Low-temperature bottom-up integration of carbonnanotubes for vertical interconnects in monolithic
3D integrated circuits
Sten Vollebregt, Ryoichi Ishihara, Johan van der Cingel and Kees BeenakkerDelft University of Technology
Delft Institute of Microsystems and Nanotechnology
Feldmannweg 17, 2628CT, Delft, The Netherlands
Email: [email protected]
Abstract—Carbon nanotubes (CNT) can be an attractive can-didate for vertical interconnects in 3D monolithic integration,due to their excellent thermal and electrical properties. In thispaper we investigate the use of a true bottom-up approach tofabricate CNT vias, for application in 3D monolithic integration.This circumvents metal deposition in high aspect ratio holes,and also allows the use of bundle densification techniques toincrease CNT density. Using this approach we fabricated four-point probe electrical measurement structures for both as-grownand densified CNT bundles, and performed I-V measurements.The resulting I-V curves display non-linearities due to a non-Ohmic top contact. The measured resistivities of 10-20 mΩ-cmare among the better values found in literature.
I. INTRODUCTION
Three dimensional integration can offer high integration
density, shorter routing and increased functionality by com-
bining for instance sensors and logic. Of the different ap-
proaches that exist for 3D integration, monolithic (device-
level) integration offers the highest integration density and
shortest interconnect length. In this approach transistor layers
are stacked on top of each other on the same die across the
entire wafer.
Many challenges still remain to be solved in order to achieve
monolithic integration. One of the biggest is creating the active
silicon layers. Our group has previously been working on
monolithic integration using single-grain thin-film transistors
(SG-TFTs) [1]. These transistors offer silicon-on-insulator
(SOI) level performance, while processing temperature re-
mains below 550 ◦C. Using two layers of transistors we
demonstrated memory circuits and combined image sensor
with read-out [2].
Two other major remaining challenges are manufacturing
reliable low resistance vertical interconnects and thermal man-
agement. Already in modern circuits vertical interconnects
(vias) display many reliability issues due to difficulties in
filling the small openings. For monolithic integrated circuits
we can expect these issues to be worse, as the oxide required
to isolate the different active layers is relative thick. Thermal
management is an issue as the amount of transistors is in-
creased drastically, while the surface area of the die remains
roughly the same.
Carbon nanotubes (CNT) can be an attractive candidate to
tackle both issues. It has been shown that CNT can have
unrivalled electric properties, allowing ballistic transport over
a length of several μm [3], [4]. Moreover, CNT have been
demonstrated to have very good reliability properties, allowing
a current density up to three order of magnitude higher than
copper [5]. Beside good electrical properties CNT can be
grown with very high aspect ratios. Finally, CNT have been
demonstrated to posses very good thermal properties with
thermal conductivities reaching 3500 W/mK [6]. Much interest
has already been shown in using CNT as vias for regular 2D
integrated circuits [7]–[10].
We propose to use CNT as via material in monolithic
integrated 3D circuits. For this a bottom-up approach as
suggested by Li et al. [11] should be most suitable. In this
approach the CNT are grown before the deposition of the
dielectric layer, thus circumventing all the issues involved in
filling high aspect ratio holes. As CNT can be grown vertically
well aligned integration density can be high.
In this paper we created low-temperature and high density
multi-walled CNT (MWCNT) using this bottom-up approach
and performed electrical resistance measurements. Moreover,
we investigated the use of a bundle densification technique in
order to reduce via dimensions and resistivity by increasing
the CNT tube density.
II. EXPERIMENTAL
As substrate 4” p-type Si wafers with (100) are used. These
wafers are metallized using a Trikon Sigma sputter coater
with 500 nm of Ti and 50 nm TiN, which acts as bottom
contact to the CNT bundles. TiN is used as this is a standard
barrier material in the IC industry, doesn’t oxidise during wafer
transfer between machines (for catalyst deposition) and allows
the growth of vertically aligned MWCNT at low temperatures
in our CNT growth reactor.
Next a 5 nm thick Fe catalyst layer is deposited using an
CHA Solution e-beam evaporator and patterned using lift-off.
For this lift-off process we employ AZ nLOF pure negative
resist, which is exposed by an ASML PAS 5500/80 waferstep-
per. The actual lift-off is performed in n-methyl pyrrolidone
�����
������
Fig. 1. Process flow from first metal, including catalyst deposition, LPCVDCNT growth, optional densification, PECVD oxide deposition, CMP and finalmetallization
at 70 ◦C. After this CNT are grown using the commercially
available AIXTRON BlackMagic Pro CNT reactor. The first
phase of the growth process consists of a activation step with
700 sccm of H2 for 3 min. at 500◦C and 80 mbar. After this
50 sccm of C2H2 is added to start the CNT growth phase.
Total growth time equals 15 minutes.
After CNT growth 5 μm of PECVD oxide is deposited using
a Novellus Concept One reactor at 400 ◦C. Before performing
this step the CNT of some of the wafers are densified using
isopropanol. Oxide depoistion is followed by planarization
using chemical mechanical polishing (CMP) with a silica
slurry. A total of approximately 1 μm of oxide is removed
in this process. Finally, the second electrode consisting of 100
nm of Ti and 2 μm of Al(1% Si) is deposited and patterned
using wet etching. Finally, contact holes are etched through the
oxide in order to contact the first electrode directly. In fig. 1
an overview is given of the bottom-up process. Samples were
characterised using an FEI/Philips XL50 scanning electron
microscope (SEM) and Agilent 4156C parameter analyser.
III. RESULTS AND DISCUSSION
In fig. 2a a SEM image taken from an array of five by
five 2 μm wide CNT bundles with a spacing of 3 μm and
height of roughly 5 μm. As can be observed the CNT and
bundles display good vertical alignment. Fig. 2b displays a
single 2 μm wide bundle with a length of almost 10 μm,
demonstrating the ability to grow high aspect ratio (HAR) free
standing bundles. The density of the CNT inside the bundles
is ∼ 1011 tubes/cm2, while the average tube diameter is 5-
15 nm. The CNT tube length can be changed by varying the
deposition time, which has a linear dependency against length,
with a growth rate of 350 nm/minute.
We use PECVD oxide as it can be deposited at low temper-
atures. Previously, we reported that we didn’t observe a major
degradation in tube crystallinity due to plasma damage while
depositing the oxide [12]. The CNT are uniformly covered by
the PECVD oxide, as can be seen in fig. 2c. It can also be
clearly seen that CMP is required to be able to contact the top
of the CNT bundles.
Finally fig. 2d displays the array of CNT after approxi-
mately 1 μm of oxide was removed using CMP. As can be seen
the CNTs are cut-off at the same height as the oxide. Gaps can
(a) CNT directly after growth (b) CNT with high aspect ratio
(c) CNT covered by PECVD oxide (d) Covered CNT after CMP
(e) Crack in larger CNT bundle
Fig. 2. SEM images taken from different stages in the bottom-up process
exist between CNT bundles spaced closed to each other, due to
a non optimized PECVD oxide deposition process. Decreasing
deposition rate likely solves this issue. As shown in fig. 2e
in some larger CNT bundles cracks appear with a width of
several hundred nm, which are caused by the non-optimized
CMP process.
It must also be noted that we encountered difficulties in
polishing large bundles with widths of 6 μm or wider. Most
of these bundles are removed during CMP, which we attribute
to mechanical removal during the CMP step as the substrate
adhesion of CNT is generally low [13]. The bundles are only
covered on the outside by the PECVD oxide. Embedding
the CNT in a layer of atomic layer deposited dielectric, as
demonstrated by Chiodarelli et al. [14], to increase mechanical
stability could circumvent this problem and the formation of
cracks in large bundles.
Although the density of our bundles is relative high, it is
still orders of magnitude lower than ideal tight packing in
which case density can be as high as 1013 − 1014 tubes/cm2.
It has been demonstrated that CNT density can be increased
by dipping the tubes into isopropanol [15]. This densification
is caused by the capillary forces of the liquid combined with
strong Van der Waals interaction between the tubes upon liquid
� � �
(a) Bundle of initially 2 μm (b) Bundle of initially 5 μm
Fig. 3. SEM images taken from bundles after densification
evaporation.
We employed this technique by simply dipping the full
wafers upside-down into an isopropanol bath, followed by
drying of the wafers in air. As can be from fig. 3 the width
of the resulting bundles are 72 % and 74 % of the initial
width for the 2 μm and 5 μm bundle, respectively. This is
representative for all bundle sizes over the entire wafer. The
resulting area after densification is thus almost 50 % of the
initial bundle area, drastically reducing the footprint of the top
CNT bundle contact. The bottom contact remains the same,
due to the strong enough surface adhesion of the CNT. The tip
of the CNT bundle is less densified, but conveniently removed
during the CMP process.
We created four-point probe measurement structures using
this bottom-up process to measure the resistance of the CNT
bundles. Using the top and bottom electrode contacts a current
could be forced through the bundle, which could be current-
less measured using two different top and bottom contacts.
Using this structure only the resistance of the CNT bundle,
and the contact resistances to the top and bottom electrode
are measured. A sketch of the measurement structure can be
found in the inset of fig. 4.
Fig. 4 displays typical I-V characteristics obtained from
several bundles with a width of 1 μm and a length in between
3.8 and 4 μm. As can be seen there exist a non-linear
component in the measured I-V curves. This is most likely
caused by a non-Ohmic contact. We suspect the top contact,
as the bottom contact should still be similar to those used for
top-down results reported earlier by our group, which didn’t
suffer from any non-linearities [12]. These non-linearites are
also visible for bundles with a width of 2 and 4 μm, which
have a larger contact areas. Improving the CMP conditions,
or removing part of the silicon oxide to expose the CNT tips
could improve the top contact. Annealing might also solve
the problem, although we found that annealing the wafers
in a nitrogen atmosphere at 400 ◦C actually degraded the
resistance.
From the five best performing structures we determined the
average resistivity and standard deviation using the approxi-
mate dimensions of the CNT bundle. The same was done for
the densified bundles, assuming a final width of 75 % of that
of the initial width (1.5 and 3 μm for, respectively, the initially
2 and 4 μm bundles). The resulting resistivities are plotted in
Fig. 4. I-V characteristics obtained from several CNT bundles with 1 μmwidth. The inset shows the used 4-point probe structure, the black cilinderindicates the CNT bundle.
fig. 5 together with results obtained by top-down integration
performed before in our group and results taken from literature
about CNT vias for 2D integrated circuits.
As can be seen the results are among the better values
reported in literature, although still many orders of magnitude
higher than the bulk resistivity of for instance Cu. For small
bundle sizes the resistivity increases sharply for the as-grown
CNT bundles. We suspect that the top contact is the cause of
this deviation for these bottom-up results. This is because the
measurement results of the densified 2 μm bundle are much
better, which also display hardly any non-linearities in the I-V
characteristics indicating better contact. The CMP processed
used for the densified and non-densified bundles differ, for the
latter it was done with a pH 7 slurry, while the first uses the
standard slurry. Making the slurry pH neutral resulted in much
longer polish times, likely wearing down the tips of the tube
resulting in a higher contact resistance.
For the larger 4 μm bundles the resistivity spread is much
larger, which could be caused by the before mentioned cracks
observed in some of the larger structures. Overall, although
a different CMP approach was used, the values between
densified and non-densified bundles are not for apart. The
obtained values using bottom-up integration are close to those
reported before using the same CNT growth recipe, but a top-
down approach [12]. For the larger and densified bundles the
resulting resistivities are even lower. This confirms that the
bottom-up approach does not degrade CNT quality. The higher
resistivities measured can be attributed to a higher top-contact
resistance, which can be improved by optimizing the CMP
parameters.
IV. CONCLUSION
Carbon nanotubes can be an attractive candidate for vias
in 3D monolithic integration due to their good electrical and
thermal properties. We demonstrate, for the first time, that is
is possible to use a pure bottom-up approach using carbon
�����
Fig. 5. Calculated average resistivities and standard deviation for normaland densified bundles, compared to literature.
nanotubes to create vias. With this approach it is possible
to use bundle densification techniques, which decrease the
top area of the via. Four-point probe I-V measurements
were performed for both densified and non-densified bundles,
displaying a non-linear resistance due to a non-Ohmic top
contact. The measured resistivities of 10-20 mΩ-cm are among
the better values reported in literature.
ACKNOWLEDGMENT
The authors want to thank T.M. Michielsen, F.J.H. van der
Kruis and O.J.A. Buijk at Philips Innovation Services MiPlaza
for performing the CMP. This project is kindly funded by the
European ENIAC Joint Undertaking as part of the JEMSiP 3D
project.
REFERENCES
[1] M. R. T. Mofrad, J. Derakhshandeh, R. Ishihara, A. Baiano, J. van derCingel, and K. Beenakker, “Stacking of single-grain thin-film transis-tors,” Japanese Journal of Applied Physics, vol. 48, p. 03B015, 2009.
[2] J. Derakhshandeh, N. Golshani, R. Ishihara, M. R. T. Mofrad, M. Robert-son, T. Morrison, and C. I. M. Beenakker, “Monolithic 3-D integrationof SRAM and image sensor using two layers of single-grain silicon,”IEEE Transactions on Electron Devices, vol. 58, no. 11, pp. 3954–3961,2011.
[3] D. Mann, A. Javey, J. Kong, Q. Wang, and H. Dai, “Ballistic transportin metallic nanotubes with reliable Pd ohmic contacts,” Nano Letters,vol. 3, no. 11, pp. 1541–1544, 2003.
[4] H. J. Li, W. G. Lu, J. J. Li, X. D. Bai, and C. Z. Gu, “Multichannelballistic transport in multiwall carbon nanotubes,” Physical ReviewLetters, vol. 95, p. 086601, 2005.
[5] B. Q. Wei, R. Vajtai, and P. M. Ajayan, “Reliability and current carryingcapacity of carbon nanotubes,” Applied Physics Letters, vol. 79, no. 8,pp. 1172–1174, 2001.
[6] E. Pop, D. Mann, Q. Wang, K. Goodson, and H. Dai, “Thermalconductance of an individual single-wall carbon nanotube above roomtemperature,” Nano Letters, vol. 6, no. 1, pp. 96–100, 2006.
[7] J. Dijon, H. Okuno, M. Fayolle, T. Vo, J. Pontcharra, D. Acquaviva,D. Bouvet, A. M. Ionescu, C. S. Esconjauregui, B. Capraro, E. Quesnel,and J. Robertson, “Ultra-high density carbon nanotubes on Al-Cu foradvanced vias,” in International Electron Devices Meeting, 2010, p.33.4.1.
[8] D. Yokoyama, T. Iwasaki, K. Ishimaru, S. Sato, T. Hyakushima, M. Ni-hei, Y. Awano, and H. Kawarada, “Electrical properties of carbon nan-otubes grown at a low temperature for use as interconnects,” JapaneseJournal of Applied Physics, vol. 47, no. 4, pp. 1985–1990, 2008.
[9] Y.-M. Choi, S. Lee, H. S. Yoon, M.-S. Lee, H. Kim, I. Han, Y. Son,I.-S. Yeo, U.-I. Chung, and J.-T. Moon, “Integration and electricalproperties of carbon nanotube array for interconnect applications,” inNanotechnology, 2006. IEEE-NANO 2006. Sixth IEEE Conference on,2006, pp. 262–265.
[10] F. Kreupl, A. P. Graham, G. S. Duesberg, W. Steinhogl, M. Liebau,E. Unger, and W. Honlein, “Carbon nanotubes in interconnect applica-tions,” Microelectron. Eng., vol. 64, pp. 399–408, 2002.
[11] J. Li, Q. Ye, A. Cassell, H. T. Ng, R. Stevens, J. Han, and M. Meyyap-pan, “Bottom-up approach for carbon nanotube interconnects,” AppliedPhysics Letters, vol. 82, no. 15, pp. 2491–2493, 2003.
[12] S. Vollebregt, R. Ishihara, J. Derakhshandeh, J. van der Cingel,H. Schellevis, and C. Beenakker, “Integrating low temperature alignedcarbon nanotubes as vertical interconnects in si technology,” in Nan-otechnology, 2011. IEEE-NANO 2011. 11th IEEE Conference on, 2011,pp. 985–990.
[13] H.-C. Su, C.-H. Chen, Y.-C. Chen, D.-J. Yao, H. Chen, Y.-C. Chang, andT.-R. Yew, “Improving the adhesion of carbon nanotubes to a substrateusing microwave treatment,” Carbon, vol. 48, pp. 805–812, 2010.
[14] N. Chiodarelli, “Integration of carbon nanotubes as future intercon-nections for sub-32nm technologies,” Ph.D. dissertation, KatholiekeUniversiteit Leuven, 2011.
[15] D. N. Futaba, K. Hata, T. Yamada, T. Hiraoka, Y. Hayamizu, Y. Kaku-date, O. Tanaike, H. Hatori, M. Yumura, and S. Iijima, “Shape-engineerable and highly densely packed single-walled carbon nanotubesand their application as super-capacitor electrodes,” Nature Materials,vol. 5, pp. 987–994, 2006.
�����