x-5

A Q-FACTOR ENHANCEMENT TECHNIQUE

FOR MMIC INDUCTORS
Mina Danesh, John R. Long, R. A. Hadaway' and D. L. Harame2
University of Toronto Nortel Technology, Ottawa, Ontario. Canada
Toronto, Ontario, Canada IBM Microelectronics, Burlingtoin, VT, USA

-
Abstract An increase of 50% in the peak Q- grounded. Differential circuiit topologies are common in
factor and a wider operating bandwidth for integrated circuits, and consequently an alternative
monolithic inductors is achieved by exciting a method that is practical for integrated circuits is to
microstrip structure differentially. Conventional excite the spiral inductor differentially, using a source
excitation of a 8 nH spiral inductor fabricated in a connected between the two ends of the microstrip spiral.
production silicon IC technology resulted in a peak The peak Q-factor shows ai significant increase under
(measured) Q-factor of 6.6 at 1.6 GHz, while the differential excitation, and this high Q value is
differential connection showed a maximum Q-factor maintained over a broader bandwidth.
of 9.7 at 2.5 GHz. These experimental results
compared favorably with the behaviour predicted 11. THEORY
from simulation. A transmission line can be approximated over a
range of frequencies by a lumped element equivalent
I. I,NTRODUCTION circuit model. For a microstrip line fabricated in silicon
Microstrip inductors have been used extensively in technology, an appropriate equivalent circuit is shown in
radio frequency (RF) and monolithic microwave Fig. 1, where L is the total inductance of the line and r is
integrated circuits (MMIC). The quality factor (Q) of the series resistance due to conductor losses and
microstrip structures is limited by the series resistance dissipation arising from cuirrent flowing in the silicon
of the metallization and, in the case of silicon substrate. The shunt parasitics result from a combination
technology, losses in the conductive substrate (typically of capacitances, due to an insulating layer of silicon
1 to 10 0-cm). The Q-factor is typically less than 10 for dioxide (Cox) and the underlying substrate (C,,), and
a metal-insulator-semiconductor (MIS) structure substrate dissipation (R,i) [7]. For the spiral inductor,
fabricated in a production silicon IC technology [ 1,2]. It additional components are required to represent mutual
has been demonstrated that the quality factor can be magnetic and electric couplling between adjacent lines
improved by optimizing both the physical layout and [ll.
structure of the inductor. Thicker metallization and .--. r
stacking of metal layers reduces the conductor
dissipation, thereby improving the Q [3]. Design
Port1
0
1 iZBi- _I2 Port2
I
guidelines have been proposed to optimize the
geometric parameters of microstrip spiral inductors,
such as strip width, spacing between the strips, and the
gap between groups of coupled lines on opposing sides
[l]. Losses in the semiconducting substrate may be
reduced through the use of higher resistivity material .-...
[4], however, this is incompatible with current silicon - - - - - :.-...
---$ .- -
Z,,!
device technology (e.g., CMOS). An alternative is to
Fig. 1. Microstrip line equivalent circuit.
remove the underlying silicon by selective etching of the
substrate or by applying a thin membrane beneath the
inductor [S, 61, but this requires additional processing For single-ended excitation, Port 2 in Fig. 1 is
steps and there is a loss of mechanical strength when the grounded and the inductor is connected as a one-port.
underlying silicon is removed. The input impedance at 'Port 1 (in this case Zsd
becomes a parallel combinaition of two components: one
Microstrip spiral inductors are normally driven due to the inductance and series dissipation (L and r in
"single-ended", that is. the driving source is connected Fig. I), the other due to the shunt R-C parasitic
to one terminal of the spiral while the other end is elements, as illustrated in Fig. 2. For a differential

217
0-7803-4439-1/98/%10.000 1998 IEEE 1998 IEEE Radio Frequency Integrated Circuits Symposium
excitation, where the signal is applied between the two 8 pm wide and the spacing between conductors is
ports (Le., between Port1 and Port2), the input 2.8 pm (w and s in Fig. 4,respectively). The relatively
impedance (Z), is due to the parallel combination of narrow conductor width and spacing results in higher
2zA and Zg, where impedances ZA and ZB are defined magnetic coupling between microstrip lines and lower
in Fig. 1. Since the substrate parasitics are connected capacitive parasitics to the substrate. The inner gap
together via the ground plane, the two shunt elements, between groups of coupled lines, G , is approximately
150 pm, which minimizes negative mutual coupling on

-
are now in series (Le., ZA+ZA), resulting in the
equivalent circuit shown in Fig. 3. opposite sides of the spiral.
Port 1 L r Passivation S W
\
L

h3
h2

hl

L r

2R Fig. 4.Cross-sectional view of the inductor.

The quality factor of the inductor below the first

resonant frequency is defined by

where Zinputis the series equivalent input impedance. At

lower frequencies, the input impedance in either the
shunt or differential connections is approximately the
same, but as the frequency increases, the substrate
parasitics, C and R, come into play. For the case of
differential excitation, these parasitics have a higher
impedance at a given frequency than in the single-ended
connection, as seen from comparison of Figs. 2 and 3.
This reduces the real part and increases the reactive
Fig. 5. Spiral inductor test structure layout.
component of the input impedance. Therefore, the
inductor Q-factor (from eq. 1) is improved when driven
differentially, and moreover, a wider operating
bandwidth can be achieved. Parameter Value
I
III. ANALYSIS Oxide thickness over M2
-1

I h3 - h6 = 1.3 pm
A 5-turn square spiral inductor was fabricated and Oxide thickness below M2 I h2 = 3.61 um
tested in order to verify these predictions. A cross- Silicon relative permittivity I E ~ =11.7
sectional view of a portion of the structure is illustrated Silicon resistivitv 1 o=lSSI-cm
in Fig. 4,and the substrate and metal properties for the I Silicon thickness IhI=200um ~ I
fabrication process are listed in Table 1. The outer Topmetal M3 resistivity p = 0.031 R-pm
dimension, A, as shown in Fig. 5, is 250 pm. M3 thickness h5 = 2.07 pm
The inductor consists mainly of topmetal (M3) which is M2 thickness h 6 = 0.84 um

218
 N.RE~UCTS
AND DISCUSSION At frequencies beyond the peak, an increase of greater
than 50% can be achieved. It should be noted that Q
Three-dimensional electromagnetic simulation
values for the differential case, because they are greater
(using HP-EEsof's Momentum) and experimental
in magnitude, are much more sensitive to slight
measurements were obtained for the 2-port spiral
variations in the measured or simulated input
inductor. The results for the single-ended configuration
impedance. Thus, the relative: effect of an error in either
were derived by grounding one of the ports (either Port1
the measurement or simulation near the peak Q for the
or Port2) since the structure is symmetric. The response
differential case is more pronounced. A slight increase
due to a differential excitation can be derived from 2-
in the topmetal thickness could account for part of the
port S-parameter measurements using the relationship
discrepancy between the measured and simulated
SD = S,,-S21 (2) performance. Tiis would be consistent with the lower
series resistance and inductance observed from the
The input impedances are determined from the l-port S- measurements when compared to the simulated values.
parameters or the 2-port impedance matrix, as: The measured data shown is from a representative
sample; several measurements were performed which
gave the same results within a k5% variation.
and

z, = z,, +z22-z12-z21
= 2*z0(S) (4)
IOOO' .... s
where 2, is the system impedance (50 Q).
Comparisons between experimental measurements
and simulations for the input impedance and Q-factor n 800'
(as defined by eq. 1) are shown in Figs. 6, 7 and 8. The s
series resistance at low frequencies (i.e., f c 0.5 GHz),
5
2
-

600'
as defined by r from the lumped element model of Figs.
N-
2 and 3, was measured to be 7.3 R ,while the simulation Y

predicts 7.9 Q . Measured and simulated low frequency

inductances are 7.7 nH and 8.3 nH, respectively. At c2 400'
lower frequencies, the difference in Q between
differential and single-ended excitations is not 200
*

significant ( ~ 1 % ) This
. is because the shunt capacitive
parasitic components do not affect the low frequency
input impedance and hence, the two cases can be - n
represented by a series L-r model. However, as the "0 11 2 3 4 5 6
frequency increases, the difference between the input Frequency (GHz)
impedances becomes substantial; ZD is much lower than Fig. 6. Resistive ZSE and ZD.
ZSEby an increasingly greater factor. This is caused by
the lower substrate parasitics present in the differentially
excited case, as previously described. The difference
between Q-factors in the differential and single-ended
cases, as shown in Fig. 8, illustrates this point. The peak
in the Q-factor is a result of the shunt parasitics
resonating with the inductance. Lower parasitics for
differential excitation result in a higher peak Q-factor
and broadening of the Q peak, when compared to the
conventional single-ended connection.
For the single-ended excitation, the peak Q-factor
is 6.6 at 1.6 GHz from both measurement and
simulation. However, for a differential excitation, the
resulting peak Q occurs at a higher frequency
(2.5 GHz), with a value of 9.7 from measurement and
10.3 from simulation. This is a 47% increase in the
measured peak Q between the differential and single-
ended cases that can be realized without modification to J
0 1 2 3 4 5 6
the fabrication process. Achieving a comparable Q value Frequency (GHz)
in the single-ended connection would require an
increase of approximately twice the top-metal thickness. ~ ZD.
Fig. 7.Inductive Z Sand

219
 V. CONCLUSION
12
I Higher peak Q-factor values and a wider operating
bandwidth for monolithic inductors can be achieved by
exciting a microstrip structure differentially. This could
be exploited in many MMIC circuit applications, such
as oscillators, mixers and amplifiers, where high-quality
components are required. Measured and simulated
results presented in this paper have validated the
prediction of lower substrate parasitics that were made
from a simplified lumped element model. The
improvement in performance has been demonstrated for
an inductor fabricated in a production silicon IC
technology, however, these same results would apply to
microstrip inductors fabricated on other substrates, such
as GaAs.

ACKNOWLEDGEMENTS
U
0 I 2 3 4 5 6 This work was supported by Micronet and the
Frequency (GHz) Natural Sciences and Engineering Research Council of
Canada (NSERC).
Fig. 8. Measured and simulated Q-factor.
REFERENCES
Table 2 lists the parameters of a lumped-element
model (see Fig. 9) that was fit over a broadband of J. R. Long and M. A. Copeland, "The Modeling,
frequency (500-6000MHz) for the inductor from the Characterization, and Design of Monolithic Inductors for
simulated data for both the single-ended and differential Silicon RF ICs", IEEE J. Solid-state Circuits, Vol. 32, No.
3, March 1997, pp. 357-369.
connections. The resistive parasitic element is 2.8 times
higher for differential excitation, and the shunt [2] K. B. Ashby, I. A. Koullias, W. C. Finley, J. J. Bastek and
capacitance is a small fraction of the single-ended case S. Moinian, "High Q Inductors for Wireless Applications
(CI: 38%, C2: 70%). The modified values for the in a Complementary Silicon Bipolar Process", IEEE J.
lumped element models differ from the equivalent Solid-state Circuits, Vol. 3 1, No. 1. January 1996, pp. 4-9.
circuits shown in Figs. 2 and 3 due to the distributed [3] J. N. Burghartz, M. Soyuer, K. A. Jenkins, M. Kies, M.
nature of the actual inductor, which cannot be modeled Dolan, K. J. Stein, J. Malinowski and D. L. Harame,
accurately by a single lumped-element section for both "Integrated RF Components in a SiGe Bipolar
one and two-port configurations. Technology", IEEE J. Solid-state Circuits, Vol. 32, No. 9,
September 1997, pp. 140-1445.
[4] M. Park,S. Lee, H. K. Yu, J. G. Koo and K. S. Nam, "High
Q CMOS-Compatible Microwave Inductors using
Double-Metal Interconnection Silicon Technology", IEEE
Microwave and Guided Wave Letters, Vol. 7, No. 2,
February 1997, pp. 45-47.
[5] J. Y.-C. Chang, A. A. Abidi and M. Gaitan, "Large
Suspended Inductors on Silicon and Their Use in a 2-pm
CMOS RF Amplifier", IEEE Electron Device Letters, Vol.
14, No. 5, May 1993, pp. 246-248.
[6] A. C. Reyes, S . M. El-Ghazali, S . J. Dom, M. Dydyk, D. K.
Schroder and H. Patterson, "Coplanar Waveguides and
Microwave Inductors on Silicon Substrates", IEEE Trans.
Microwave Theory Tech., Vol. 43, No. 9, September 1995,
Fig. 9. Equivalent circuit model for both configurations. pp. 2016-2022.
Table 2. Lumped element parameter fitting. [71 H. Hasegawa, M. krukawa and H. Yanai, "Properties of
Microstrip Line on Si-Si02 System", IEEE Trans.
Excitation I C1 (iF) I C2 (€IIR1
? (a)
) Microwave Theory Tech.. Vol. 19, No. 11, November
Sinele-ended
~
_-- --
~~~ I 135 I 162 I 358 1971, pp. 869-881.
Differential I 95 I 62 I lo00

220