IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO.

3, MARCH 2013 1347

Analysis and Design of a Transformer-Feedback-

Based Wideband Receiver
Venumadhav Bhagavatula, Student Member, IEEE, and Jacques C. Rudell, Senior Member, IEEE

AbstractThis paper proposes a multistage transformer-feed- active imaging, a wideband passive-imaging receiver that col-
back-based design approach for a high fractional-bandwidth lects millimeter-wave radiation emitted by objects to render an
(fBW) IF stage in a 60-GHz heterodyne receiver. An in-depth
image has been reported [5] in the -band (70110 GHz).
analysis of source-to-gate transformer-feedback amplifiers, in-
cluding the design of the matching network, is presented. Analytic In both of the above applications, the rate of data transfer
expressions for the input resistance, quality factor, and noise figure and image resolution are enhanced by improving the bandwidth
(NF) as a function of the transformer turns-ratio ( ) and magnetic of the transceiver circuitry. At baseband frequencies, feedback
coupling factor ( ) are derived. To validate the proposed analysis, techniques may be applied to extend the bandwidth of a circuit
a wideband IF amplifier and mixer were designed in a 40-nm
CMOS process. From measured results, this device achieves a placed in a closed-loop configuration. However, in the past, lim-
16% 3-dB fBW, a peak power gain of 27.6 dB, an NF of 5.3 dB ited power gain and associated with CMOS transistors pre-
while consuming 28.8 mW from a 0.9-V supply. cluded the use of feedback techniques at RF frequencies. As
Index TermsCMOS, IF, Lange coupler, mixer, 60-GHz trans- a result, to realize a high fBW channel in CMOS, open-loop
ceiver, source-to-gate transformer feedback. circuit techniques employing high-order LC-based ladder filters
have been applied [6], [7]. Now, with advances in CMOS pro-
cessing technology through silicon scaling, the device
I. INTRODUCTION (unity power gain frequency) exceeds several hundred giga-
hertz, allowing designers to explore both resistive and reac-
T HE LAST decade has seen intense research efforts that
seek to exploit both the high device associated with
modern CMOS technologies and the large bandwidth available
tive-feedback circuit techniques.
Reactive feedback using integrated transformers, in which the
magnetically coupled windings provide a path for current-sense
at millimeter-wave frequencies to realize high data rate trans-
current feedback, has received considerable interest [8][10].
ceivers [1][4]. Specifically, the 7 GHz of unlicensed bandwidth
There are three fundamental transformer-feedback topologies
from 57 to 64 GHz has attracted attention as a potential solution
[11]: drain-to-source, drain-to-gate, and source-to-gate. The
for short-range high-speed data communication links [2], [3].
first two, drain-to-source and drain-to-gate, have been applied
Development of circuit techniques capable of handling modu-
in single-ended amplifiers to neutralize the device capac-
lated data with fractional bandwidths (fBWs), defined as (abso-
itance and improve reverse-isolation over a wide bandwidth.
lute bandwidth/center frequency), on the order of 15%20% is
The source-to-gate transformer-feedback (SGTxFB) topology
crucial for these high data rate communication standards. The
is more suited for wideband matching network design and
targeted data rates of up to 7 Gb/s have the potential to en-
has been applied over a wide range of operating frequencies,
able applications such as streaming uncompressed high-defini-
ranging from ultra-wideband (UWB) [8] to -band [9].
tion video [3] from a set-top box to a television, and communi-
However, currently available analytic models for SGTxFB-
cation between a portable smart-phone and a Wireless Kiosk.
based matching-network design provide little intuition for op-
Imaging systems for medical and security applications have
timization. This paper provides a generic and systematic ap-
also benefited from wide bandwidth silicon integrated circuits
proach to the design of SGTxFB amplifiers. The input admit-
(ICs). An active-imaging transmitter with pulse-widths as low
tance is modeled as a function of transformer and transistor
as 26 ps, enabling free-space depth resolution of less than 4 mm,
parameters. The models are then used to assess the impact of
has been proposed in [4] for early breast-cancer detection. This
the circuit parameters on the bandwidth, noise figure (NF), and
transmitter requires circuits with a bandwidth and center fre-
gain. In addition, application of SGTxFB in prior art has been
quency in excess of 30 and 75 GHz, respectively. In addition to
restricted to the interface between the antenna and low-noise
amplifier in integrated receivers. In contrast, this paper presents
Manuscript received October 09, 2012; revised January 12, 2013; accepted
a multistage SGTxFB-based high-fBW IF stage for the 60-GHz
January 15, 2013. Date of publication February 19, 2013; date of current version heterodyne receiver shown in Fig. 1. Transformer feedback is
March 07, 2013. This work is an expanded paper from the IEEE RFIC Sympo- employed to provide a wide bandwidth load on each circuit
sium, Montreal, QC, Canada, June 1719, 2012.
The authors are with the Electrical Engineering Department, University of block in the receiver. The challenges involved in designing a
Washington, Seattle, WA 98195 USA (e-mail: [email protected]; jcrudell@uw. multistage SGTxFB down-converter are described.
edu). This paper is organized as follows. First, an ideal cur-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. rent-feedback amplifier (CFA) model is introduced, and related
Digital Object Identifier 10.1109/TMTT.2013.2244904 to a simplified first-order SGTxFB stage in Section II. Next,

0018-9480/$31.00 2013 IEEE

1348 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 3, MARCH 2013

Fig. 1. Heterodyne phase-array receiver: low-noise amplifier (LNA), power-

combiner (PC) followed by a 1113-GHz IF stage comprised of an IF amplifier,
quadrature down-conversion mixer, and lumped-element Lange coupler for I/Q
generation. Fig. 3. (a) Common-source (CS) amplifier with inductor degeneration. (b) CS
amplifier with inductor degeneration and current feedback. (c) Input impedance
with . (d) with .

the circuit in Fig. 3(a), it is straightforward to prove that the

input impedance is described by

Fig. 2. (a) Generic feedback circuits. (b) Transformer-based feedback circuit. (1)

Accordingly, the circuit appears to be a series-RLC network

with , resonant frequency
guidelines for matching network design and accurate input , and a quality factor , where
admittance models for SGTxFB amplifiers are discussed in
Section III. This is followed with a derivation of analytic (2)
expressions for the noise and gain of SGTxFB amplifiers
in Section IV. The design of an IF stage, operating over a Next, consider the CFA in Fig. 3(b), where describes the
frequency range of 1113 GHz, is described in Section V. ratio of the source current to the current fed back to the gate.
Measured results from a prototype chip implemented in a The CFA and inductor-degenerated amplifier have identical ,
40-nm CMOS process are presented in Section VI. Finally, , and ; however, the devices are biased differently, and
in Section VII, this paper concludes with some summary hence, have differing transconductance. The input impedance
comments. is given by

II. CURRENT FEEDBACK

The generic model of a feedback system is shown in Fig. 2(a).
The forward path consists of a high-gain amplifier, . A
fraction of the output voltage (or current) of is sampled (3)
by the feedback circuit and fed back to the input. In the cir-
cuit shown in Fig. 2(b), the feedback circuit is a transformer;
At . Accordingly, for circuits
hence, the name transformer feedback. The secondary windings
with quality factor , the complex term in the denominator
of the transformer samples the output current of and the
is approximately equal to 1; reducing (3) to
current induced in the primary is fed back to the input. In the spe-
cific case of SGTxFB, the output current is sampled at the source
(4)
and induced current is fed back to the gate of the MOSFET.
As a first-order approximation, an SGTxFB amplifier can
Thus, similar to (1), the input impedance of the CFA ap-
be modeled as the CFA shown in Fig. 3(b). In this ideal
pears as a series RLC circuit resonant at . In addition, it can
current-feedback model, the feedback current induced in the
be observed that the CFA has a input impedance
primary is included; however, the feed-forward current
and quality factor , where
induced in secondary is ignored. To study the impact of
current feedback on bandwidth, consider the inductor-degener-
ated common-source amplifier (CSA) without and with current
feedback shown in Fig. 3(a) and (b), respectively. Starting with (5)
BHAGAVATULA AND RUDELL: ANALYSIS AND DESIGN OF TRANSFORMER-FEEDBACK-BASED WIDEBAND RECEIVER 1349

Several important observations can be made based on the

above result. Assume the circuit in Fig. 3(a) is designed to match
with an antenna with resistance , i.e., . The
first case to consider is wherein the transconductance of
is identical in both circuits . The resulting input
impedance as a function of frequency is plotted in Fig. 3(c).
From (2) and (5), one can observe that, with equal transcon-
ductance, both the amplifiers display an identical quality factor
. However, , as a result the
CFA is not power matched to the antenna.
To correct the antenna mismatch, without altering the value
of passive components and , transistor in the CFA
is biased such that . From (4), one observes Fig. 4. Schematic and small-signal model of an SGTxFB amplifier.
that the new bias condition ensures . Fur-
thermore, since the quality factor of the matching network is
inversely proportional to the transconductance, the -boost re- is ignored. The transformer model of [12] is adopted and the
sults in a reduction in the quality factor. Therefore, as body effect is neglected. Initially, to simplify the mathematical
compared to the inductor-degenerated CSA, the CFA effectively analysis, inductors and are assumed to be ideal. How-
achieves a higher matching bandwidth at the expense ever, after deriving the model, a technique to include inductor
of higher current (assuming square-law devices). This non-idealities will also be described. Applying Kirchoffs cur-
relative performance is shown in Fig. 3(d). rent law (KCL) to the circuit in Fig. 4 yields
The desire to exploit current feedback to achieve a wide input
matching bandwidth motivates application of SGTxFB. The
SGTxFB amplifier in Fig. 4 contains two feedback loops. In (6a)
the first feedback loop, at frequency , senses the (output)
current flowing through and converts it to a voltage, which (6b)
controls the (input) of the device. In the second loop, (6c)
current through is fed back to the input via the antiphase (6d)
mutual magnetic coupling between and . In effect, the
transformer formed by and provides current-sense Solving (6a)(6d), the input admittance can be
current feedback. shown to be given by (6e) as follows:
To design a matching network using SGTxFB, an accurate
model of the input impedance of the amplifier as a function of
transformer and transistor parameters is derived in Section III.

III. SGTxFB-BASED MATCHING NETWORKS

Although receiver front-end circuitry realized with SGTxFB
amplifiers has been reported in recent literature [8][10], in-
sightful and compact analytic expressions that assist matching (6e)
network design and model the noise performance are yet to be
presented. This is primarily due to relatively open design space,
comprising multiple variables such as device transconductance
To verify (6e), a test circuit (TC1) with
, self-inductance , and mutual inductance . To
GHz mS pH was designed. A compar-
simplify the calculations, it is common to assume perfect mag-
ison of and obtained from circuit simulation and
netic coupling; a coupling coefficient of one.
from (6e) are plotted as a function of frequency in Fig. 5.
However, for a large turns ratio , a coupling co- However, while (6e) is an accurate and exact solution to the
efficient close to unity is difficult to achieve. To investigate the KCL equations, it fails to provide insight on how to select the
tradeoffs involved in the design of an SGTxFB, an input admit- component values to achieve the target admittance.
tance model is computed next. To simplify (6e), the design space must be constrained as to
reflect normal operating conditions. Toward this goal, a reso-
A. Input Admittance
nance condition is enforced: at frequency resonates
The small-signal model of the SGTxFB amplifier is presented with leakage inductance, . As mentioned earlier,
in Fig. 4. Transistor is assumed to have zero output conduc- includes a shunt capacitance , which can be appropriately
tance. In the small-signal model, is the parallel combination selected to ensure satisfies the resonance condition mathe-
of the gate-to-source capacitance of and an extra ca- matically described in (7) as follows:
pacitance . For the input admittance analysis, capacitance
, which appears in parallel with the ideal voltage source, , (7)
1350 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 3, MARCH 2013

B. Matching Network Design

Since input impedance of the SGTxFB amplifier is a function

of the transformer parameters, applying transformer feedback to
match the amplifier with its preceding driver stage is possible.
The driver stage could be an on-chip pre-amplifier, mixer, off-
chip transmission-line (T-line) driver, or antenna. To maximize
the power gain of the driver stage with an output conductance
, a matching network is designed to ensure .
In order to match the SGTxFB amplifier to a driver with purely
real admittance, all imaginary terms in (9b) must be eliminated.
Manipulating the design variables and to achieve
cancellation of imaginary terms is not possible due to the
already established resonance condition (7). Therefore, an
additional capacitance must be added in parallel to to
achieve . From (9b), it can be shown that

(10)

After the addition of , the input admittance of the circuit

is purely real and given by

(11)
Fig. 5. Real and imaginary admittance: model versus circuit simulations.

Further intuition regarding the impact of feedback for band-

width extension can be obtained by using (7) to reformulate
Using (7), the simplifications given in (8a)(8c) can be made (11), resulting as follows in (12) (derivation in Appendix II):

(12)

In (12), one should notice the expression derived is similar

(8a) to the input impedance of a CSA with inductor degeneration.
For the circuit in Fig. 3(a), from (1), the real component of the
input impedance is given by . Thus,
verifying that the narrowband inductor degeneration is a spe-
cial case of transformer feedback. In the SGTxFB amplifier, the
(8b) magnetic coupling between the windings of the transformer re-
duces the shunt input impedance and provides a wideband
match.
While the aforementioned results have been derived using an
(8c) ideal model inductor and , the results can be extended to
include the effects of a finite factor and self-resonance fre-
Finally, using (8a)(8c), a simpler expression for , (9a) and quency. The loss in can be modeled by a shunt resistance
(9b), , which appears in parallel with input resis-
tance derived in (12). The loss in can be modeled as a
series resistance . As a first-order approxima-
tion, can be absorbed into the amplifier transconductance
by defining an effective . Finally, the
(9a) parasitic capacitance associated with and the windings be-
tween and can also be absorbed in capacitances and
, respectively.
(9b)

C. Quality Factor
is derived. Equation (9b) has a clear physical interpretation: the
input impedance of an SGTxFB amplifier appears as a parallel A key metric for wideband matching networks is the factor.
circuit, where and are functions of . For a parallel RLC circuit, the quality factor is or
BHAGAVATULA AND RUDELL: ANALYSIS AND DESIGN OF TRANSFORMER-FEEDBACK-BASED WIDEBAND RECEIVER 1351

is accompanied by a reduction in , thereby reducing the

gain of the amplifier.
Using the values of obtained from (14) to solve for the
factor of the matching network (13), a plot of the factor as a
function of is plotted in Fig. 6(b). As expected from the result
given in (12), increasing of the transformer, while maintaining
a power match, results in a lower . In addition, it is interesting
to note that as increases, so too does the matching network ,
thereby requiring a high- transformer for a wideband match.

IV. TRANSCONDUCTANCE AND NOISE

A. Effective Transconductance
The bandwidth extension provided by transformer feedback,
described in Section III, is accompanied by a suppression of the
closed-loop effective transconductance of the
SGTxFB amplifier. can be derived using the small-signal
model described in Fig. 4. Assuming a transistor with infinite
output impedance, the short-circuit output current of the
amplifier is a product of and gate-to-source voltage
. The relationship between the voltage at the primary
and secondary of the transformer is derived in Appendix I.
Using (A.3) and (7), it can be proven that is inversely
Fig. 6. Design space for matching TC1 to a 50- source resistance with proportional to . As a result, at ,
mS mS mS mS mS and . (a) Transformer
turns-ratio. (b) Matching network factor. (15)

. From the input admittance expression in (9b), the With , the effective transconductance reduces
of the matching network can be proven to be to . This is similar to the inductively degenerated
matching network, Fig. 3(a), where
(13)
(16)
As described earlier, the factor and are func-
It is important to note, though is not an explicit function of
tions of . However, in order to design for
in (15), specifying implicitly constrains .
, only three among can be
For the test circuit TC1, the obtained from circuit simulation
uniquely specified. is primarily limited by the current
is compared with the analytic model (15) in Fig. 8(a).
budget of the target application and an upper limit on is
Further insight regarding the relationship between and
placed by the inductors self-resonance frequency. As a result,
can be obtained by considering the expression for , as
to obtain an optimal power match, the transformer parameters
given in (17). If the SGTxFB amplifier is perfectly matched to
cannot be selected independently. Using (11) as a
, using (9a) and (17), it is straightforward to prove that
starting point, the following interdependence between and
is the geometric mean of and
can be shown:

(14) (17)

(18)
Defining the design space of the circuit as all sets of
, which satisfy (14), the design space for the test Equation (18) offers intuition on the impact of SGTxFB
circuit TC1, with pH at GHz, is plotted in loading on the gain of a two-stage amplifier. Consider a
Fig. 6. For values of ranging from 25 to 125 mS, Fig. 6(a) two-stage amplifier, the first being the th stage followed
plots as a function of for an SGTxFB circuit designed to by a th stage. The input admittance of the th stage is
match a 50- driving source. Two important observations can designed to be wideband using transformer feedback in order
be made from this graph. First, for a fixed turns ratio , an to provide a wideband load to the th stage. The gain
amplifier with higher current (higher ) requires larger to of the th stage, transconductance load impedance ,
achieve a power match. This is important because obtaining a is inversely proportional to . However, from (18), or
high using spiral-inductor-based transformers for a non-unity the gain of th stage is directly proportional to . As a
turns ratio is quite challenging. Secondly, for a fixed coupling result, the cascaded gain of the th and th stage is only
coefficient, the power match achieved by increasing values of proportional to .
1352 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 3, MARCH 2013

Fig. 7. Small-signal model used for noise calculation.

B. Noise Contributions
To employ SGTxFB at the front-end of the receiver, the im-
pact of transformer feedback on the noise performance must
be considered. Two important noise contributors, thermal noise
from the source resistance and transistor ,
are considered in this section.
Assuming a driving source with an input resistance , noise
power , and an SGTxFB perfectly matched to the source
resistance, it is straightforward to prove that the available noise
power density is [13]. Using the effective transcon-
ductance defined previously, the source-induced current-noise
power density at the output of the SGTxFB amplifier
is given by Fig. 8. (a) Effective transconductance. (b) Transistor Thermal noise contribu-
tion in TC1.

(19)
C. NF
To analyze the current noise contributed by , con- The NF of the SGTxFB amplifier is given by
sider the small-signal model in Fig. 7. The source impedance
of the driver stage is modeled by the resistor . Due to (22)
transformer feedback, the thermal noise in the drain current
is coupled to the gate of the transistor. The resulting Substituting, (19) and (21) in (22),
gate voltage noise is amplified, inverted, and fed back to the
drain. The draincurrent noise and the current-noise fed back
via the SGTxFB are fully correlated. As a result, the inverted (23)
phase of the two noise components suppresses the output noise
contribution of . Applying KCL and Kirchoffs voltage law
(KVL), For a circuit that is perfectly matched to the source resistance
(conductance ), is governed by (18). Thus, with a
source noise power, , and drain thermal
(20a) noise current, ,
(20b)
(24)
(20c)

With the help of (A.1), (A.2), and (20a), it can be proven that
(derived in Appendix III) The NF of the SGTxFB is a function of and the trans-
former parameters . As a point of reference, consider a
(21) common-gate amplifier (CGA) and a resistor-terminated CSA
with and , re-
spectively. To highlight the noise contribution of the MOSFET,
a noiseless resistor termination has been assumed in the CSA.
To validate (21), the output thermal-noise power density for In a CGA, the transistor is uniquely specified by the ad-
test circuit TC1 is calculated and compared with the noise-sim- mittance of the driving source, i.e., . For a power-
ulation result. The results are shown in Fig. 8(b). At 12 GHz, the matched circuit, the minimum is independent of
output noise is modeled with an accuracy of 5% of the simu- and only a function of . From (24), one observes that
lation result. for all values of .
BHAGAVATULA AND RUDELL: ANALYSIS AND DESIGN OF TRANSFORMER-FEEDBACK-BASED WIDEBAND RECEIVER 1353

Fig. 9. Two-stage stagger-tuned IF-amplifier with SGTxFB driving the mixer

transconductance (MX ).

Fig. 10. Simulated frequency response of IFA1 and IFA2.

For a resistively terminated CSA, is inversely pro-
portional to , and therefore suffers from a noise versus power
tradeoff. In order to minimize the NF, the current must be maxi-
mized while maintaining power and linearity requirements. The
SGTxFB relaxes this tradeoff by introducing additional design
variables via the feedback transformer. However, a quali-
tative comparison between the relative noise performance of the
CSA and the SGTxFB amplifier is difficult because is also
dependent on the choice of .
In summary, analytic closed-form expressions for the input
admittance, quality factor, effective transconductance, and NF
have been derived in this section. As a test vehicle to validate the
results, the design of a three-stage SGTxFB-based IF down-con-
verter operating over a frequency range of 1113 GHz is de-
scribed next. The challenges involved in the design and phys-
ical implementation of a multistage SGTxFB is presented in
Section V.

V. CIRCUIT DESIGN

A. IF Amplifier (IFA)
The IFA consists of two stagger-tuned SGTxFB amplifiers,
IFA1 and IFA2. The circuit diagram of the IFA and mixer Fig. 11. Compact floor-plan for multiple transformer designs. (a) Transformer-
transconductance stage is shown in Fig. 9. In this circuit, coupled circuit. (b) Layout of multiple transformer-coupled stages. (c) Trans-
three transformers are included for bandwidth extension. The former-feedback circuit. (d) Layout of multiple transformer-feedback stages.
transformer-feedback network in IFA1 is designed to match
the amplifier to the 50- impedance of the off-chip measure-
ment circuitry. The overlay transformer in the input-matching Theoretically, the gain of a generic -stage amplifier can be
network has a coupling coefficient of 0.7 and is formed increased by increasing the number of stages cascaded. How-
using spiral inductors of 780 pH and 2 nH . ever, in the case of the SGTxFB-based IFA, cascading stages
At the interface between IFA1 and IFA2, is designed becomes challenging from the perspective of unwanted para-
to provide a wide-bandwidth high-gain load for IFA1. From sitic elements due to the rather complicated routing between
(14), it can be observed that to increase the input impedance transistors and transformers from stage to stage. This problem
of IFA2, the transformer turns ratio should be increased. Thus, is better illustrated by drawing a parallel between transformer-
and are 1-to-4 transformers to maximize the gain of coupled and transformer-feedback amplifiers. In the three-stage
IFA1 and IFA2. The simulated frequency response of IFA1 and transformer-coupled amplifier shown in Fig. 11(a), the output
IFA2 are shown in Fig. 10. The center frequencies of IFA1 and of stage 1 (drain of the amplifier) and the input of stage 2 (gate
IFA2 are tuned to 11 and 12 GHz, respectively, to reduce the of the amplifier) are completely isolated by transformer TR1.
in-band gain variation. Furthermore, to mitigate the impact of A popular technique to efficiently layout the cascaded config-
the cascode pole on the frequency response, transistor and uration relies on using the transformer to route the output of
are sized equally to allow a shared-junction layout. The one stage to the input of the next stage, hence the name trans-
gain of each stage of the amplifier varies by less than 1 dB over former-coupled amplifiers [14], [15]; this is shown in Fig. 11(b),
the 2 GHz of signal bandwidth. The two-stage IFA achieves a with the transformers between the active devices. The parasitic
peak gain of 19.2 dB while consuming 20 mA of current from routing from the amplifier and transformer can be tightly con-
a 0.9-V supply. trolled and minimized. Conversely, in the SGTxFB amplifier,
1354 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 3, MARCH 2013

Fig. 12. Quadrature down-conversion IF mixer.

Fig. 13. Layout of the three-winding transformer to couple the mixer transcon-
ductance and switching stages.
Fig. 11(c), both the primary and secondary of a given trans-
former must be routed to both the gate and source of one set of
C. Quadrature LO Generation
devices associated with a single stage. As a result, the standard
layout from Fig. 11(b) applied to SGTxFB results in a signif- To simplify the measurement setup, quadrature LO signals
icant amount of extra routing and a high corresponding para- for the mixer are generated on-chip using a single external
sitic capacitance. An alternate approach, presented in this work, 12-GHz sinusoidal signal source. Several active [17] and
minimizes the routing between the active devices and the trans- passive [18][20] techniques for generating quadrature signals
formers by placing the MOSFET devices of each cascaded stage have been proposed in the literature. Passive I/Q generation
in a centralized cluster, Fig. 11(d). All of the transformers are circuits including RC poly-phase filters and T-line-based
then wrapped around the clump of active devices, with the pri- branch line and Lange couplers are more favorable. However,
mary and secondary of each transformer oriented to facilitate both T-line and RC-based filters are associated with significant
transistor access. This compact layout for a cascaded SGTxFB design challenges. In cases where the C values are more than
optimizes the layout for minimal routing between stages; e.g., an order of magnitude higher than the layout parasitics, RC
between IFA1, IFA2, and the mixer transconductance stage. poly-phase filters are suitable for I/Q generation. In contrast,
T-line-based structures are better suited for frequencies above
60 GHz; frequencies at which the physical dimensions of the
B. IF Mixer T-line is sufficiently small to allow implementation on-chip. At
12 GHz, the RC poly-phase implementation is highly sensitive
The output of the IFA is mixed with a 12-GHz in-phase to parasitic routing capacitances, and the T-line-based quadra-
quadrature (I/Q) local oscillator (LO) signal and down-con- ture generation technique is area intensive. At the intersection
verted to baseband. The schematic of the IF mixer is shown of these two approaches, the lumped-element implementation
in Fig. 12. Two transformers, and , are included of a T-line-based I/Q generator is found to be the most optimal.
within the mixer for wideband down-conversion. The SGTxFB Two important lumped-element coupler topologies are the
network using transformer at the interface, between the branch-line and Lange coupler [20]. Branch-line couplers use
mixer transconductance-stage and IFA2, has been described only capacitive coupling, have narrower bandwidth, and require
large area to ensure zero magnetic coupling between the induc-
earlier. The three-winding transformer couples the
tors in the circuit. With the goal of optimizing area, a Lange cou-
output of the transconductance stage into the switching stage
pler has been implemented in this design. Lange couplers use
[16]. performs two important functions. First, inductance
both capacitive and inductive coupling for the I/Q generation.
resonates with the drain-to-bulk capacitance of , and
However, to ensure the amplitude of the I and Q outputs are per-
resonates with the device capacitance of the switching
fectly matched, a well-controlled magnetic coupling coefficient
transistors . Secondly, the isolation provided by TR4
is required between the inductors. Values of deviating
prevents the flow of dc current from into the switching from 0.707 increase the amplitude mismatch. The schematic for
transistors, thereby reducing the flicker-noise contribution of the Lange coupler and balun is shown in Fig. 14. From elec-
the switch at the baseband output. In addition, isolating the dc tromagnetic (EM) simulations, LO signals have quadrature ac-
current from the switching stage allows for a higher load-resis- curacy within 0.5 and an amplitude mismatch of less than
tance and gain. 0.8 dB.
The layout of is shown in Fig. 13. An overlay trans-
former structure was used to increase the coupling between the VI. MEASUREMENT RESULTS AND COMPARISON
inductor pairs , and , . Compared to the switching The chip [21] was fabricated in a six-metal-layer 40-nm
stage, higher current flows through the transconductance stage, CMOS process with a top level ultra-thick metal (UTM) layer.
thus has been implemented in an UTM layer. and The die photograph is shown in Fig. 15 and occupies an area
carry significantly lower switching current and have been im- of 1 mm 0.6 mm (pads included). The three transformers
plemented in the aluminium passivation (AP) layer. used for SGTxFB, the three-winding transformer in the mixer,
BHAGAVATULA AND RUDELL: ANALYSIS AND DESIGN OF TRANSFORMER-FEEDBACK-BASED WIDEBAND RECEIVER 1355

Fig. 14. Transformer-based lumped-element Lange coupler.

Fig. 16. (a) Input matching (dB). (b) IF-section down-conversion gain.

Fig. 15. Chip micrograph.

and the Lange-coupler-based quadrature generator circuit are

highlighted on the chip micrograph. On-chip wafer probing
was done to measure the performance. A balun probe provides
a differential input signal; and a single-ended off-chip 12-GHz
LO signal drives the I/Q generation circuitry.
The chip consumes 30 mA of current from a 0.9-V supply.
The input matching ( ) of the SGTxFB amplifier is shown
in Fig. 16(a). The matching bandwidth ( dB) is
1 GHz. The conversion gain of the IF section is plotted as a
function of frequency in Fig. 16(b). The measured peak conver-
sion gain is 27.6 dB and the 3-dB bandwidth is 2.1 GHz. The
center frequency of the IF section is at 11.6 GHz; a 400-MHz
offset from the desired frequency. The NF is measured using the
test setup shown in Fig. 17(a). Over the baseband signal band-
width of 1.08 GHz, the total NF variation is less than 0.8 dB with
a peak NF of 6.1 dB and a minimum of 5.3 dB. The linearity
of the receiver is characterized by a two-tone test. The mea-
sured third-order intermodulation intercept point (IIP3), with Fig. 17. (a) NF measurement setup. (b) NF versus frequency.
two tones at 10-MHz offset from 12 GHz, is 22 dBm.
The performance of the transformer-feeedback-based wide-
band receiver has been compared with high fBW -band, single-ended LNA presented in [23] uses shunt peaking to
-band and 60-GHz direct-conversion receivers in Table I. achieve an fBW of 8%. In addition, the inductor-degenerated
The 30% fBW UWB pulsed-radar proposed in [22] em- input matching-network results in narrowband input power
ploys high-order LC bandpass filters for input and inter-stage match. References [24] and [25] propose wideband amplifiers
matching. As a result, the two-stage LNA occupies an on-chip using capactively coupled and magnetically coupled resonant
area of 0.93 mm (estimated from a die micrograph), ap- tanks, respectively. While [24] achieves a gain, fBW, and NF
proximately 8 larger than the area consumed by the three similar to this work, [25] achieves double the fBW. However, it
transformers , , and in Fig. 15. The two-stage is important to note that [25] only includes a single-stage LNA
1356 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 3, MARCH 2013

TABLE I
PERFORMANCE SUMMARY AND COMPARISON WITH PRIOR ART

and has 9-dB lower gain. Cascading multiple stages to enhance APPENDIX II
the gain would result in a reduction of the bandwidth. From (9a), input admittance of the SGTxFB amplifier at the
resonant frequency is
VII. CONCLUSION
Analytic expressions for the input admittance, factor, and (A.5)
NF of SGTxFB amplifiers have been derived as a function of de-
sign variables . The impact of high and low
on the factor of the matching network has been described. Rearranging the terms in (A.5) and with ,
Using transformer-feeedback-based bandwidth extension tech-
niques, a 16% fBW IF section consisting of a two-stage stagger- (A.6)
tuned IF-amplifier, a transformer-coupled quadrature mixer, and
a Lange coupler has been presented. The challenges associated From (A.6) and , it can be shown
with the layout in multistage SGTxFB are highlighted and a that
strategy for compact layout has been proposed.

APPENDIX I (A.7)

The primary and secondary transformer currents can be ex-

pressed as a function of and using (6a) to obtain (A.1) and
APPENDIX III
(A.2),
The output noise analysis is performed using the small-signal
(A.1) model shown in Fig. 7. The transformer current (A.1) and (A.2)
are valid even for this model. Substituting (A.1) in (20a),
(A.2)
(A.8)
From (A.1) and (6a), the relation between and can be
expressed as
At , with the help of (7) in (A.8) can be simplified to

(A.9)

(A.3)
Next, from (20a) and (A2),

(A.10)
Substituting (A.2) in (6a),
Using , the terms in (A.10) can be rearranged to
express as a function of . Again, using (7), it can be shown
that
(A.4)

(A.11)
Finally, substitute (A.3) in (A.4) to obtain (6e).
BHAGAVATULA AND RUDELL: ANALYSIS AND DESIGN OF TRANSFORMER-FEEDBACK-BASED WIDEBAND RECEIVER 1357

Now, with the help of (21), (A.9), and (A.11), is ex- [7] A. Bevilacqua and A. M. Niknejad, An ultrawideband CMOS low-
pressed as a function of only in (A.12c), noise amplifier for 3.110.6-GHz wireless receivers, IEEE J. Solid-
State Circuits, vol. 39, no. 12, pp. 22592268, Dec. 2004.
[8] M. T. Reiha and J. R. Long, A 1.2 V reactive-feedback 3.110.6 GHz
low-noise amplifier in 0.13 m CMOS, IEEE J. Solid-State Circuits,
vol. 42, no. 5, pp. 10231033, May 2007.
[9] M. Khanpour, K. W. Tang, P. Garcia, and S. P. Voinigescu, A wide-
(A.12a) band -band receiver front-end in 65-nm CMOS, IEEE J. Solid-State
Circuits, vol. 43, no. 8, pp. 17171730, Aug. 2008.
[10] A. C. Heiberg, T. W. Brown, T. S. Fiez, and K. Mayaram, A 250
mV, 352 W GPS receiver RF front-end in 130 nm CMOS, IEEE J.
Solid-State Circuits, vol. 46, no. 4, pp. 938949, Apr. 2011.
[11] V. Bhagavatula and J. C. Rudell, Transformer-feedback based CMOS
(A.12b) amplifiers, in IEEE Int. Circuits Syst. Symp., 2012, pp. 237240.
[12] J. R. Long, Monolithic transformers for silicon RF IC design, IEEE
J. Solid-State Circuits, vol. 35, no. 9, pp. 13681382, Sep. 2000.
[13] J. C. Rudell, J. A. Weldon, J. J. Ou, L. Lin, and P. Gray, An integrated
GSM/DECT receiver: Design specifications, Univ. of California at
Berkeley, Berkeley, CA, USA, ERL Memo UCB/ERL M97/82, 1997.
[14] M. Boers, A 60 GHz transformer coupled amplifier in 65 nm digital
CMOS, in IEEE RFIC Symp. Tech. Dig., 2010, pp. 343346.
[15] D. Chowdhury, P. Reynaert, and A. M. Niknejad, Design considera-
tions for 60 GHz transformer coupled CMOS power amplifiers, IEEE
(A.12c) J. Solid-State Circuits, vol. 44, no. 10, pp. 27332744, Oct. 2009.
[16] J. Paramesh, R. Bishop, K. Soumyanathan, and D. J. Allstot, A four-
antenna receiver in 90-nm CMOS for beamforming and spatial diver-
Since the SGTxFB amplifier is perfectly matched to the sity, IEEE J. Solid-State Circuits, vol. 40, no. 12, pp. 25152524, Dec.
source resistor , using from (9a), it can be shown 2005.
that [17] A. Rofougaran, J. Rael, M. Rofougaran, and A. Abidi, A 900 MHz
CMOS LC-oscillator with quadrature outputs, in IEEE Int. Solid-State
Circuits Tech. Dig., 1996, pp. 392393.
(A.13) [18] F. Behbahani, Y. Kishigami, J. Leete, and A. A. Abidi, CMOS mixers
and polyphase filters for large image rejection, IEEE J. Solid-State
Circuits, vol. 36, no. 6, pp. 873887, Jun. 2001.
[19] R. C. Frye, S. Kapur, and R. C. Melville, A 2-GHz quadrature hybrid
implemented in CMOS technology, IEEE J. Solid-State Circuits, vol.
Finally, to compute the noise power, 38, no. 3, pp. 550555, Mar. 2003.
[20] D. Ozis, J. Paramesh, and D. J. Allstot, Integrated quadrature couplers
and their applications in image-reject receivers, IEEE J. Solid-State
Circuits, vol. 44, no. 5, pp. 14641476, May 2009.
(A.14) [21] V. Bhagavatula, M. Boers, and J. C. Rudell, A transformer-feedback
based wideband IF-amplifier and mixer for a heterodyne 60 GHz re-
ceiver in 40 nm CMOS, in IEEE RFIC Symp. Tech. Dig., 2012, pp.
167170.
[22] V. Jain, S. Sundararaman, and P. Heydari, A 2229-GHz UWB pulse-
radar receiver front-end in 0.18- m CMOS, IEEE Trans. Microw.
Theory Techn., vol. 57, no. 8, pp. 19031914, Aug. 2009.
ACKNOWLEDGMENT [23] A. Mazzanti, M. Sosio, M. Repossi, and F. Svelto, A 24 GHz sub-
harmonic direct conversion receiver in 65 nm CMOS, IEEE Trans.
The authors would like to thank M. Boers, P. Sen, Circuits Syst. I, Reg. Papers, vol. 58, no. 1, pp. 8897, Jan. 2011.
[24] F. Vecchi et al., A wideband receiver for multi-Gbit/s communica-
D. J. Allstot, E. Pepine, T. Zhang, S. Dey, and A. Ravish tions in 65 nm CMOS, IEEE J. Solid-State Circuits, vol. 46, no. 3, pp.
Suvarna for many helpful discussions and comments. 551561, Mar. 2011.
[25] C. H. Li, C. N. Kuo, and M. C. Kuo, A 1.2-V, 5.2-mW, 2030 GHz
wideband receiver front-end in 0.18 m CMOS, IEEE Trans. Microw.
Theory Techn., vol. 60, no. 11, pp. 35023512, Nov. 2012.
REFERENCES
[1] E. Laskin, M. Khanpour, S. T. Nicholson, A. Tomkins, P. Garcia, A.
Cathelin, D. Belot, and S. P. Voinigescu, Nanoscale CMOS trans-
ceiver design in the 90170-GHz range, IEEE Trans. Microw. Theory
Techn., vol. 57, no. 12, pp. 34773490, Dec. 2009. Venumadhav Bhagavatula (S10) received the
[2] K. Okada et al., A 60-GHz 16QAM/8PSK/QPSK/BPSK direct-con- B.E. degree in electronics and communication
version transceiver for IEEE802.15.3c, IEEE J. Solid-State Circuits, engineering from the University of Delhi, Delhi,
vol. 46, no. 12, pp. 29883004, Dec. 2011. India, in 2005, the M.Tech. degree in electronic
[3] A. Siligaris et al., A 65-nm CMOS fully integrated transceiver module design engineering from the Indian Institute of
for 60-GHz wireless HD applications, IEEE J. Solid-State Circuits, Science, Bangalore, India, in 2007, and is currently
vol. 46, no. 12, pp. 30053017, Dec. 2011. working toward the Ph.D. degree at the University
[4] A. Arbabian, S. Callender, S. Kang, B. Afshar, J.-C. Chien, and A. M. of Washington (UW), Seattle, WA, USA.
Niknejad, A 90 GHz hybrid switching pulsed-transmitter for medical He has held an internship with the Broadcom
imaging, IEEE J. Solid-State Circuits, vol. 45, no. 12, pp. 26672681, Corporation, Irvine, CA, USA, where he was in-
Dec. 2010. volved with wideband millimeter-wave receivers.
[5] J. W. May and G. M. Rebeiz, Design and characterization of -band Prior to joining the UW, he was an IC Designer with the Audio-Circuits Group,
SiGe RFICs for passive millimeter-wave imaging, IEEE Trans. Mi- Cosmic Circuits Pvt. Ltd., Bangalore, India. His research interests include
crow. Theory Techn., vol. 58, no. 5, pp. 14201430, May 2010. RF/millimeter-wave and low-power mixed-signal circuits.
[6] A. Ismail and A. A. Abidi, A 310 GHz low-noise amplifier with Mr. Bhagavatula was the recipient of the 2007 CEDT Design Medal of the
wideband LC-ladder matching networks, IEEE J. Solid-State Circuits, Indian Institute of Science and the 2012 Analog Devices Outstanding Student
vol. 39, no. 12, pp. 22692277, Dec. 2004. Designer Award.
1358 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 3, MARCH 2013

Jacques C. Rudell (S94M00SM09) received Engineering (CSNE), University of Washington. His research interests are
the B.S. degree in electrical engineering from The related to topics in the area of analog, RF, and mixed-signal systems. His
University of Michigan at Ann Arbor, Ann Arbor, recent areas of interest include RF and millimeter-wave circuits, in addition to
MI, USA, and the M.S.E.E. and Ph.D. degrees from interface electronic solutions toward biomedical applications.
the University of California at Berkeley, Berkeley, Dr. Rudell is a member of Tau Beta Pi and Eta Kappa Nu. He was on the
CA, USA. Technical Program Committee for the International Solid-State Circuits Con-
From 1989 to 1991, he was an Integrated Circuit ference (ISSCC) from 2003 to 2011. He is currently the 2013 general chair
(IC) Design Engineer with Delco Electronics (now for the IEEE Microwave Theory and Techniques Society (IEEE MTT-S) In-
Delphi), where his research focused mainly on ternational Microwave Symposium (IMS) and Radio Frequency Integrated Cir-
bipolar analog circuits for automotive applications. cuits (RFIC) Symposium. He is an associate editor for the IEEE JOURNAL OF
From late 2000 to 2001, he was a Postdoctoral SOLID-STATE CIRCUITS. He was the recipient of the 2000 Demetri Angelakos
Researcher with the University of California at Berkeley, in addition to holding Memorial Achievement Award, a citation given to one student per year by the
consulting positions with several Silicon Valley firms. In late 2001, he joined Electrical Engineering and Computer Science Department, University of Cal-
Berkana Wireless (now Qualcomm), San Jose, CA, USA, as an Analog/RF IC ifornia at Berkeley. He was the recipient of the 1998 ISSCC Jack Kilby Best
Design Engineer, and later became the Design Manager of the Advanced IC Student Paper Award. He was the corecipient of the 2001 ISSCC Lewis Best
Development Group. From September 2005 to December 2008, he was with Paper Award and the Best Student Paper award of the 2011 RFIC Symposium.
the Advanced Radio Technology (ART) Group, Intel Corporation. In 2009, He was also the recipient of an award presented at the 2008 ISSCC for Best
he joined the Faculty of Electrical Engineering as an Assistant Professor with Evening Session.
University of Washington, Seattle, WA, USA. He is currently a member of
the National Science Foundation (NSF) ERC Center for Sensorimotor Neural