See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/336028861

Theory and Design of Impedance Matching Network Utilizing a Lossy On-Chip

Transformer

Article in IEEE Access · September 2019

DOI: 10.1109/ACCESS.2019.2943512

CITATIONS READS

11 3,797

2 authors:

Trinh Van Son Jung-Dong Park

Dongguk University Dongguk University
14 PUBLICATIONS 61 CITATIONS 66 PUBLICATIONS 1,258 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Trinh Van Son on 04 March 2020.

The user has requested enhancement of the downloaded file.

Received September 3, 2019, accepted September 20, 2019, date of publication September 24, 2019,
date of current version October 9, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2943512

Theory and Design of Impedance Matching

Network Utilizing a Lossy On-Chip Transformer
VAN-SON TRINH , (Student Member, IEEE), AND JUNG-DONG PARK , (Senior Member, IEEE)
Division of Electronics and Electrical Engineering, Dongguk University, Seoul 04620, South Korea
Corresponding author: Jung-Dong Park ([email protected])
This work was supported in part by the National Research Foundation of Korea (NRF) through the Korean Government (MSIP) under
Grant 2018R1C1B5045481, in part by the Korea Institute of Energy Technology Evaluation and Planning (KETEP), and in part by the
Ministry of Trade, Industry and Energy (MOTIE), South Korea, under Grant 20194030202320.

ABSTRACT In this paper, we present a study on a transformer-based impedance matching network. We use
a simplified transformer model comprising two magnetically coupled coils, which are driven by a source
and terminated by a load. The formulae of the load and the source impedance for conjugate matching of
both sides of the transformer are presented, and a figure of merit is proposed for the evaluation of the
power transfer efficiency of the transformer under conjugate matching conditions. Analytical expressions are
provided for constructing the widely used transformer network consisting of a resistive load and a parallel
tuning capacitor. To verify the proposed work, we examined various on-chip transformers implemented
in 0.18 µm CMOS technology. Simulation and measurement results for a matching network synthesized
using the aforementioned analytical expressions corresponded well with the result of analysis for operating
frequencies up to 72% of the self-resonant frequency of the transformer. The presented results confirm that
the proposed analytical formulae based on the simplified transformer model are useful for the design and
optimization of transformer-based impedance matching networks in the microwave and millimeter-wave
regimes.

INDEX TERMS CMOS technology, impedance matching, power efficiency, transformers.

I. INTRODUCTION the transformer should be optimally designed to provide max-

At present, numerous radio frequency integrated cir- imum power transfer from the source to the load. It is well
cuit (RFIC) designs are being implemented with transformers known that the input and output ports of a transformer should
for various purposes such as impedance matching, impedance be simultaneously conjugate matched to the source and load
transformation, and signal conversion between single-ended impedances [16]. The general solution of the load and the
and differential signals in various frequency bands ranging source to the simultaneous conjugate match for a two-port
from the radio frequency (RF) to terahertz regimes. Trans- network using Z-parameters was presented originally in [17],
formers are used in many applications, including power or recently in [18]. However, little analytical details can be
amplifiers, low-noise amplifiers, voltage-controlled oscilla- seen for specific structures used in RF circuit design such
tors, mixers, and power-combining circuits [1]–[9]. There- as the impedance matching network consisting of a lossy
fore, many studies have been published on modeling and transformer and a shunt tuning capacitor for calculating the
analyzing transformers, new transformer structures, and new optimum source and load impedances. Relevant work on syn-
methods for enhancing transformer efficiency [10]–[15]. thesizing lossless impedance matching networks was given
However, most of these studies are not only complex but in [19] which gave the detailed analysis and synthesis of the
also application-specific, making it difficult to directly apply lossless matching network based on the transmission phase
the proposed techniques to the practical design of a highly shift from a given source to a load. Nevertheless, the work
efficient transformer. did not cover the impedance matching network using a lossy
In particular, when we use a transformer for a specific pur- transformer.
pose such as impedance transformation or signal balancing, Considerable effort has been devoted to optimizing the
The associate editor coordinating the review of this manuscript and
efficiency of transformers. In [20], the authors presented
approving it for publication was Bora Onat. formulae for a series load (a resistor in series with a capacitor)

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
140980 VOLUME 7, 2019
V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

for achieving the maximum efficiency. However, they can- between two or more conductors, which are called windings.
not be extended to a transformer-based matching network Notably, when we implement a winding transformer on a
with a resistive load in parallel with a capacitor, which is silicon substrate, the series resistance of each winding is
widely used in RFICs [21]–[24]. In the wireless power trans- quite significant because of the fabrication of the windings
fer (WPT) field, an inductive power link involves a load with a on relatively thin metal layers within back-end-of-the-line
parallel tuning capacitor; the load resonates with the inductive (BEOL) dielectric layers, and the skin effect in the metal
part of the transformer [25]. Although optimum load equa- windings in modern silicon technologies.
tions have been proposed, the assumptions made for deriv- Several secondary effects, such as capacitive coupling and
ing them render them impractical for a typical transformer. magnetic coupling to the substrate, lower the quality factor
In addition, these studies have not considered impedance Q of each coil. For the accurate modeling of the on-chip
matching at the source, which is as important as impedance transformer, an enormously large number of lumped-element
matching at the load for power transfer from the source to the parameters should be considered [27]. Therefore, this type
load. Based on the work in [17], a comprehensive solution of sophisticated modeling might not be quite appealing in
of the load and the source for inductively coupled coils for the early stages of design optimization. Instead, we extracted
the wireless power transfer application was presented in [26]. the effective transformer parameters for the two magnetically
However, it did not cover the transformer-based impedance coupled coils from Z-parameters, and the extracted param-
matching network with a shunt tuning capacitor which has eters were then used in the derived analytical formulae to
been widely used in RFIC design. design a simultaneous conjugate matching network, which
In this study, we developed a systematic approach to the demonstrated a promising accuracy of the maximum power
design and analysis of a transformer-based impedance match- transfer by the frequency response up to 72 % of the self-
ing network. We derived general conditions for the source resonant frequency (SRF) of the transformer within a percent-
and load from the characteristic parameters of the transformer age error of 10 %.
for optimal power transfer. For a simplified transformer Typically, a transformer can be considered as two magnet-
model involving two magnetically coupled coils, analytical ically coupled coils, as shown in Fig. 1. In this simplified
equations were derived for simultaneous conjugate match- model, the transformer is characterized by only five param-
ing, and the transformer parameters were extracted from eters: the series resistances (R1 and R2 ), the inductances
Z-parameters. On the basis of an analysis of the maximum (L1 and L2 ) of the primary and secondary windings, and
power transfer condition, the product of the coupling coef- the coupling coefficient, which indicates the strength of the
ficient k and the quality factors of the primary (Q1 ) and magnetic coupling between the two windings [27]. The turn
secondary (Q2 ) windings k 2 Q1 Q2 was used as a figure of ratio between the two windings is defined as n = L1 /L2 . It
merit to evaluate the quality of the transformer. is noteworthy that the parasitic coupling capacitance between
This paper consists of five sections. We present a detailed the two coils was neglected in this work since the complexity
analysis of the impedance matching of the on-chip trans- of the model and the analysis were considerably increased
former based on two magnetically coupled coils in Section II. while the effect is marginal at the frequency of interest (below
To demonstrate the validity and applicability of the derived SRF) when it was considered. (Some of the effects of this
equations for various transformer parameters, a typical coupling capacitance can be found in [28].) The simple low-
2:1 on-chip transformer designed in a 0.18 µm CMOS pro- frequency model has been widely used as a core circuit to
cess was evaluated with the High Frequency Structure Simu- characterize transformers in many studies (e.g., [10]–[15])
lator (HFSS). A comparison of the calculated values with the since the physical size of transformers is usually designed to
simulation results is presented in Section III to demonstrate be noticeably less than the guided wavelength at the operating
the validity of the presented analysis. Next, in Section IV, frequency [27]. Therefore, it can reflect dominant physical
we present a design for impedance matching with a trans- phenomena occurring in a transformer with an inductance and
former containing a parallel tuning capacitor at the load magnetic coupling of the windings at operating frequencies
and source, along with an analysis of the impedance match-
ing. In Section V, the fabrication of the on-chip transformer
(mentioned in Section III) in a 0.18 µm CMOS process is
discussed, and our works are compared with measurement
results to verify the applicability of the proposed work on the
on-chip transformer to RFIC design in the gigahertz regime.
Finally, conclusions are provided in Section VI.

II. IMPEDANCE MATCHING FOR A GENERAL LOAD

A. LOW-FREQUENCY TRANSFORMER MODEL WITH TWO
MAGNETICALLY COUPLED COILS
In a passive transformer, an input signal or input power from FIGURE 1. Low-frequency transformer model with two magnetically
the source is transferred to the load by the magnetic coupling coupled coils with load and source terminations.

VOLUME 7, 2019 140981

 V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

well below the SRF of the transformer. The validity of a

low-frequency model is discussed in Section III.

B. IMPEDANCE MATCHING WITH TWO MAGNETICALLY

COUPLED COILS
An important criterion that designers should consider when
using a transformer is the amount of power delivered from
the source to the load, especially in applications such as
WPT [3], [25], and impedance transformation in power
amplifiers [1], [2], [4], [21], [22], [29]. We can assess the
efficiency of the network with regard to the power from the FIGURE 2. Maximum transducer power efficiency (GTmax [%]) versus
k 2 Q1 Q2 .
transducer power gain GT , which is defined as the ratio of
the power delivered to load PL to the power available from
source Pavs [16]:
secondary inductor (L2 ). The quality factors Q1 and Q2 of the
PL PL Pin
GT = = = η(1 − |0s |2 ), (1) primary and secondary inductors are calculated as
Pavs Pin Pavs
ωL1 ωL2
where Pin is the input power delivered from the source to Q1 = ; Q2 = . (6)
the network (including the transformer and load) and η = R1 R2
PL /Pin is the power efficiency (operating power gain) of the As shown in Appendix A, simultaneous conjugate match-
network [20]. 0S is the source reflection coefficient, and it is ing on both sides of the transformer can be achieved either
given by when the transformer is ideal (R1 = R2 = 0) or when ZS and
Zin − ZS∗ ZL satisfy the following conditions:
0S = , (2)
Zin + ZS 
XL = −ωL2 , XS = −ωL1 (7.1)
p p
where ZS and Zin are the source and network impedances, RL = R2 1 + k Q1 Q2 , RS = R1 1 + k Q1 Q2 . (7.2)
2 2
respectively. Thus, impedance matching is important to
obtain an optimal transducer power gain for a given pas- The equations in (7) are the derivation of the general con-
sive network. On the source side, Zin should be the con- jugate matching condition based on Z-parameters presented
jugate of the source impedance: Zin = ZS∗ . Similarly, in [17] for the transformer model with the impedance matrix
the load impedance ZL should be the conjugate of the output in (4).
impedance Zout on the load side: ZL = Zout ∗ . The former

condition maximizes the power delivered from the source to C. FIGURE OF MERIT FOR A TRANSFORMER, OBTAINED
the network when the source is given, and the latter condition FROM THE TRANSDUCER POWER GAIN
maximizes the power delivered to the load when the source Under the simultaneous conjugate matching conditions on
and transformer are given. both sides of the transformer, the maximum of the transducer
We can consider the transformer as a two-port network power gain GTmax is calculated in Appendix B as
driven by a source and terminated by a load, as shown p
in Fig. 1. The transformer is modeled using two magnetically k 2 Q1 Q2 + 1 − 1
GT max = 1 − 2 . (8)
coupled inductors, and the load and the source impedances k 2 Q1 Q2
are given by
It is noteworthy that because the optimal transducer power
ZL = RL + jXL and ZS = RS + jXS , (3) gain in (8) is obtained under the condition 0s = 0, it even-
tually equals the optimal power efficiency derived in [20],
where RL and XL are the equivalent series resistance and which is presented in (1). Fig. 2 shows the maximum trans-
reactance of the load, respectively, while RS and XS are ducer power gain as a function of k 2 Q1 Q2 . (The simulation
the equivalent series resistance and reactance of the source, and measurement setups are described in Sections III and V,
respectively. The relationship between the input (V1 ) and respectively.) Clearly, GTmax increases rapidly as k 2 Q1 Q2
output (V2 ) voltages for the input and output currents, given increases in the low-value region of the x-axis and saturates at
by I1 and I2 , respectively, can be written as a sufficiently large value. Therefore, k 2 Q1 Q2 is a reasonable
candidate for the figure of merit, which can be used to assess
    
V1 (R1 + jωL1 ) −jωM I1
= , (4) the quality of the designed transformer.
V2 jωM −(R2 + jωL2 ) I2
V2 = ZL I2 , (5) For a given transformer, the load and source impedances
calculated from (7) can be used for performing simultaneous
where ω is the angular frequency, and M (= k(L1 L2 )1/2 ) is the conjugate matching for both source and load (0L = 0S = 0),
mutual inductance between the primary inductor (L1 ) and the which would maximize the transducer power gain.

140982 VOLUME 7, 2019

V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

FIGURE 3. On-chip 2:1 transformer structure used for 3D electromagnetic

simulation: (a) the front face, (b) a 3D view in HFSS, and (c) a side view of
the layer stacks used in HFSS.

III. VERIFICATION OF THE PROPOSED ANALYSIS WITH

AN ON-CHIP TRANSFORMER IN A 3D EM SIMULATION
In this section, we discuss the examination of a typical on-
chip winding transformer (presented in Fig. 3) that is widely
used for interstage matching in millimeter-wave circuits in
silicon technologies [1], [2], [29]–[31]. In this study, an on-
chip transformer was designed in a 0.18 µm CMOS pro- FIGURE 4. Simulated (dashed line) and measured (solid line) values of
cess by using an 8.625 µm thick back-end-of-line (BEOL) the effective parameters of the on-chip 2:1 transformer.
stack and a 300 µm thick silicon substrate. The complex
dielectric stacks from the BEOL process were simplified into
14 equivalent dielectric layers, each of which was calculated
on the basis of a series capacitance approximation verified in
previous works [2], [4], [32], [33]. The conductivity of the
substrate was set to 10 S/m. An on-chip winding transformer
with a side-coupled structure was implemented using a 2 µm
ultra-thick metal. The primary winding inductor had one turn,
and its width was set to 8 µm, while the secondary inductor
had two turns and was 6 µm wide. The gap between two
adjacent turns was set to 3 µm, and the inner diameter (din )
of the octagonal winding inductor was 90 µm. In microwave
inductor and transformer design, a definite return path is
necessary to achieve good agreement between the HFSS
simulation and measurements. Therefore, we placed ground
tiles all over the area, which was modeled as a square ring
around the transformer, as shown in Fig. 3. FIGURE 5. Simulation results for several power gains of the 2:1
We performed 3-D electromagnetic (EM) simulations transformer as a function of the operating frequency.
with HFSS for the transformer design to attain its
two-port Z-parameters. Subsequently, the parameters of the
low-frequency transformer model R1 , R2 , L1 , L2 , and k were Fig. 5 shows the simulation results for various power
extracted as below (depicted in Fig. 4) gains of the on-chip transformer when the source and load
√ impedances were conjugately matched at 10 GHz. The
Im {zii } Im {z12 } × Im {z21 }
Ri = Re {zii } ; Li = ; k=√ analytical expressions given in (7) were used to achieve
ω Im {z11 } × Im {z22 }
simultaneous conjugate matching. Here, GTmax (or Gmax ) is
(9)
the maximum transducer power gain with 0S = S11 ∗ and

where i = {1, 2}; zij with i, j = {1, 2} is the element of the 0L = S22 ; GA (= Pavn /Pavs ) is the available power gain,
∗

extracted Z-matrix. which is defined as the ratio of the power available from
The simulated SRF was 51.8 GHz for the designed the transformer (Pavn , or the maximum power that can be
2:1 on-chip transformer. The effective inductance of the two delivered to the load) to the power available from the source;
windings increased rapidly near the SRF because of the reso- and GP (= PL /Pin ) is the operating power gain (or the power
nance between the winding inductor and the parasitic capac- efficiency η of the transformer in [20]) [16]. As evident,
itances seen from the winding inductor. When the operating all the mentioned power gains had the same value when
frequency exceeded the SRF, the effective reactance of each both source and load were simultaneously conjugate matched
inductor became capacitive. at 10 GHz. Fig. 5 presents the simulation results of the

VOLUME 7, 2019 140983

 V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

FIGURE 6. Plot of optimum power efficiency versus frequency for the

2:1 transformer with SRF = 51.8 GHz.

S-parameters. S11 on the source side and S22 on the load side
were less than −30 dB, while S21 reached its maximum value
at 10 GHz. FIGURE 7. Impedance transformation in a transformer with a load, series
Fig. 6 shows a comparison between GTmax calculated from tuning capacitors, and an extra tuning capacitor in parallel to the load.
(8) (which is also the maximum power efficiency ηmax ) and
GTmax simulated with Spectre TM ; both parameters are plotted
against the normalized frequency (fnor = fo /SRF) for SRF =
51.8 GHz. We determined the percentage error, defined as
the difference between the simulated and calculated values
divided by the simulated value, to quantify the limitation
of the low-frequency model. At a relatively low frequency
(fnor < 0.72 or f < 37.3 GHz) compared with the SRF of
the transformer, the calculated GTmax matched well with the FIGURE 8. Configuration for impedance transformation in a transformer
with a tuning capacitor in parallel with the load, and the series equivalent
simulated GTmax within an error of 10%. circuit of the configuration.
It is quite encouraging that such a simple low-frequency
model can provide promising results in a frequency range
of around 72% of the SRF of the transformer coil. As the transformer network with three additional capacitors. Such
operating frequency approaches the SRF of a transformer, a network is shown in Fig. 7(b). The series capacitor CL in
the parasitic coupling capacitances between the primary and series with the load is tuned to maximize the transformer’s
the secondary windings and between them and the substrate power efficiency, and another capacitor COUT in parallel
start to play a vital role [27]. Furthermore, with an increase with the load is used to reduce the turn ratio n, which
in the operating frequency, the mutual resistance associated makes it possible to use a transformer with a reasonable turn
with the eddy currents induced by the coupled magnetic ratio while keeping the PER unchanged. On the source side,
flux of another inductor increases the power loss [34]–[36]. a shunt capacitor CS is added to adjust the input reactance
Therefore, the operating frequency of the on-chip transformer to the desired value. More recently, a configuration with a
is typically chosen to be well below the SRF since the input tuning capacitor in parallel with the load has been widely
and output impedances of the transformer change drastically used [21]–[24]; the configuration is shown in Fig. 8. In this
around the SRF. Accordingly, the low-frequency model used configuration, the parallel capacitor can increase the effi-
in this study is acceptable for this typical case. ciency of the transformer (if it is tuned to a proper value)
as well as transform the equivalent series load resistance to
IV. TRANSFORMER MATCHING CONFIGURATION WITH Req <RL , which relaxes the requirement for the turn ratio n.
A PARALLEL TUNING CAPACITOR Moreover, when an active device is used as the load, the tun-
A. TRANSFORMER MATCHING NETWORK ing capacitor can absorb uncalculated parasitic capacitances
CONFIGURATIONS of the device.
Fig. 7(a) shows a compact lumped circuit model for a typ- Although this configuration with a parallel capaci-
ical monolithic transformer derived from two magnetically tor (Fig. 8) has been widely used in implementing a
coupled coils and intended for operation at a relatively low transformer-matching network, to the best of the authors’
frequency compared with the SRF [27]. Aoki et al. trans- knowledge, no analytical design equations have been reported
formed this low-frequency model into a T-model, and they so far. Such equations must be readily applicable in the hand
used the T-model for investigating the power efficiency (η = calculation at the early stage of design. It should be noted that
PL /Pin ) as well as the power enhancement ratio (PER) of a the configuration in Fig. 8 is different from that in Fig. 7(b).
typical transformer for impedance transformation [20]; they When we transform the parallel network configuration (RL
presented conditions for the optimal power efficiency of a in parallel with CL or a tuning capacitor Ct ) in Fig. 7 to

140984 VOLUME 7, 2019

V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

the series equivalent circuit (Req in series with Ceq ),

it is easy to confuse the parallel configuration (shown
in Fig. 8) with its series counterpart (shown in Fig. 7b).
Thus, the optimization conditions of the load with a parallel
capacitor are different from those for the network presented
in Fig. 8.
In the following, for a transformer network with a capacitor
in parallel with a resistive load, we present design formulae
that can be used for achieving optimum power efficiency by
choosing the optimal parallel capacitor.

B. CALCULATION OF OPTIMAL LOAD AND SOURCE

IMPEDANCES FOR A GIVEN TRANSFORMER
For the load impedance, given by ZL = RL + jXL ,
in Fig. 7(b) (excluding COUT ), the power efficiency (calcu-
lated in Appendix C) is given by
RL
η= . (10)
(RL + R2 ) + R1 (RL +R2 ) 2 +XL )
2 +(ωL 2
FIGURE 9. Plot of the power efficiency versus the value of the parallel
(ωM )2 tuning capacitor for the on-chip 2:1 transformer at 10 GHz.

If RL is independent of XL (= −1/ωCL ), η is maximized

at XL = −ωL2 , which is identical to the condition presented
in [20]. where
Let us consider the transformer with a load and a tuning R2 R1 R22 + (ωL2 )2 R1 L2
capacitor (CL = Ct ) in parallel with the load, as shown a= + ; b= ;
RL RL (ωM )2 ωM 2
in Fig. 8. The equivalent series reactance and resistance of (2R2 + RL ) R1
the load are given by c = a+ . (14)
(ωM )2
RL ωR2L Ct
Req = ; Xeq = − . (11) Because a > 0, the ratio Ploss /PL is minimized at x = b/a,
1 + (ωRL Ct )2 1 + (ωRL Ct )2 where we can obtain the optimal parallel tuning capacitor
By replacing the calculated values from (11) with the (Ctopt ) as
corresponding quantities in (10), the power efficiency can be
expressed as a function of Req and Xeq . However, Req is a L2 1 Q1 =Q2
Ctopt = ≈ Q1 Q2 (1+k 2 )1
2 2 ω2 L2


function of Xeq when Ct is tuned for fixed RL . Therefore, R2 (1 + Q2 +k 2 Q1 Q2 ) k 2 +1
we should not substitute Xeq with the aforementioned opti- (15)
mum value (−ωL2 ) of XL for maximizing η.
In order to calculate the value of Ct that maximizes η, At this point, the maximum power efficiency is given by
we express η as follows: 1
η(Ctopt ) = (16)
Ploss 1+c− b2
a .
PL
As evident from (15), the optimum value of the parallel
1 1 + (ωRL Ct )2
= −1= capacitor Ctopt is less sensitive to the quality factors of the
η RL transformer if Q1 is close to Q2 and the product Q1 Q2 (k 2 +1)
 2 
is sufficiently higher than unity. By applying Ctopt in (15)

RL
R +
 
 2 
to (11), we obtain Xeqopt , which differs from −ωL2 used in
(ωR )2
 
 R1  1 + C
L t

× R2 + ,
 
!2 Fig. 7(b). Fig. 9 shows plots of the calculated and simulated
(ωM )2  ωR2L Ct
 
+ ωL2 −
 

 
 efficiency versus Ct for two values of RL at 10 GHz. Except
1 + (ωRL Ct )2
 
for the tuning capacitor (CL = Ctopt ) in parallel with the
(12) given load resistance RL , the simulation setup was the same as
before. Owing to the parasitic capacitor of the two windings,
where Ploss is the power dissipated in the transformer. The a peak discrepancy of around 2.5% was observed between
ratio Ploss /PL should be minimized to maximize η. the efficiency calculated from (15) and the simulated value.
By denoting x = ωRL Ct , this ratio can be expressed as Interestingly, Ctopt is not a function of RL .
Ploss We can also obtain the same design formulae by using a
= ax 2 − 2bx + c, (13) different approach. When both source and load values are
PL
VOLUME 7, 2019 140985
 V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

FIGURE 10. Plot of simulated power gains versus the value of the parallel
tuning capacitor (Zs is calculated from (7) and RL is given by (18)). FIGURE 11. Photographs of the on-chip transformers fabricated in
0.18 µm CMOS technology: (a) the standalone 2:1 transformer, and
(b) transformers with parallel tuning capacitors.

set, it is desirable to find the impedance matching condi-

tion on both sides of the transformer so that the maximum
transducer power gain, which is also the maximum power
efficiency (ηmax = GTmax ) is obtained. The source and
load impedances are given by (7) to achieve this condition.
Therefore, the equivalent series resistance Req and equivalent
series reactance Xeq should satisfy the conditions:

RL p
Req =


2
= R2 1 + k 2 Q1 Q2
1 + (ωRL Ct )
(17)
ωR2L Ct FIGURE 12. Plots of optimum power efficiency versus frequency for the
Xeq = − = −ωL .

 2 on-chip 2:1 transformer.
1 + (ωRL Ct )2
By solving these two equations, RLopt and Ctopt are calcu- din = 9µm. Fig. 11(a) presents a photograph of a fabri-
lated as cated transformer with two signal pads. The S-parameters
 !
 p Q22 of the transformer were measured and de-embedded from
2
RLopt = R2 1 + k Q1 Q2 1 +

 the pads and transmission lines that did not belong to
1 + k 2 Q1 Q2
(18) the transformer. Parameters of the low-frequency model
L2
.

 C
 topt
 = transformer—k, L1 , L2 , Q1, and Q2 —were then extracted and
R22 (1 + Q22 + k 2 Q1 Q2 ) compared with those extracted from the simulation, which
Naturally, the optimum value of Ctopt in (18) is the same are shown in Fig. 4. The maximum power gain (GTmax ) was
as that in (15). Fig. 10 shows plots of the simulated power also extracted from the measured S-parameters; it is depicted
gains—GTmax , GT , and the operating power gain (or power in Fig. 2 along with plots of calculated and simulated values.
efficiency η)—around Ct = Ctopt . In this verification, ZS and Fig. 4 shows that the measured inductances and coupling
RL are chosen so that 0S = 0L = 0 at the optimum value coefficients fit the simulation results quite well. By contrast,
Ctopt , which is calculated using (18); ZS is computed using simulation results in the high-frequency region (f > 7 GHz)
(7) and (18) gives RL = RLopt . showed that the measured quality factors of the two coils were
Similarly, if the source has the same structure, which means degraded. The optimistic loss consideration in the model,
that the source consists of a resistor (RSp ) in parallel with which does not hold true in a real environment, may cause this
a tuning capacitor (CSp ), the optimum values of parameters discrepancy. Specifically, the loss tangents of the dielectric
RSp_opt and CSp_opt to maximize the transducer power gain materials depend on the frequency in the real case while
are given by they were assumed to be constants in HFSS which were
 ! extracted at low frequency region. In addition, the rough
 p
2
Q21 metal surfaces in the fabricated transformer could cause more
RSp_opt = R1 1 + k Q1 Q2 1 +

1 + k 2 Q1 Q2

(19) loss at high-frequency region [37].
L1 Because the maximum power gain is suppressed quickly
.

CSp_opt = 2

when the product k 2 Q1 Q2 becomes large (as shown in Fig. 2),

R1 (1 + Q21 + k 2 Q1 Q2 )
the GTmax extracted from measurements was still comparable
V. EXPERIMENTAL VERIFICATION to the simulated value (as shown in Fig. 12), regardless of
To demonstrate the validity of the proposed approach, the degradation of the measured quality factors Q1 and Q2
we implemented various transformers with and without par- compared to those simulated (depicted in Fig. 4). In addition,
allel capacitors in a 0.18 µm CMOS process. A reference in the high-frequency region (compared to the SRF), the par-
transformer was also included to confirm the validity of the asitic capacitor also plays a role in the increase in GTmax ,
transformer model in HFSS, which is shown in Fig. 3 for as shown in Fig. 6.

140986 VOLUME 7, 2019

V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

To verify the analysis proposed in section IV, we measured If simultaneous conjugate matching conditions are met
several fabricated transformers with parallel capacitors at the (ZS = Zin∗ and ZL = Zout
∗ ), then we have

load; they are presented in Fig. 11(b). The S-parameters were

(ωM )2 (R1 + RS )

de-embedded from the RF pads and extra transmission lines. RL = R2 +

(R1 + RS )2 + (ωL1 + XS )2


Subsequently, power efficiencies for specific loads of 50 and 
(ωM )2 (ωL1 + XS )


100  were extracted (Fig. 9). The measured data indicated

XL = −ωL2 +


that the optimum Ct was the same for the two loads and (R1 + RS )2 + (ωL1 + XS )2 (A3)
that it deviated from the simulation value by around 20 fF 
R = R + (ωM )2 (R2 + RL )
 S 1
(R2 + RL )2 + (ωL2 + XL )2

(nearly 7%). The shift in the measured Ctopt is because of the 

(ωM )2 (ωL2 + XL )


difference between the fabricated transformer and its model; 
XS = −ωL1 +
 .
this is shown by the extracted parameters in Fig. 4. Another (R2 + RL )2 + (ωL2 + XL )2
reason could be that the nominal capacitance specified by the If we express parameters as rL = RL +R2 , xL = XL +ωL2 ,
manufacturer was different from the actual value because of rs = Rs + R1 , and xs = Xs + ωL1 , then (A3) becomes
process variations. From the extracted parameters of the fab-
ricated transformer, the optimum parallel load resistor RLopt
(ωM )2 rS

was calculated to be 103  by using (18). Therefore, the case 
 r L = 2R 2 + , (A4.1)
rS2 + xS2


of RL = 100 still provided a higher power efficiency, which 


(ωM )2 xS

was confirmed by measurements when Ct was around its 
,

x = (A4.2)

 L

optimum value. 
rS2 + xS2
(ωM )2 rL
,

VI. CONCLUSION r = 2R + (A4.3)

S 1

rL2 + xL2


We present a systematic analysis of and design formulae for



 xS = (ωM ) xL .
2

an impedance matching network with a two-winding trans-


(A4.4)

former. To develop design guidelines for the optimization of 2
rL + xL2

a transformer network, we investigated the role of the resistive
Case 1: Consider xs 6 = 0 (⇔ xL 6 = 0).
and the reactive parts of the source and load in achieving the
By replacing xs in (A4.4) with (A4.2), we obtain
maximum power transfer. We present the design formulae,
obtained from the aforementioned analysis, for the optimum (ωM )4
rS2 + xS2 = . (A5)
source and load impedance of a given transformer for the case rL2 + xL2
where a parallel tuning capacitor is used to achieve optimal
By using this expression for (A4.1) after replacing rs in
power transfer in the transformer network. The validity of the
(A4.1) with that in (A4.3), we obtain
proposed formulae was verified for a 2:1 on-chip transformer  
simulated and measured in a 0.18 µm CMOS process with (ωM )2 rL
(ωM ) 2R1 + r 2 +x 2
2
a 3D EM simulator (HFSS). The results of the present study L L
rL = 2R2 +
are widely applicable to various sectors in the fields of RFICs, 4(ωM )
WPT, and any transformer network operating below the trans- rL2 +xL2
2R1 rL2 + xL2


former SRF.
= 2R2 + + rL
(ωM ) 2
APPENDIX A R2
From (4) and (5), we can obtain the input impedance as ⇒ rL2 = − (ωM )2 .
+ xL2 (A6)
R1
Apparently, there is no solution for rL and xL that satisfies
( )
V1 (ωM )2 (R2 + RL )
Zin = = R1 + (A6), and therefore, the equations in (A3) have no solution.
I1 (R2 + RL )2 + (ωL2 + XL )2
( ) Case 2: Consider xs = xL = 0. Therefore, (A4) becomes
(ωM )2 (ωL2 + XL ) (ωM )2

+ j ωL1 − . (A1) 
rL = 2R2 +
(R2 + RL )2 + (ωL2 + XL )2

rS
(ωM )2
.

r = 2R +
Because the transformer is symmetric, Zout can be derived  S 1
rL
from Zin by replacing R1 , L1 , and ZL with R2 , L2 , and ZS , (
respectively, as follows: rL rS = 2R2 rS + (ωM )2 R1
⇒ ⇒ rS = rL (A7)
( ) rS rL = 2R1 rL + (ωM ) 2 R2
(ωM )2 (R1 + RS ) By substituting this expression in the first expression in (A7),
Zout = R2 +
(R1 + RS )2 + (ωL1 + XS )2 we obtain
( )
(ωM )2 (ωL1 + XS ) (ωM )2 R2
+ j ωL2 − . (A2) rL = 2R2 + ⇒ rL2 − 2R2 rL − (ωM )2 = 0.
(R1 + RS )2 + (ωL1 + XS )2 rL RR12 R1

VOLUME 7, 2019 140987

 V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

We know that rL = RL + R2 . By substituting this expression REFERENCES

in the above equation, we can write [1] T. Xi, S. Huang, S. Guo, P. Gui, D. Huang, and S. Chakraborty, ‘‘High-
efficiency E-band power amplifiers and transmitter using gate capacitance
s linearization in a 65-nm CMOS process,’’ IEEE Trans. Circuits Syst., II,
R2 (ωM )2 Exp. Briefs, vol. 64, no. 3, pp. 234–238, Mar. 2017.
R2L − R22 − (ωM )2 = 0 ⇒ RL = R2 1 + [2] V.-S. Trinh, H. Nam, and J.-D. Park, ‘‘A 20.5-dBm X -band power amplifier
R1 R1 R2
p with a 1.2-V supply in 65-nm CMOS technology,’’ IEEE Microw. Wireless
= R2 1 + k 2 Q1 Q2 . Compon. Lett., vol. 29, no. 3, pp. 234–236, Mar. 2019.
[3] R. Wu, N. Liao, X. Fang, and J. K. O. Sin, ‘‘A silicon-embedded
transformer for high-efficiency, high-isolation, and low-frequency on-
Similarly, chip power transfer,’’ IEEE Trans. Electron Devices, vol. 62, no. 1,
p pp. 220–223, Jan. 2015.
RS = R1 1 + k 2 Q1 Q2 . [4] V.-S. Trinh and J.-D. Park, ‘‘An X -band single-pull class A/B power
amplifier in 0.18µm CMOS,’’ Microw. Opt. Technol. Lett., vol. 61, no. 7,
pp. 1736–1740, Jul. 2019.
From (A4), one can quickly point out that if R1 = R2 = [5] A. Medra, D. Guermandi, K. Vaesen, S. Brebels, A. Bourdoux,
0 (an ideal transformer), then solutions for xL and rL in W. Van Thillo, P. Wambacq, and V. Giannini, ‘‘An 80 GHz low-noise
(A4.1) and (A4.2) automatically satisfy (A4.3) and (A4.4), amplifier resilient to the TX spillover in phase-modulated continuous-
wave radars,’’ IEEE J. Solid-State Circuits, vol. 51, no. 5, pp. 1141–1153,
respectively. This shows that when a transformer is ideal, May 2016.
a conjugate match on the source side leads to one on the load [6] S.-L. Jang, ‘‘Complementary current reuse quadrature voltage-controlled
side. Intuitively, we can see that if the transformer is lossless, oscillators,’’ IET Microw. Antennas Propag., vol. 10, no. 7, pp. 756–763,
May 2016.
the transfer of the maximum power from the source to the [7] S.-W. Kang, H.-J. Kim, and B.-H. Cho, ‘‘Adaptive voltage-controlled
network (including the transformer and load) implies that the oscillator for improved dynamic performance in LLC resonant converter,’’
maximum power is delivered to the load. IEEE Trans. Ind Appl., vol. 52, no. 2, pp. 1652–1659, Mar./Apr. 2016.
[8] N. Mazor, B. Sheinman, O. Katz, R. Levinger, E. Bloch, R. Carmon,
R. Ben-Yishay, and D. Elad, ‘‘Highly linear 60-GHz SiGe downconver-
APPENDIX B sion/upconversion mixers,’’ IEEE Microw. Wireless Compon. Lett., vol. 27,
From (4) and (5), we can write no. 4, pp. 401–403, Apr. 2017.
[9] S.-L. Jang, T.-C. Kung, and C.-W. Hsue, ‘‘Wide-locking range divide-by-
(R2 + RL ) + j (ωL2 + XL ) 4 injection-locked frequency divider using linear mixer approach,’’ IEEE
I1 = I2 . (B1) Microw. Wireless Compon. Lett., vol. 27, no. 4, pp. 398–401, Apr. 2017.
jωM [10] C. Wang, H. Liao, Y. Xiong, C. Li, R. Huang, and Y. Wang, ‘‘A Physics-
Based Equivalent-Circuit Model for On-Chip Symmetric Transformers
Under the condition 0S = 0, we can use (1) to obtain With Accurate Substrate Modeling,’’ IEEE Trans. Microw. Theory Techn.,
vol. 57, no. 4, pp. 980–990, Apr. 2009.
PL RL |I2 |2 [11] A. Ghadiri and K. Moez, ‘‘Bandwidth Enhancement of On-Chip Trans-
GT = η = =
formers Using Negative Capacitance,’’ IEEE Trans. Circuits Syst., II, Exp.
Pavs 2Pnet Zin = Zs∗ Briefs, vol. 59, no. 10, pp. 648–652, Oct. 2012.
[12] H.-M. Hsu, S.-H. Lai, and C.-J. Hsu, ‘‘Compact layout of on-chip trans-
RL |I2 |2 RL |I2 |2
=
= . former,’’ IEEE Trans. Electron Devices, vol. 57, no. 5, pp. 1076–1085,
Rin |I1 |2 Zin = Zs∗ RS |I1 |2 May 2010.
[13] L. F. Tiemeijer, R. M. T. Pijper, C. Andrei, and E. Grenados, ‘‘Anal-
Using (7) along with the above formula and (B1), we can ysis, design, modeling, and characterization of low-loss scalable on-
chip transformers,’’ IEEE Trans. Microw. Theory Techn., vol. 61, no. 7,
obtain pp. 2545–2557, Jul. 2013.
p [14] Z. Gao, K. Kang, C. Zhao, Y. Wu, Y. Ban, L. Sun, W. Hong, and Q. Xue,
k 2 Q1 Q2 + 1 − 1 ‘‘A broadband and equivalent-circuit model for millimeter-wave on-chip
GT max = 1 − 2 . (B2) M:N six-port transformers and baluns,’’ IEEE Trans. Microw. Theory
k 2 Q1 Q2 Techn., vol. 63, no. 10, pp. 3109–3121, Oct. 2015.
[15] H.-M. Hsu, C.-W. Tseng, and K.-Y. Chan, ‘‘Characterization of on-chip
APPENDIX C transformer using microwave technique,’’ IEEE Trans. Electron Devices,
Because the mutual inductance is modeled as a pure imag- vol. 55, no. 3, pp. 833–837, Mar. 2008.
[16] D. M. Pozar, Microwave Engineering, 4th ed. Hoboken, NJ, USA: Wiley,
inary value, the power loss in the transformer is the total 2011.
resistive loss in the two windings expressed as [17] S. Roberts, ‘‘Conjugate-Image Impedances,’’ Proc. IRE, vol. 34, no. 4,
pp. 198–204, Apr. 1946.
Ploss = (1/2)R1 |I1 |2 + (1/2)R2 |I2 |2 . [18] J. Rahola, ‘‘Power Waves and Conjugate Matching,’’ IEEE Trans. Circuits
Syst., II, Exp. Briefs, vol. 55, no. 1, pp. 92–96, Jan. 2008.
[19] R. Sinha and A. De, ‘‘Theory on matching network in viewpoint of
The power consumed by the load is given by transmission phase shift,’’ IEEE Trans. Microw. Theory Techn., vol. 64,
no. 6, pp. 1704–1716, Jun. 2016.
PL = (1/2)RL |I2 |2 . [20] I. Aoki, S. D. Kee, D. B. Rutledge, and A. Hajimiri, ‘‘Distributed active
transformer-a new power-combining and impedance-transformation tech-
nique,’’ IEEE Trans. Microw. Theory Techn., vol. 50, no. 1, pp. 316–331,
Using (B1), we can calculate the power efficiency as Jan. 2002.
[21] G. Liu, P. Haldi, T.-J. K. Liu, and A. M. Niknejad, ‘‘Fully integrated CMOS
PL PL RL |I2 |2 power amplifier with efficiency enhancement at power back-off,’’ IEEE
η= = = J. Solid-State Circuits, vol. 43, no. 3, pp. 600–609, Mar. 2008.
Pnet PL + Ploss R1 |I1 |2 + R2 |I2 |2 + RL |I2 |2 [22] P. Haldi, D. Debopriyo, P. Min, D. Reynaert, G. Liu, and A. M. Niknejad,
RL ‘‘A 5.8 GHz 1 V linear power amplifier using a novel on-chip transformer
= (C1)
(RL + R2 ) + R1 (RL +R2 ) +(ωL 2 +XL )
2 2 power combiner in standard 90 nm CMOS,’’ IEEE J. Solid State Circuits,
(ωM )
2 vol. 43, no. 5, pp. 1054–1063, May 2008.
.

140988 VOLUME 7, 2019

V.-S. Trinh, J.-D. Park: Theory and Design of Impedance Matching Network

[23] S. Goswami, H. Kim, and J. L. Dawson, ‘‘A frequency-agile RF frontend VAN-SON TRINH (S’18) received the B.Sc.
architecture for multi-band TDD applications,’’ IEEE J. Solid-State Cir- degree from the Hanoi University of Science and
cuits, vol. 49, no. 10, pp. 2127–2140, Oct. 2008. Technology (HUST), Hanoi, Vietnam, in 2015.
[24] N. Ryu, S. Jang, K. C. Lee, and Y. Jeong, ‘‘CMOS doherty amplifier with He is currently pursuing the Ph.D. degree in
variable balun transformer and adaptive bias control for wireless LAN electronics and electrical engineering with Dong-
application,’’ IEEE J. Solid-State Circuits, vol. 49, no. 6, pp. 1356–1365, guk University, Seoul, South Korea.
Jun. 2014. His current research interests include various
[25] R. Wu, W. Li, H. Luo, J. K. O. Sin, and C. P. Yue, ‘‘Design and character-
analog and RF integrated circuits.
ization of wireless power links for brain–machine interface applications,’’
IEEE Trans. Power Electron., vol. 29, no. 10, pp. 5462–5471, Oct. 2014.
[26] N. Inagaki, ‘‘Theory of image impedance matching for inductively coupled
power transfer systems,’’ IEEE Trans. Microw. Theory Techn., vol. 62,
no. 4, pp. 901–908, Apr. 2014.
[27] J. R. Long, ‘‘Monolithic transformers for silicon RF IC design,’’
IEEE J. Solid-State Circuits, vol. 35, no. 9, pp. 1368–1382, Sep. 2000.
[28] B. Razavi, RF Microelectronics, 2nd ed. Upper Saddle River, NJ, USA:
Prentice-Hall, 2011.
[29] J. Chen and A. M. Niknejad, ‘‘A compact 1V 18.6dBm 60GHz power
amplifier in 65nm CMOS,’’ in IEEE ISSCC Dig. Tech. Papers, Feb. 2011,
pp. 432–433.
[30] T. LaRocca, J. Y.-C. Liu, and M.-C. F. Chang, ‘‘60 GHz CMOS amplifiers JUNG-DONG PARK (M’15–SM’18) received the
using transformer-coupling and artificial dielectric differential transmis- B.Sc. degree from Dongguk University, Seoul,
sion lines for compact design,’’ IEEE J. Solid-State Circuits, vol. 44, no. 5,
South Korea, in 1998, the M.S. degree from
pp. 1425–1435, May 2009.
the Gwangju Institute of Science and Technol-
[31] J.-D. Park, S. Kang, and A. M. Niknejad, ‘‘A 0.38 THz fully integrated
transceiver utilizing a quadrature push-push harmonic circuitry in SiGe ogy (GIST), Gwangju, South Korea, in 2000, and
BiCMOS,’’ IEEE J. Solid-State Circuits, vol. 47, no. 10, pp. 2344–2354, the Ph.D. degree in EECS from the University
Oct. 2012. of California at Berkeley, Berkeley, CA, USA,
[32] C. Marcu, D. Chowdhury, C. Thakkar, J.-D. Park, L.-K. Kong, M. Tabesh, in 2012.
Y. Wang, B. Afshar, A. Gupta, A. Arbabian, S. Gambini, R. Zamani, From 2000 to 2002, he was with the Institute
E. Alon, and A. M. Niknejad, ‘‘A 90 nm CMOS low-power 60 GHz for Advanced Engineering (IAE), Yongin, South
transceiver with integrated baseband circuitry,’’ IEEE J. Solid-State Cir- Korea, where he was involved with the design of 35 GHz radar/radiometer
cuits, vol. 44, no. 12, pp. 3434–3447, Dec. 2009. transceivers. From 2002 to 2007, he was a Senior Researcher with the
[33] J.-D. Park, S. Kang, S. V. Thyagarajan, E. Alon, and A. M. Niknejad, Agency for Defense Development (ADD), Daejeon, South Korea, where
‘‘A 260 GHz fully integrated CMOS transceiver for wireless chip-to- he was responsible for the development of millimeter-wave (mmW) pas-
chip communication,’’ in Proc. Symp. VLSI Circuits, Honolulu, HI, USA, sive/active sensors and related mmW modules. From 2007 to 2012, he was
Jun. 2012, pp. 48–49. with the Berkeley Wireless Research Center (BWRC), where he involved
[34] R. Wu and J. K. O. Sin, ‘‘High-efficiency silicon-embedded coreless in silicon-based RF/millimeter-wave/terahertz circuits and systems. From
coupled inductors for power supply on chip applications,’’ IEEE Trans.
2012 to 2015, he was with Qualcomm Inc., San Jose, CA, USA, where
Power Electron., vol. 27, no. 11, pp. 4781–4787, Nov. 2012.
he designed various RF/analog integrated circuits. He is currently an Asso-
[35] L. H. Dixon, ‘‘Eddy current losses in transformer winding and circuit
wiring,’’ Texas Instruments, Dallas, TX, USA, Tech. Rep. TI 2001 Mag- ciate Professor with the Division of Electronics and Electrical Engineering,
netic Design Handbook-MAG100A, 2001. Dongguk University, Seoul. His current research interests include wireless
[36] W. B. Kuhn and N. M. Ibrahim, ‘‘Analysis of current crowding effects in communications, remote sensors, microwave electronics, analog, RF, mixed-
multiturn spiral inductors,’’ IEEE Trans. Microw. Theory Techn., vol. 49, signal, and millimeter wave circuits.
no. 1, pp. 31–38, Jan. 2001. Prof. Park was a recipient of the 2017 Most Frequently Cited Papers Award
[37] B. H. Lee, Y. T. Keum, and R. H. Wagoner, ‘‘Modeling of the friction as a Lead Author, from 2010 to 2016, at the 2017 IEEE Symposium on VLSI
caused by lubrication and surface roughness in sheet metal forming,’’ Circuits, Kyoto.
J. Mater. Process. Technol., vols. 130–131, pp. 60–63, Dec. 2002.

VOLUME 7, 2019 140989

View publication stats