Beruflich Dokumente
Kultur Dokumente
4, APRIL 2001
737
AbstractIn order to model the RF behavior of a device-under-test (DUT), e.g., active and passive devices, dedicated
on-wafer test-structures are required. However, parasitic components in the test-structure stemming from the contact pads, the
metal interconnections and the silicon substrate, largely influence
the RF behavior of the actual DUT. They need to be subtracted
from the measurement results if one wants to model the RF behavior of the actual DUT accurately. This subtraction procedure
is referred to as de-embedding. In this paper, we propose an improved three-step de-embedding method to subtract the influence
of parasitics. The de-embedding method has been applied not only
to -parameter measurement results on MOSFETs but also, for
the first time, to large-signal vectorial RF measurements.
Index TermsIntegrated circuit modeling, microwave measurements, MOSFETs, nonlinear network measurement system
(NNMS), scattering parameters measurement, semiconductor
device measurements, semiconductor device modeling.
I. INTRODUCTION
ESIGNING RF circuits requires accurate models to describe the RF behavior of the devices used in the circuit.
For example, to model the RF behavior of MOS transistors,
compact models, like BSIM3v3 [1] and MM9 [2], or equivalent circuit-based models, like those proposed in [3], can be
used. Their model parameters are extracted from -parameter
measurements on dedicated on-wafer test-structures. These RF
test-structures not only consist of the actual device-under-test
(DUT) for which the model parameters need to be extracted,
e.g., a MOSFET, a bipolar transistor, or a passive component,
but also of parasitic components that largely influence the electrical behavior of the DUT. The parasitic components mainly
originate from the contact pad, which connects the RF measurement probe and the silicon wafer, and from the metal interconnections between these contact pads and the DUT. In order to
model the RF behavior of the DUT accurately, the influence of
the parasitic components must be subtracted from the measureManuscript received February 21, 2000; revised August 18, 2000. The work
of E. P. Vandamme was supported by IWT. The work of D. M. M.-P. Schreurs
was supported by the FWO. The review of this paper was arranged by Editor G.
Baccarani.
E. P. Vandamme is with IMEC vzw, B-3001 Leuven, Belgium and also with
the Katholieke Universiteit Leuven, ESAT-INSYS, B-3001 Heverlee, Belgium
(e-mail: vandamme@imec.be).
D. M. M.-P. Schreurs is with the Katholieke Universiteit, Leuven, ESATTELEMIC, B-3001 Heverlee, Belgium.
C. van Dinther is with Philips Semiconductors, 6534 AE Nijmegen, The
Netherlands.
Publisher Item Identifier S 0018-9383(01)02353-X.
ments on the test-structure. The procedure to correct for the influence of the on-wafer parasitic components is called de-embedding. After de-embedding, the correct , , or -parameters
of the DUT are obtained. These parameters can then be used for
example to extract RF model parameters.
It should be noted here that, contrary to IIIV technologies
which are manufactured on isolating substrates, the parasitic
components in silicon-based RF test-structures are very difficult
to calculate accurately by electromagnetic simulations. Therefore, an on-wafer de-embedding technique for silicon-based
technologies is mandatory.
In their paper, Cho and Burk [4] already pointed out that the
-parameter subtraction technique is not a sufficient de-embedding method, because the parasitic components in series
with the DUT are not negligible. To de-embed the on-wafer
parasitic components, they proposed a three-step de-embedding
technique that requires four additional on-wafer de-embedding
structures, namely, an open, a short1, a short2, and a through.
Based on their analysis, we propose an improved three-step
de-embedding technique. Our method proves to be more
accurate at higher frequencies, which is important in view of
the ever-increasing RF performance of silicon-based transistors
and the corresponding model accuracy required at these high
frequencies. Besides, we also show for the first time how to
de-embed large-signal vectorial RF measurements.
This paper is organized as follows. In the next section, we
propose our modifications to the three-step de-embedding technique by Cho and Burk. We show how to de-embed the effect of
parasitic contributions from -parameter measurements as well
as from vectorial large-signal measurements. Section III discusses the main advantages of our improved three-step de-embedding technique compared to the method by Cho and Burk
[4]. Small-signal and large-signal measurement results, that are
de-embedded using our improved de-embedding method, are
also shown and will be compared with modeling results.
738
(2)
can be calculated as
from which the value of
. All parasitic admittance and impedance
values can be calculated accordingly from the measured -paand
of the open structure
, of
rameters
, of the short1 structure, and
the through structure
of the short2 structure. They are given by
(3)
(4)
(5)
(6)
Fig. 1. (a) Layout of the RF test-structure containing the DUT in the
area enclosed by the dashed line. (b) Magnified view of the layout of the
de-embedding structures. The pad layout and interconnection layout are equal
to the test-structure. The shaded areas represent the metal.
(7)
(8)
As pointed out in [4], it can be useful to extract an equivalent circuit for these parasitic elements. For example, this greatly simplifies the model implementation of the complete test-structure,
i.e., DUT and parasitic elements, in a SPICE-like circuit simulator. Then also the behavior of the complete test-structure could
be easily analyzed at frequencies outside the measurement frequency range. Nevertheless, for the purpose of de-embedding
the measurement results and subsequent modeling of the DUT
itself, which is the focus of this paper, the extraction of an equivalent circuit for the parasitic elements is not required as such.
B. Model Assumptions and Improvements Compared to the
Original Three-Step De-Embedding Method
In the derivation of (6)(8) we assumed that
(9)
. Our measurements indicate that the ratio
decreases with frequency, but it is always
larger than 60 in the measured frequency range. Equation (9)
can be used to check the validity of the de-embedding method.
Another assumption is that the contact impedance is identical
with
or
739
Fig. 2. (a) Equivalent circuit of the RF test-structure. The DUT in this paper
is a MOS transistor. This circuit represents the actual layout of the complete RF
test-structure, as shown in Fig. 1(a). (b) Equivalent circuits of the open, short1,
short2, and through de-embedding structures.
740
741
erties of the silicon substrate. Therefore, the best possible verification is to use the measurement results of those -parameters
from the de-embedding structures that are not used in (3)(8)
up to
. These -parameters are
to calculate values for
then compared with their theoretically calculated value from the
equivalent circuit in Fig. 2(b). This comparison, presented in
and
, shows that both
Fig. 6 for
and
are modeled accurately up to 50 GHz. This analysis also indicates that the layout of the de-embedding structures
is correctly represented by the equivalent circuits in Fig. 2(b).
show large deviations, but this
Modeled and measured
is because the measured -parameters are below the noise floor,
.
especially
Fig. 6. Measured (solid line) and modeled (symbols) y -parameters of the
de-embedding structures to check the validity of the de-embedding procedure.
Modeling results are obtained using the equivalent circuit of the de-embedding
structures, shown in Fig. 2, and the calculated values for G up to Z , (3)(8).
742
extracted using Chos method. This is related to the limited frequency range (20 GHz) of the present RF large-signal measurement equipment. Nevertheless, because no information is found
in literature about the de-embedding of large-signal vectorial RF
measurements, the aim here is to show how these measurements
can be de-embedded up to the level of the DUT.
IV. CONCLUSIONS
We proposed an improved de-embedding method to accurately account for the influence of pad parasitics in silicon RF
test-structures. Our method proves to be more accurate at high
frequencies than the existing three-step de-embedding method.
We successfully applied our method to -parameter measurements and, for the first time, to vectorial large-signal RF measurements.
REFERENCES
[1] W. Liu, R. Gharpurey, M. Chang, U. Erdogan, R. Aggarwal, and J.
Mattia, R. F. MOSFET modeling accounting for distributed substrate
and channel resistances with emphasis on the BSIM3v3 SPICE model,
in IEDM Tech. Dig., 1997, pp. 309312.
[2] R. Vanoppen, J. Geelen, and D. Klaassen, The high-frequency analogue
performance of MOSFETs, in IEDM Tech. Dig., 1994, pp. 173176.
[3] C. Biber, M. Schmatz, T. Morf, U. Lott, and W. Bachtld, A nonlinear microwave MOSFET model for spice simulators, IEEE Trans.
Microwave Theory Tech., vol. 46, no. 5, pp. 604610, 1998.
[4] H. Cho and D. Burk, A three step method for the de-embedding of high
frequency S -parameter measurements, IEEE Trans. Electron Devices,
vol. 38, no. 6, pp. 13711375, 1991.
Cees van Dinther studied electrical engineering with an emphasis on telecommunication at the Technical University of Eindhoven, Eindhoven, The Netherlands, and received the M.Sc. degree in 1999.
In 1999, he joined Philips Semiconductors, where he has been working on
the development of MMICs. Low-noise amplifiers and VCOs are his main field
of work, in silicon and recently also in SiGe. His activities include technology
characterization, modeling, designing, and testing.