IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 59, NO. 6, JUNE 2012 1607
Improved Methodology for Integrated
k -Value Extractions
Ivan Ciofi, Gianpaolo Borrello, Oreste Madia, Christopher J. Wilson, Bart Vereecke, and Gerald P. Beyer
Abstract—We present an improved methodology for integrated
k-value extractions that is based on the use of interconnect
schemes with deep hanging low-k trenches and reduced passi-
vation as a test vehicle. We perform a rigorous analysis for the
calculation of the error bar and the evaluation of the impact
of each source of uncertainty. We demonstrate the effectiveness
of our methodology in reducing the error bar on the extracted
integrated k-value at 90 nm half-pitch and discuss the limitations
at narrow pitches.
Index Terms—Damascene integration, dielectric constant,
k-value, low-k dielectric materials.
I. I NTRODUCTION
N
EW HIGHLY porous insulating materials with an ul-
tralow relative dielectric constant (k-value) are being
currently screened in order to identify the most promising
candidates for next-generation Cu damascene interconnects [1].
The increased porosity makes these materials more prone to
degradation during processing. Therefore, the impact of the
damascene flow on the dielectric properties and, in particu-
lar, on the k-value must be taken into careful consideration.
Blanket wafers are commonly used for quick and cost-effective
screening of low-k dielectrics and their compatibility with
process steps, such as plasma etch or ash, chemical–mechanical
polishing (CMP), and wet cleaning [2]. However, process con-
ditions can be only partially reproduced on blanket wafers [3].
Hence, characterization on patterned wafers is finally required
to qualify a certain material for a given technology node. The
investigated material is integrated in Cu damascene structures,
such as interdigitated interconnect capacitors, and the related
postprocessing characteristics are evaluated [4]. In this conclu-
sive phase, the accurate determination of the integrated k-value
is fundamental for the estimation of the actual gain in inter-
connect performance (RC delay) that can be achieved by the
use of the new material. Based on the ITRS roadmap, the inte-
grated k-value is expected to scale down by 0.2 with each new
technology generation. Therefore, accuracy of at least ±0.1 is
Manuscript received December 12, 2011; revised February 20, 2012;
accepted March 8, 2012. Date of publication May 5, 2012; date of current
version May 23, 2012. The review of this paper was arranged by Editor
V. R. Rao.
The authors are with the Imec, 3001 Leuven, Belgium (e-mail:
Ivan.Ciofi@imec.be).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TED.2012.2196436
Fig. 1. Schematic cross sections of the inspected interconnect schemes:
(a) and (b) hanging low-k trenches of different depth; (c) landing low-k
trenches on the SiCN etch-stop layer; and (d) shallow hanging oxide trenches,
included as a control. The passivation consists of a thin layer of SiCN.
required in order to be able to benchmark the new material
against the existing ones from previous generation. The method
for the evaluation of the integrated k-value is rather complex. In
fact, the k-value is extracted from the measured line-to-line ca-
pacitance by means of 2-D model simulations. The 2-D model
is defined on the base of transmission electron microscopy
(TEM) cross sections and realistic assumptions on the dielectric
constants of the other layers in the damascene stack. The
simulations are carried out by a 2-D static field solver in order
to determine the dependence of the line-to-line capacitance on
the k-value of the investigated low-k material. The integrated
k-value is obtained by matching the model-based capacitance
with the measured capacitance. Commonly, single-damascene
structures are used as a test vehicle. In this case, the etch stop
and passivation layers of a higher k-value will significantly con-
tribute to the line-to-line capacitance [see Fig. 1(c)]. Therefore,
small uncertainties on the dielectric constants and thicknesses
of these parasitic layers will result in significant errors on
the extracted k-value. Errors may also originate from a poor
modeling of the line cross section, resulting, for instance, from
a basic polygon approximation of an irregular trench profile. To
the best of our knowledge, problems of accuracy in integrated
k-value extractions have never been thoroughly investigated in
literature. Furthermore, the error bar, which must be taken into
account for fair comparisons between different low-k materials
and processes, is typically not provided. In previous work,
we reported on the accuracy limitations of industry standard
procedures. As a result, an insufficient accuracy of ±0.3 was
evaluated for integrated k-value extractions at 90 nm half-pitch
for 2.2-2.5 low-k materials, when conventional damascene
structures are used [5].
In this paper, we evaluated an unconventional interconnect
scheme with deep hanging trenches as an optimized test ve-
hicle for integrated k-value extractions [see Fig. 1(a)]. The
0018-9383/$31.00 © 2012 IEEE