz direct downconverter, consisting of a preamplifier, I/Q double-balanced mixers, a frequency tripler, and a quadrature generator; a 60-GHz power amplifier (PA); and 60and 20-GHz VCOs. The circuits have been implemented in a 0.12- m SiGe bipolar technology (SiGe8T) [10]. The techof 200 GHz, nology includes NPN bipolar transistors with of 240–290 GHz, V, and V. Also metal–insulator–metal (MIM) capacitors, metal-film resistors, and four levels of metal are provided. Substrate resistivity is 11–16 -cm. This technology is the predecessor of a BiCMOS technology (SiGe8HP) which has the same features as above plus 0.12- m MOS transistors and associated front-end-of-the-line passive components. II. MMW DESIGN TECHNIQUES AND MEASUREMENT SETUPS A. Transmission Lines On-chip microstrip transmission lines (t-lines) are used for a variety of purposes in our circuits, including inductive and capacitive stubs, RF chokes, single-stub tuners for matching, and branch-line couplers. The impedance looking into a lossless t-line of length and terminated with impedance is as follows [28]: (1) Manuscript received April 9, 2004; revised May 28, 2004. This work was supported in part by NASA. The authors are with the IBM Thomas J. Watson Research Center, Yorktown Heights, NY 10598 USA (e-mail: brianfl@us.ibm.com). Digital Object Identifier 10.1109/JSSC.2004.837250 and are the characteristic impedance and propawhere gation constants of the line [28]. From (1), it can easily be shown that short-circuited stubs are inductive for lengths less 0018-9200/$20.00 © 2005 IEEE FLOYD et al.: SiGe BIPOLAR TRANSCEIVER CIRCUITS OPERATING AT 60 GHz than , a high impedance or RF choke for a length equal to , and capacitive for lengths between and . Likewise, open-circuited stubs are capacitive for lengths less than and inductive for lengths between and . Since our t-lines are implemented solely in the back-end-of-the-line, silicon dioxide rather than the silicon substrate appears in the is 600 m at 60 GHz; t-line “stack-up.” Accordingly, thus, RF chokes can easily be integrated, as can the stubs. Matching networks are implemented in a number of our circuits using single-stub shunt tuning [28]. This network consists of a series t-line followed by a stub connected in parallel, and can be found in the LNA and PA schematics. An advantage of the single-stub tuner is that it does not require prohibitively small series capacitors (e.g., 10s of fF), since the network is realized solely with t-lines. Operation of the tuner is best described through an example of matching a given admittance, to 50 . The series line first transforms to , i.e., rotating along a constant VSWR circle on the Smith chart. which adds The shunt stub then generates an admittance of directly to i.e., moving along the 0.02 conductance circle to the center of the Smith chart, resulting in a matched condition. Models for the microstrip transmission lines were included in the design kit [29]. We further verified these models up to 110 GHz using electromagnetic (EM) simulations and subsequent measurements. Note that bends and junctions were ignored in our simulations, since earlier measurements showed small discontinuities from these structures [29]. In most places, the microstrip was implemented using top-level metal (AM) with either metal-1 (M1) or metal-2 ground planes. The AM layer is much thicker than the 0.34- m skin depth at 60 GHz. Side ground shields were used in a number of places to minimize coupling to adjacent structures. This resulted in a maximum characteristic impedance of 65 for the minimum line width of 4 m. Finally, the quality factor of a short-cir[28], which equals at cuited stub can be shown to be 60 GHz for the AM-over-M1 microstrip ( mm and Np/mm, where dB). B. Simulation Methodology The circuits were designed using a standard Cadence-based methodology, with SpectreRF as the simulator for the LNA, VCO, and mixer, and ADS as the simulator for the PA. In addition, EM simulations (method-of-moments using IE3D from Zeland Software) were used to simulate more complicated passive structures such as coplanar waveguide tapers or 90 hybrids (e.g., a branch-line coupler), as well as to verify the t-line model and to estimate via inductance. At MMW frequencies, the difference between a device and a parasitic can be small; therefore, accurate parasitic extraction is crucial. For these designs, most of the circuit was covered with a ground plane, allowing many interconnects to be accounted for using the design-kit’s t-line models. Parasitic capacitance was then extracted on local (i.e., nondistributed) nodes. Parasitic via inductance was included wherever practical. An initial value of 7 pH was included for the AM-to-M1 via stack; however, subsequent measurements show that the via inductance is closer pH. to 157 C. Measurement Setups All measurements were made on-chip through waferprobing. Standard measurement setups were employed for network analysis, spectrum analysis, and power measurements using a 65- or 110-GHz vector network analyzer, a 40-GHz spectrum analyzer with external WR-15 harmonic mixer (50–75 GHz), and a power meter with a 1.85-mm, 65-GHz power sensor. Custom measurement setups were constructed to measure noise figure (NF) and intercept points, and to implement V-band baluns. The NF setup consists of a WR-15 noise source with isolator, the device-under-test with probes/cables/adaptors, an external 59–64 GHz downconverter, an LO signal generator for the downconverter, and the noise-figure measurement system. Mixer double-sideband NF measurements were obviously made without the external downconverter. The external downconverter consists of an LNA, mixer, and active frequency quadrupler, and has 5-dB -dB image rejection. To generate NF, 20-dB gain, and two tones for an IM3 or IM2 measurement, a 65-GHz signal generator was used to produce one tone, while the other tone was produced using a 20-GHz generator and a frequency quadrupler. The tones were combined with a WR-15 directional coupler, and then adjusted with an attenuator. Finally, 60-GHz baluns were built using magic-Ts [28], WR-15 variable phase shifters, and WR-15 to 1.85-mm adaptors. A simple routine was developed to calibrate the baluns such that 180 phase shift was obtained at the probe tips. III. MILLIMETER-WAVE CIRCUITS A. 60-GHz System Overview and Proof of Concept The planned 60-GHz transceiver architecture consists of a direct-conversion receiver (RX) and a variable-IF heterodyne transmitter (TX). Fig. 1 shows a simplified block diagram of the components in the transceiver which have currently been implemented. Direct conversion was selected for the RX to avoid the need for integrating an image-reject filter, though a superheterodyne RX is also being investigated [30]. The large signal bandwidth envisioned for high rate 60-GHz data transmissions enables a highpass filter to be used to filter away any LO-RF leakage effects in the direct-conversion RX. Although this dc-block filter can affect the switching time between TX, RX, and standby operating modes in the transceiver, the cutoff frequency can be relatively high with wideband modulation which minimizes the duration of the settling transient. For the TX, direct-conversion seems ill-advised due to carrier feedthrough in the mixer at 60 GHz, which most likely would need to be addressed using a nulling approach to maintain desired carrier suppression. To avoid the need for an LO-feedthrough compensation based upconversion design, a two-stage superheterodyne transmitter design is used. A variable-IF concept was chosen to allow the use of a single TX-VCO operating at two-sevenths the RF frequency. This results in an IF of RF/7. The two LO frequencies for the TX are then generated by tripling or halving the TX-VCO. To investigate the feasibility of 60-GHz links, “breadboard” versions of a superheterodyne TX and superheterodyne RX were constructed using commercially available components 158 Fig. 1. IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 40, NO. 1, JANUARY 2005 Block diagram of four 60-GHz transceiver components—LNA, direct downconverter, PA, and 20-GHz VCO. Dashed boxes show boundaries of each chip. Fig. 2. Simplified schematic of 60-GHz low-noise amplifier. [31]. The demonstrators have about 8-dB NF and 10-dBm output power at the RX and TX antennas, respectively. Using direct-sequence spread-spectrum based QPSK modulation, a 200-Mb/s link was demonstrated over two meters using omni-directional antennas even after totally blocking the line-of-sight. In [31], a 900-Mb/s link was demonstrated with the same breadboard system using 20-dBi horn antennas. B. Low-Noise Amplifier A two-stage single-ended LNA was implemented as shown in Fig. 2. The first stage common-base amplifier was chosen for its simple input matching, its higher gain compared to an inductively degenerated common-emitter amplifier (due to the lack of degeneration), and its high reverse isolation which decouples the input and inter-stage matching networks. The second stage consists of a common-emitter cascode amplifier with emitter degendB of gain when biased at 3 mA eration. Each stage provides V. Single-stub tuners are used for the input, and inter-stage, and output matching networks, with stub lengths of m and series t-line lengths of m for the output and inter-stage networks. MIM ac coupling capacoperating beyond self-resonance are used between itors the two stages and at the output. At 60 GHz, the silicon substrate exhibits low impedance; thus, care was taken in circuit design and layout to ensure adequate reverse isolation and stability. Metal-1 ground shields are used throughout the amplifier, and substrate ties are included wherever possible. Also, metal-2 planes are used together with many MIM bypass capacitors to realize a low-impedance supply. A die photograph of the LNA is shown in Fig. 3. The die size is 0.9 0.6 mm . Diamond-shaped pads are used at the input and output to reduce the parasitic capacitance of the pad. The measured S-parameters for the LNA are shown in Fig. 4, together with the simulated results. The LNA is unconditionally stable over 30–110 GHz. At 61.5 GHz, the gain is 14.7 dB and the reverse isolation is 40 dB, when biased at 6 mA from FLOYD et al.: SiGe BIPOLAR TRANSCEIVER CIRCUITS OPERATING AT 60 GHz Fig. 3. Die photograph of 60-GHz low-noise amplifier. Die size is 0.9 Fig. 4. Measured and simulated S-parameters of the LNA for V 159 2 0.6 mm . = 1:8V, I a 1.8-V supply. The input return loss is 6 dB, while the output return loss is 17 dB. Model-to-hardware correlation is excellent , and . The miss in , though, was unexpected, for particularly given that the output match was right on target. An alternate version of the LNA which had 50- coplanar waveguide (CPW) tapers at the input and output [27] showed and better than 12 dB, as expected. These CPW tapers absorb pad parasitics, providing a 50- impedance directly at the microstrip ports of the LNA. Since the CPW-LNA input match was on target, we know that the input matching network have been correctly modeled. This then leaves and transistor the bondpad as the source of mismatch. Simulations reveal that when the bondpad capacitance is significantly reduced, simuagree very well, while the other three lated and measured = 6 mA. S-parameters still agree. This is shown in Fig. 4 by the curve plot. Work is ongoing to charlabeled “sim. w/o pad” in the acterize the bondpad using on-chip TRL (Thru-Reflect-Line) calibration structures. Fig. 5 shows the measured and simulated NF of the LNA (with pads) at 6 mA and 1.8 V. No on-chip loss has been de-embedded from this result. Note that a better-calibrated WR-15 noise source over that used in [27] resulted in less ripple across the band, but slightly higher NF values. The measured NF is 4.5 dB at 61.5 GHz, while the simulated NF is 4.6 dB. The taper-LNA also has a NF of 4.5 dB when the insertion loss of the tapers is de-embedded. The simulated minimum NF of the of a single transistor is 3.1 dB. LNA is 4.2 dB, while the From simulation, the input common-base device contributes 160 Fig. 5. Measured and simulated noise figure (NF) of the LNA for V V, I = 6 mA. IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 40, NO. 1, JANUARY 2005 = 1:8 80% of the added noise (split nearly equally between collector shot noise and thermal noise from base resistance), while the second-stage cascode contributes 10%. The remaining 10% comes from other assorted sources. Biasing of the two stages in the LNA is independent and adjustable. By changing the second-stage current, adjustable gain is obtained while NF remains relatively constant. Specifically, with the first-stage current held constant at 3 mA and the second-stage current adjusted between 0.5 and 5 mA, measurements show that the gain varies by 10 dB while the NF reremains below 9 dB. The gain mains below 5.5 dB and and NF of the LNA were measured across multiple dies and dB over temperature. Die-to-die gain and NF variations of and dB, respectively, were observed for 23 LNA samples. Fig. 6 shows the measured gain, NF, and current of the taper-LNA over a 5 C–95 C temperature range. With a constant voltage applied to the current mirror, the resultant bias currents increase faster than proportional-to-absolute-temperature (PTAT). The measured gain varies a total 1.5 dB while NF varies 1 dB. Finally, Fig. 7 shows the measured 1-dB compresand IIP3 (at 100-MHz tone spacing) of the sion point LNA, revealing a 20-dBm and 8.5-dBm IIP3 for both LNA versions when biased at 6 mA and 1.8 V. Simulations predict a 22-dBm iCP and 12-dBm IIP3. C. Downconverter The architecture of the direct-conversion downconverter is shown in Fig. 1. Direct conversion is often used at microwave frequencies because it facilitates high integration, eliminating image-reject and IF filters [32]. However, one of the issues in a direct-conversion receiver is leakage of the local oscillator (LO) signal into the RF signal path, which results in dc offsets at the mixer output. LO leakage becomes a greater issue at 60 GHz because of coupling through the silicon substrate. In the present design, coupling was controlled by covering most of the substrate with a ground plane on the first metal level (M1), placing substrate contacts every 5 m. Referring to Fig. 1, LNA2 has 12 dB of gain and serves as an active balun, providing a differential RF signal to the two Fig. 6. Measured gain, NF, and current of the taper-LNA versus temperature. Fig. 7. Two-tone IP3 measurement of the LNA. double-balanced Gilbert-cell mixers, which have 4 dB of gain. The mixers are driven with quadrature LO signals from a differential branch-line directional coupler. A pilot signal comes on-chip at a third of the desired LO frequency, and a frequency tripler generates the final LO signal to drive the branch-line coupler. This architecture minimizes LO-RF coupling through the probes and pads, and the inclusion of the frequency tripler will ultimately allow the frequency synthesizer to run at one-third of the LO frequency. LNA2 consists of a cascoded differential pair with inductive load and degeneration, as shown in Fig. 8. The four inductors are realized with ac short-circuited stubs, realizing an effective load inductance of 80 pH and degeneration of 60 pH. Inductive degeneration creates a 50- real part to the input impedance, and the input match is completed by series inductor . is made from a straight segment of top-metal (AM) over the substrate, and losses are reduced by cross-hatching the substrate under with oxide-filled trenches. LNA2 draws 8 mA from 2.7 V. The mixers are conventional resistively degenerated doublebalanced Gilbert cells, as shown in Fig. 9. Each mixer and associated LO buffer together draw 12 mA. Double emitter followers are used to provide a low-impedance LO signal to the mixer FLOYD et al.: SiGe BIPOLAR TRANSCEIVER CIRCUITS OPERATING AT 60 GHz 161 Fig. 9. Simplified schematic of 60-GHz double-balanced Gilbert-cell. Fig. 8. Simplified schematic of LNA2 used in downconverter. switches and to isolate the load capacitance of the switches from the branch-line coupler. The mixer load resistors are 300 , so the mixers are followed by unity-gain buffers to provide 100differential baseband signals to drive off-chip. The baseband 3-dB bandwidth is 1.5 GHz. The buffers, which draw 26 mA each, are for test only, since an analog baseband is planned to be integrated with the downconverter. The frequency tripler block in Fig. 10 consists of two stages: the actual tripler, and an LO amplifier following the tripler and preceding the branch-line coupler. A 0-dBm, 20-GHz differential LO pilot signal is applied to the tripler (a cascoded differential amplifier with a tuned load), which produces third-harmonic distortion. The LO amplifier has high input-impedance emitter followers so that isolated, high-Q (5–10) tuned loads are pro. This provides gain for the third duced at the collectors of harmonic around 60 GHz while attenuating the 20-GHz fundamental. The LO amplifier also has a tuned load to further amplify the third harmonic and reject the fundamental; it supplies dB down. Com2 dBm (simulated) with the fundamental match the input impedance of the ponents - and tripler to 100- differential at 20 GHz. The tripler alone draws 4 mA and LO amplifier 16 mA. Fig. 11 is a die photograph of the direct downconverter, which has a die size of 1.9 1.6 mm . The branch-line coupler, folded into the shape of a peanut to reduce its size, is located in the center. The measured gain and NF of the downconverter are shown in Fig. 12. All measurements of conversion gain, NF, , and linearity refer to the unbalanced 50- input of LNA2. Measured cable and probe losses in the test setup have been de-embedded from the measurements, but no on-chip losses (such as those due to the pads) have been de-embedded. Data was taken from a wafer on which transistors had a peak of 290 GHz, whereas simulations were done with models that of 240 GHz. Therefore, measured percorrespond to an formance is sometimes better than simulated performance. At 60 GHz, gain is 18.6 dB and NF is 13.3 dB, while simulated gain is 17.1 dB and NF is 14.9 dB. Some samples, such as chip F6 in Fig. 12, show a dip in conversion gain at 64 GHz, which we believe is due to interaction between the LO amplifier and is 17 dBm, with a corbranch-line coupler. Measured responding IIP3 of 7 dBm and IIP2 in the range of 9 to 20 dBm, depending on the sample. Average IIP2 for 4 samples was 13 dBm. IIP3 and IIP2 were measured with a 19-GHz LO-pilot signal applied to the frequency tripler and RF input is 12 dB or better from 57 tones at 57.1 and 57.12 GHz. to 65 GHz. Fig. 13 shows LO leakage measured at LNA2’s input with a spectrum analyzer; it varies little from chip-to-chip. The leakage generally rises with frequency and reaches 50 dBm at 60 GHz, implying approximately 50 dB of isolation. Mixer offset voltages are another measure of LO-RF isolation. Measured I and Q offsets on eight chips have an average offset-voltage magnitude of 56 mV (49 mV for I-channel, 64 mV for Q-channel), inmV. The remaining six chips cluding 2 chips with offsets had an average offset of 21 mV. These offsets are 100 times higher than we measured on an earlier 2.1-GHz design [32], but still indicate that direct conversion is practical for a 60-GHz WPAN, given 650-mV mixer output swing and ac coupling. 162 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 40, NO. 1, JANUARY 2005 Fig. 10. Simplified schematic of frequency tripler block. Fig. 11. Die photo of the direct downconverter. Die size is 1.9 2 1.65 mm . AC coupling at the mixer outputs can prevent the offsets from unbalancing subsequent stages and will have little impact on a wideband WPAN baseband signal. The offsets do consume some of the mixers’ dynamic range and decrease IIP2, but linearity requirements for a 60-GHz WPAN are minimal, due to a lack of adjacent-channel interference. Fig. 14 shows the variation in gain and NF over temperature with bias currents held PTAT by adjusting an off-chip bias supply. The gain is held fairly constant up to 60 C, but then falls off rapidly. Simulations reveal that the LO amplifier which drives the branch-line coupler has reduced output above 60 C, and the mixer conversion gain drops with the reduced LO drive. Fig. 14 also shows measured NF rising from 12.5 to 16.5 dB as temperature increases from 5 to 80 C. Figs. 15 and 16 reveal the performance of the branch-line coupler. Ideally, the I and Q channels would have the same gain and be exactly 90 out of phase. But the outputs of the branch-line coupler vary in amplitude and phase over the frequency range, and thus the I and Q channel mixer outputs also vary. Fig. 15 shows that the coupler produces LO outputs of roughly equal Fig. 12. Measured conversion gain and NF for the direct downconverter. Fig. 13. Measured local oscillator leakage to LNA2 RF input versus frequency. amplitude from 54 to 62 GHz, and the simulations match the measurements quite closely. Fig. 16 shows that the simulated of phase difference between I and Q channels is within 90 from 44 to 66 GHz, but the measured results show a phase FLOYD et al.: SiGe BIPOLAR TRANSCEIVER CIRCUITS OPERATING AT 60 GHz Fig. 14. Measured and simulation conversion gain and NF of downconverter versus temperature with PTAT bias. 163 Fig. 15. Measured I-channel to Q-channel gain balance versus frequency. Fig. 16. Measured I-channel to Q-channel phase difference versus frequency. difference of 87 to 105 from 57 to 64 GHz. These results indicate that our simulation model for the coupler is not as accurate as we would like. However, our system-level simulations indicate that even without improvement, the measured quadrature accuracy is still adequate for a 60-GHz WPAN, since it has a small impact on the bit-error rate of the demodulated signal. Smulders [25] discusses some of the modulation schemes that might be used in a 60-GHz WPAN. D. Power Amplifier The PA is a two-stage class-AB balanced amplifier, designed to directly feed a differential antenna, eliminating any need for an on-chip balun. The balanced amplifier is implemented with two unbalanced amplifiers in parallel. Each unbalanced amplifier, shown in Fig. 17, consists of two common-emitter amplifiers, with input, inter-stage, and output matching networks. Single open-circuited stub tuning networks are used at the input and output, while the inter-stage match uses a MIM capacitor instead of a stub. Quarter-wavelength RF chokes supply to each stage. Each of the four power transistors has a separate temperature-compensated bias circuit [33], which is located nearby for good thermal matching. Two major problems associated with the design of on-chip power amplifiers in an advanced SiGe bipolar technology are the low breakdown voltages and the high loss of the on-chip impedance transformation. In order to operate the transistor at higher voltages, to the power transistor’s each bias circuit provides base terminal. Thus, the NPN is limited by operation , where is slightly above 4 V for rather than equal to 300 . Fig. 18 shows a die photograph of the PA, which has a size of 2.1 0.8 mm . Clearly visible left and right are the two identical unbalanced amplifiers. The PA was biased at 150 mA from a 2.5-V supply. Fig. 19 shows the measured transducer power gain and output power versus input power for the balanced amplifier. At 61.5 GHz, are 10.8 dB and 11.2 dBm, respectively. the gain and The saturated output power level is 16.2 dBm—measured for an unbalanced amplifier to which 3 dB is added. The maximum power-added efficiency (PAE) is 4.3%. Fig. 20 shows the measured gain and output power versus temperature, with constant-current biasing. Across 5 C–135 C, the gain each drop around 4 dB, yet the amplifier still delivers and 9-dBm output power at 100 C. Input and output return losses at 61.5 GHz are 6.2 and 8.9 dB, respectively, with respect to a 100- differential impedance. E. Voltage-Controlled Oscillators VCOs have been implemented at both 20 GHz, for use with the frequency tripler, and at 60 GHz, for use as a fundamental-frequency oscillator. A fully differential Colpitts topology is used at both frequencies. Fig. 21 shows a schematic of the 60-GHz VCO. The hallmark capacitive divider of the Colpitts architecture is formed by the base-to-emitter capaciand the coupling capacitor between both sides tance of of the VCO. Microstrip transmission lines are used to form a resonant circuit, connecting the bases to a current mirror and realizing an inductance of 40 pH. Frequency tuning is provided by junction varactors located in the base. No explicit output buffers are included—the output is obtained directly from the collector, which has a microstrip load. The 20-GHz VCO is identical in architecture except that a slab inductor is used in place of the base microstrip, a spiral inductor is used in place 164 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 40, NO. 1, JANUARY 2005 Fig. 17. Simplified schematic of unbalanced amplifier used for power amplifier. Fig. 18. Die photograph of the PA. Die size is 2.1 2 0.8 mm . Fig. 19. Measured transducer power gain and output power at 61.5 GHz versus input power for the balanced PA. of the collector microstrip, and a MIM capacitor is added between the base and emitter. Note that an output buffer would be required to drive an off-chip load, and the buffer would likely degrade the phase noise [17]. The buffer is required since unmatched external loads cause the two halves of the VCO to oscillate at slightly different frequencies, resulting in a discontinuous and hysteretic tuning response. The 60-GHz VCO consumes 8 mA from 3 V. The 60-GHz VCO was designed to operate between 60 and 63 GHz, using a parasitic via inductance value of 7 pH. Measurements show Fig. 20. Measured gain and output power (P temperature. ) of the PA versus the tuning range to instead be 65.8 to 67.9 GHz (3.1%), for from 0 to 3 V. This is consistent with a via inductance value of 4 pH, which gives a 64.3 to 67.7 GHz simulated tuning range. The differential output power, measured with a power meter, is 8 dBm across the band. The measured phase noise at the tuning range edges is 98 dBc/Hz at 1-MHz offset for 50- loads. This improves to 104 to 102 dBc/Hz (shown in Fig. 22) in the middle of the tuning range. The corresponding figure of merit for the VCO [34] is between 181 and 187 dB. FLOYD et al.: SiGe BIPOLAR TRANSCEIVER CIRCUITS OPERATING AT 60 GHz Fig. 21. Fig. 22. offset. 165 Simplified schematic of the 60-GHz VCO. Output spectrum of 60-GHz VCO, showing phase noise at 1-MHz The 20-GHz VCO can be tuned from 21.2 to 22.4 GHz (4.5%), and the differential output power is 5.5 dBm. Power consumption is 9 mA from 3 V. Phase noise at 1-MHz offset is between 109 and 116 dBc/Hz. Note that this measurement helps to verify the 67-GHz phase noise measurements, since or adjusting for frequency, we should see at least a 9.5-dB improvement. More than this improvement is seen since varactor Q should be three times larger at 22 GHz than at 67 GHz. IV. SUMMARY AND CONCLUSIONS This paper has described key RF front-end components in silicon for a 60-GHz transceiver, including an LNA, a direct-downconverter, a PA, and VCOs. The performance of these circuits is summarized in Table I. Excellent model-to-hardware correlation has been achieved for each circuit. At 61.5 GHz, the LNA achieves 4.5-dB NF, 14.7-dB gain, and 10.8-mW power dissipation, and is the first known demonstration of a silicon LNA at V-band. In comparison, III-V LNAs have typically achieved NFs in the 2 to 4 dB range, gains greater than or equal to 20 dB, and power dissipations between 20 and 160 mW [1]–[5]. Our silicon V-band LNA achieves similar NF to GaAs at lower power consumption. The III-V LNAs, though, do achieve higher compression points than this SiGe LNA. At 61.5 GHz, the downconverter achieves 16-dB gain, 14.8-dB NF, and 150-mW power dissipation. Once again, this is the first known demonstration of active mixers in silicon at V-band. In comparison, III-V mixers are often passive or diode-based [35], requiring LO-drive levels of 10 to 20 dBm and thereby consuming a large amount of power in the LO amplifiers (e.g., 1 W in [3]). Thus, the silicon downconverter described here represents a very low power consumption. Also, the downconverter block contains 80 transistors and 43 inductors or transmission lines, which is a very high level of integration at this frequency. The PA provides a 10.8-dB gain, a 11.2-dBm 1-dB compression point, and a maximum PAE of 4.3%. The saturated output power is 16 dBm. This is the second known demonstration of a silicon PA at these frequencies, where Li [21] has demonstrated a 14-dBm silicon amplifier at 77 GHz with % efficiency. In comparison, III-V PAs for V-band deliver 23–30 dBm of power with 8–15 dB of gain and 20%–40% of power-added efficiency [6]–[9]. Finally, 60- and 20-GHz VCOs have been demonstrated with very low phase noise, albeit with narrow tuning ranges. The 60-GHz VCO achieves an excellent phase noise at 1-MHz offset between 98 and 104 dBc/Hz, which is comparable to the performance achieved in [17]. Based on these results, a fully integrated transceiver with less than 6-dB NF and greater than 11-dBm output power should be possible. Specifically, cascading the LNA and downconverter mathematically results in a receiver front-end with a . 5.8-dB NF, 31-dB gain, 22-dBm IIP3, and 32-dBm The utility of such hardware has already been demonstrated in 166 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 40, NO. 1, JANUARY 2005 TABLE I SUMMARY OF MEASURED PERFORMANCE OF LNA, DOWNCONVERTER, PA, AND VCO Section III-A and in [31], where discrete III-V based 60-GHz radios with 8-dB NF and 10-dBm output power (both antenna-referred) were used to implement a 200-Mb/s omnidirectional link and a 900-Mb/s directional link in our lab. This hardware, then, is the first step toward the implementation of a fully integrated 60-GHz transceiver in SiGe technology. Clearly, it is in the integration level where silicon can truly stand out over III-V implementations. SiGe’s high level of integration and potentially low power consumption of the MMW downconverter/upconverter could open the door to new MMW applications and market opportunities, including high data-rate wireless personal-area networks. ACKNOWLEDGMENT The authors thank IBM’s SiGe Technology group for chip fabrication. Additionally, the authors thank D. Beisser, D. Goren, D. Friedman, M. Soyuer, and M. Oprysko of IBM Research, and Y. Tretiakov, and G. Freeman of IBM Microelectronics for their contributions. REFERENCES [1] K. Ohata et al., “1.25 Gb/s wireless gigabit ethernet link at 60 GHzband,” in IEEE Int. Microwave Symp. Dig., Jun. 2003, pp. 373–376. [2] L. Tran et al., “High performance, high yield millimeter-wave MMIC LNA’s using InP HEMTs,” in IEEE Int. Microwave Symp. Dig., Jun. 1996, pp. 9–12. [3] M. Siddiqui et al., “GaAs components for 60 GHz wireless communication applications,” presented at the GaAs Mantech Conf., Apr. 2002. [4] A. Fujihara et al., “High performance 60-GHz coplanar MMIC LNA using InP heterojunction FET’s with AlAs/InAs superlattice layer,” in IEEE Int. Microwave Symp. Dig., Jun. 2000, pp. 21–24. [5] K. Nishikawa et al., “Compact LNA and VCO 3-D MMIC’s using commercial GaAs PHEMT technology for V-band single-chip TRX MMIC,” in IEEE Int. Microwave Symp. Dig., Jun. 2002, pp. 1717–1720. [6] O. Tang et al., “A 560 mW, 21% power-added efficiency V-band MMIC power amplifier,” in GaAs IC Symp. Tech. Dig., Nov. 1996, pp. 115–118. [7] R. Kasody et al., “A high efficiency V-band monolithic HEMT power amplifier,” IEEE Microw. Guided Wave Lett., vol. 4, no. 9, pp. 303–304, Sep. 1994. [8] W. Kong et al., “Very high efficiency V-band power InP HEMT MMICs,” IEEE Electron Device Lett., vol. 21, no. 11, pp. 521–523, Nov. 2000. [9] Y. Chen et al., “A single-chip 1-W InP HEMT V-band module,” in GaAs IC Symp. Technical Dig., Oct. 1999, pp. 149–152. [10] B. Jagannathan et al., “Self-aligned SiGe NPN transistors with 285 GHz f and 207 GHz fT in a manufacturable technology,” IEEE Electron Device Lett., vol. 23, no. 5, pp. 258–260, 2002. [11] T. Meister et al., “SiGe bipolar technology with 3.9 ps gate delay,” in Proc. Bipolar/BiCMOS Circuits and Technology Mtg., Sep. 2003, pp. 103–106. [12] T. Hashimoto et al., “Direction to improve SiGe BiCMOS technology featuring 200-GHz SiGe HBT and 80-nm gate CMOS,” in Int. Electron Device Mtg. Tech. Dig., Dec. 2003, pp. 129–132. [13] H. Rucker et al., “SiGe:C BiCMOS technology with 3.6 ps gate delay,” in Int. Electron Device Mtg. Tech. Dig., Dec. 2003, pp. 121–124. [14] J.-S. Rieh et al., “SiGe HBT’s for millimeter-wave applications with simultaneously optimized f and f of 300 GHz,” in IEEE RFIC Symp. Dig. Papers, Jun. 2004. [15] W. Winkler et al., “60 GHz and 76 GHz Oscillators in 0.25 m SiGe:C BiCMOS,” in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2003, pp. 454–455. [16] Y. Baeyens and Y. Chen, “A monolithic integrated 150 GHz SiGe HBT push-push VCO with simultaneous differential V-band output,” in IEEE Int. Microwave Symp. Dig., Jun. 2003, pp. 877–880. [17] H. Li and H.-M. Rein, “Wide-band VCO’s in SiGe production technology operating up to about 70 GHz,” IEEE Microw. Wireless Compon. Lett., vol. 13, no. 10, pp. 425–427, Oct. 2003. [18] H. Li, H.-M. Rein, and M. Schwerd, “SiGe VCO’s operating up to 88 GHz, suitable for automotive radar sensors,” Electron. Lett., vol. 39, no. 18, pp. 1326–1327, Sep. 2003. [19] W. Perndl, H. Knapp, K. Aufinger, T. Meister, W. Simburger, and A. Scholtz, “A 98 GHz voltage controlled oscillator in SiGe bipolar technology,” in Proc. Bipolar/BiCMOS Circuits and Technology Mtg., Sep. 2003. [20] H. Li, H.-M. Rein, and T. Suttorp, “Design of W-band VCO’s with high output power for potential application in 77 GHz automotive radar systems,” in GaAs IC Symp. Tech. Dig., Oct. 2003, pp. 263–266. FLOYD et al.: SiGe BIPOLAR TRANSCEIVER CIRCUITS OPERATING AT 60 GHz [21] Y. Li, M. Bao, M. Ferndahl, and A. Cathelin, “23 GHz front-end circuits in SiGe BiCMOS technology,” in IEEE RFIC Symp. Dig. Papers, Jun. 2003, pp. 99–102. [22] I. Gresham et al., “Ultra wide and 24 GHz automotive radar front-end,” in IEEE Int. Microwave Symp. Dig., Jun. 2003, pp. 369–372. [23] H. Hashemi, X. Guan, and A. Hajimiri, “A fully integrated 24 GHz 8-path phased-array receiver in silicon,” in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2004, pp. 390–391. [24] S. Hackl, J. Bock, M. Wurzer, and A. L. Scholtz, “40 GHz monolithic integrated mixer in SiGe bipolar technology,” in IEEE Int. Microwave Symp. Dig., Jun. 2002, pp. 1241–1244. [25] P. Smulders, “Exploiting the 60 GHz band for local wireless multimedia access: Prospects and future directions,” IEEE Commun. Mag., no. 1, pp. 140–147, Jan. 2002. [26] H. Xu, V. Kukshya, and T. S. Rappaport, “Spatial and temporal characteristics of 60-GHz indoor channels,” IEEE J. Select. Areas Commun., vol. 20, no. 3, pp. 620–630, Apr. 2002. [27] S. Reynolds, B. Floyd, U. Pfeiffer, and T. Zwick, “60 GHz transceiver circuits in SiGe bipolar technology,” IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, pp. 442–443, Feb. 2004. [28] D. M. Pozar, Microwave Engineering, 2nd ed. New York: Wiley, 1998. [29] D. M. Goren et al., “On-chip interconnect-aware design and modeling methodology, based on high bandwidth transmission line devices,” in Proc. 2003 Design Automation Conf., Jun. 2003, pp. 724–727. [30] S. K. Reynolds, “A 60-GHz superheterodyne down-conversion mixer in silicon-germanium bipolar technology,” IEEE J. Solid-State Circuits, vol. 39, no. 11, pp. 2065–2068, Nov. 2004. [31] B. Gaucher et al., “Silicon monolithic broadband millimeter wave radio technology,” in Proc. Int. Conf. Space Mission Challenges for Information Technology, Jun. 2003, pp. 113–121. [32] S. K. Reynolds, B. A. Floyd, T. Beukema, T. Zwick, U. Pfeiffer, and H. Ainspan, “A direct conversion receiver IC for WCDMA mobile systems,” IEEE J. Solid-State Circuits, vol. 38, no. 9, pp. 1555–1560, Sep. 2003. [33] E. Järvinen, S. Kalajo, and M. Matilainen, “Bias circuits for GaAs HBT power amplifiers,” in IEEE Int. Microwave Symp. Dig., Jun. 2001, pp. 507–510. [34] E. Hegazi, H. Sjoland, and A. A. Abidi, “A filtering technique to lower LC oscillator phase noise,” IEEE. J. Solid-State Circuits, vol. 36, no. 12, pp. 1921–1930, Dec. 2001. [35] T. H. Oxley, “50 years development of the microwave mixer for heterodyne reception,” IEEE Trans. Microw. Theory Tech., vol. 50, no. 3, pp. 867–876, Mar. 2002. Brian A. Floyd (S’98–M’01) received the B.S. (with highest honors), M.Eng., and Ph.D. degrees in electrical and computer engineering from the University of Florida, Gainesville, in 1996, 1998, and 2001, respectively. While at the University of Florida, he held the Intersil/SRC Graduate Fellowship and the Pittman Fellowship. His doctoral research on wireless interconnects was a Phase One winner and a Phase Two first runnerup in the 2000 SRC Copper Design Contest. In 2001, he joined IBM and is presently a Research Staff Member at the IBM T. J. Watson Research Center, Yorktown Heights, NY, engaged in millimeter-wave, RF, and high-speed wired integrated circuit design. Scott K. Reynolds received the B.S.E.E. degree from the University of Michigan in 1983, the M.S.E.E. degree from Stanford University in 1984, and the Ph.D. degree in electrical engineering, also from Stanford in 1987. He joined IBM in 1988 and is presently a Research Staff Member at the IBM T. J. Watson Research Center in Yorktown Heights, New York. His job responsibilities have involved analog and mixed-signal circuit design for high speed communication systems, including optical, wired, and RF wireless systems, and disk drive channels. Currently, he is engaged primarily in development of RFICs for high data rate wireless communication links. 167 Ullrich R. Pfeiffer received the diploma degree in physics and the Ph.D. degree in physics from the University of Heidelberg, Germany, in 1996 and 1999, respectively. In 1997, he worked as a Research Fellow at the Rutherford Appleton Laboratory, Oxfordshire, U.K., where he developed high-speed multi-chip modules. In 2000, his research was based on high-integrated real-time electronics for a particle physics experiment at the European Organization for Nuclear Research (CERN), Switzerland. He joined IBM in 2001 and is presently a Research Staff Member at the IBM T. J. Watson Research Center. His research involves RF circuit design, high-power amplifier design at 60 GHz and 77 GHz, high-frequency modeling and packaging for 60-GHz and 3G cellular systems. Thomas Zwick (M’00) received the Dipl.-Ing. (M.S.E.E.) and the Dr.-Ing. (Ph.D.E.E.) degrees from the Universität Karlsruhe (TH), Germany in 1994 and 1999, respectively. From 1994 to 2001 he was a Research Assistant at the Institut für Höchstfrequenztechnik und Elektronik (IHE) at the Universität Karlsruhe (TH), Germany. Since February 2001, he has been with the IBM T. J. Watson Research Center, Yorktown Heights, NY. His research topics include wave propagation, stochastic channel modeling, channel measurement techniques, material measurements, microwave techniques, wireless communication system design and millimeter wave antenna design. He participated as an expert in the European COST231 Evolution of Land Mobile Radio (Including Personal) Communications and COST259 Wireless Flexible Personalized Communications. For the Carl Cranz Series for Scientific Education, he served as a lecturer for Wave Propagation. Dr. Zwick received the Best Paper Award from the International Symposium on Spread Spectrum Technology and Applications (ISSSTA) 1998. Troy Beukema received the B.S.E.E. and M.S.E.E. degrees from Michigan Technological University in 1984 and 1988, respectively. From 1984 to 1988, he was an R&D Engineer with Hewlett-Packard in the area of communications test equipment. He joined Motorola in 1989 and contributed to development of digital cellular wireless systems with a focus on digital signal processing algorithm design and implementation. In 1996, he joined IBM, where he is presently a research staff member involved in communications system research. His research interests include communication link system design and simulation, with an emphasis on signal processing algorithms for wireless and high-speed wireline channels. Brian Gaucher performed his undergraduate work at the Univesity of Massachusetts and graduate work at Northeastern University. From 1982 to 1983, he worked at Alpha Industries R&D Laboratory designing microwave GaAsFET amplifiers, switches detectors limiters, filters, and supercomponents. In 1984, he joined GTE Communication Systems Division, working in the area of research and development of secure spread spectrum communication and radar systems for the military, across the 900 MHz to 60 GHz frequency bands. In 1993, he joined IBM and is presently a Research Staff Member at the IBM T. J. Watson Research Center, Yorktown Heights, NY, where he manages a communication system design and characterization group. His present research interests include 60 GHz Gb/s wireless communication design and biomedical applications of wireless technology. His group has helped more than five products come to market. Dr. Gaucher is an IBM master inventor and holds two outstanding technical achievement awards and one corporate award.