Electrical properties of Ta
2
O
5
thin ®lms deposited on Cu
S. Ezhilvalavan
1, Tseung-Yuen Tseng*
Department of Electronics Engineering and Institute of Electronics, National Chiao Tung University, Hsinchu 300, Taiwan Received 11 February 1999; received in revised form 15 September 1999; accepted 13 October 1999
Abstract
The electrical and dielectric properties of reactively sputtered Ta2O5thin ®lms with Cu as the top and bottom electrodes forming a simple
metal insulator metal (MIM) structure, Cu/Ta2O5/Cu/n-Si, were studied. Ta2O5®lms subjected to rapid thermal annealing (RTA) at 8008C for
30 s in N2ambient crystallized the ®lm, decreased the leakage current density and resulted in reliable time-dependent dielectric breakdown
characteristics. The conduction mechanism at low electric ®elds (,100 kV/cm) is due to Ohmic conduction; however, the Schottky mechanism becomes predominant at high ®elds (.100 kV/cm). Present studies demonstrate the use of Cu as a potential electrode material to replace the conventional precious metal electrodes for Ta2O5storage capacitors. q 2000 Elsevier Science S.A. All rights reserved.
Keywords: Electrical properties; Copper; Tantalum; Oxides
1. Introduction
New capacitor dielectric materials with high dielectric constants are needed for advanced dynamic random access memory (DRAM) cell technologies if they are to keep up with the scaling rule. Ta2O5thin ®lm capacitors are
consid-ered as one of the best alternatives to conventional ultra-thin silicon dioxide which has reached its physical limits below 4 nm, or other thin ®lm insulators such as oxide±nitride± oxide structures in terms of good dielectric properties [1±3]. As the DRAM generation goes 256 Mbit and beyond, the DRAM fabrication process has become more and more complicated. This will cause the production cost of the high density DRAMs to become unacceptably high and will signi®cantly degrade the device reliability. Thus it is essential to develop a process technology that is simple and yet ensures high performance and high reliability.
In order for ultra large scale integrated circuit (ULSI) manufacturing to minimize the cost of ownership aspect in the metallization process, several metallization technol-ogies have been proposed. The evidential criteria in choos-ing the most probable methods are physical or material limitations (e.g. step-coverage and resistivity) and manufac-turing requirements such as process complexity, reliability, throughput and total cost.
Matsuhashi et al. [4] investigated the effects of top
elec-trode materials, metals (W, Mo, Ti and Ta) and their nitrates (WN, MoN, TiN and TaN) on the leakage current in Ta2O5
®lms before and after annealing and proposed Mo and MoN as better electrodes for high temperature processes. Poly-Si, Pt, TiN and W have been evaluated as bottom electrodes for Ta2O5dielectric ®lms [5]. It was reported that TiN and W
bottom electrodes had interfacial oxide layers which are smaller in thickness than that of the Si electrode, whereas Pt does not show any appreciable formation of an interfacial layer. The presence of an interfacial SiO2layer is
responsi-ble for the reduction of the dielectric constant in polycrystal-line Ta2O5®lms [6]. A poly-Si/TiN double layer was also
introduced as an upper electrode in Ta2O5capacitors where
TiN served as a barrier layer to prevent reaction between Ta2O5and poly-Si electrode under high thermal budget [7].
The application of poly-Si/TiN double electrode however has to be limited to relatively simple capacitor structures because of inherent poor conformability. Cu based intercon-nect metallization technology could be incorporated into devices by the turn of this century owing to ease of proces-sing and high reduction in production cost of DRAMs. Recently, there have been signi®cant improvements in various elements of Cu metallization process technology, including improved material properties, diffusion barriers, Cu deposition, Cu integration etc. Successful fabrication of semiconductor devices with good electrical performance, integrated with copper metallization has been demonstrated by Awaya et al. [8]. Current generation interconnect mate-rials are Al and Al±Cu alloy. They could be replaced in the future by Cu and Cu alloy. Cu is favorable as an
intercon-0040-6090/00/$ - see front matter q 2000 Elsevier Science S.A. All rights reserved. PII: S0040-6090(99)00873-1
www.elsevier.com/locate/tsf
* Corresponding author. Fax: 188-6-35-724-361. E-mail address: [email protected] (T.Y. Tseng)
1Present address: Physics and Astronomy, Michigan State University,
nect material since the upper limit of the current density to prevent electromigration for Cu is 5 £ 106 A/cm2, whereas
for Al it is 2 £ 105 A/cm2[9]. In this paper we report the
effect of Cu as a top and bottom electrode material on the electrical and dielectric properties of reactively sputtered Ta2O5®lms for the ®rst time. We also evaluated the
possi-bility of using Cu as the top and bottom electrode material for ULSI storage capacitors.
2. Experimental
The n-type silicon wafer was cleaned by a standard clean-ing process. The Cu bottom electrode on n-Si substrate with a thickness of 200 nm was deposited using a separate sput-tering system. The Cu ®lm was prepared at a ®xed power of 80 mW and at a constant pressure of 10 mTorr with Ar as the sputtering gas. Ta2O5®lms were deposited on the
Cu/n-Si bottom storage node electrode by dc-magnetron sputter-ing from a high purity tantalum metal target (2.5 inch diameter). More details on the deposition technique may be found in [10,11]. The sputtering gas was 80% Ar and 20% O2mixture with a total pressure of 10 mTorr. During
deposition, the chamber was ®rst back ®lled with Ar gas and used to pre-sputter clean the target for at least 5 min. Then the Ar/O2gases were introduced into the chamber to reach a
total pressure of 10 mTorr. Film thickness was estimated to be 100 nm using a Tencor Alpha-step 200 pro®lometer. The Cu top electrode with a thickness of 200 nm and diameters of 150, 250 and 350 mm were patterned by a shadow mask process. The current±voltage (I±V) characteristics of the Ta2O5 ®lms were measured on the MIM structure with a
HP4145B semiconductor parameter analyzer. The capaci-tance±voltage (C±V) characteristic and the dielectric loss tangent were recorded at frequencies ranging from 100 Hz to 1 MHz with 0.5 V ac sweeping signal using a HP4194A impedance-gain phase analyzer. The rapid thermal anneal-ing (RTA) of the Ta2O5 ®lm was performed in a RTA
furnace (Ulvac Sinku-Rico, HPC 700) at 8008C for 30 s in N2ambient, before the formation of the top electrode. The
heating rate used was the maximum heating rate of about 1008C/s. Crystalline phases of the Ta2O5®lms were
identi-®ed by X-ray diffractometry (XRD, model D5000, Siemens, Munich, FRG).
3. Results and discussion
Fig. 1 shows the XRD patterns of as-deposited and Ta2O5
®lms subjected to RTA at 8008C for 30 s in N2 ambient.
XRD results indicate that as-deposited ®lms were amor-phous and the annealed ®lms crystallized into b-Ta2O5.
The crystallinity increased with temperature in terms of an increase in the intensity of the diffracted peaks, while the thickness of the ®lm was kept constant. In addition, Fig. 1 also indicates the appearance of additional peaks corre-sponding to Cu (from the bottom electrode) and Cu2O.
The presence of Cu2O peaks is due to the formation of an
interfacial Cu2O layer at the Ta2O5/Cu interface. However,
the intensities of Cu2O peaks decreased after the RTA N2
annealing at 8008C. Ellipsometry measurements show that the refractive index (n) for the N2 annealed Ta2O5
(n 2:239) was higher than that of as-deposited Ta2O5
(n 2:149). The thickness of as-deposited Ta2O5 was
close to 100 nm and was reduced by ,2.5% after crystal-lization. Since we expect that Dn/n is proportional to Dr/r (wherer denotes the density) through the Lorentz±Lorentz formula [12], the proximity of the variations observed is satisfying (Dn=n , 4%) and con®rms the hypothesis of densi®cation induced during crystallization.
Fig. 2a,b depicts the relative counts of Cu and O atoms obtained from the X-ray photoelectron spectroscopy (XPS) analyses performed on the surfaces of as-deposited and N2
annealed Cu bottom electrode deposited on Si substrate. The samples for the XPS analyses were prepared as follows in order to keep the exact surface conditions of the Cu bottom electrode as used during the deposition of Ta2O5®lms, so
that the formation or reduction of Cu2O layer at the Ta2O5/
Cu interface can be envisaged. Ta2O5®lms were sputtered
onto the as-deposited Cu bottom electrode for 5 min to form a very thin layer of Ta2O5 of thickness #10 nm. One of
these samples was subjected to 8008C N2 RTA for 30 s.
Then, both the samples were ion etched for repeated cycles until the top Ta2O5layer was removed completely and XPS
analyses were performed on the freshly exposed as-depos-ited and N2 annealed Cu bottom electrode on Si. We
observed similar XPS spectra for the detected signal of Cu photoelectron, Cu(2p3/2), from the outermost surface of both
as-deposited and N2 annealed Cu bottom electrodes (Fig.
2a). The peak position for Cu(2p3/2) was detected at
Fig. 1. XRD spectra of Ta2O5®lm deposited on Cu: (a) as-deposited and (b)
932.5 eV as compared to the standard value of 932 eV, which indicates that the Cu bottom electrode preserved a high degree of Cu-elemental chemical state. The O(1s) spectra presented in Fig. 2b showed that the oxygen photo-electrons were in the Cu2O state. The O(1s) peak position
located at ,531 eV compares with the standard value. It is to be noted that the peak intensity of the O(1s) signal was decreased drastically for the N2annealed Cu bottom
trode in comparison to the as-deposited Cu bottom elec-trode. Therefore, XPS and X-ray analyses clearly demonstrate that the formation of Cu2O took place mostly
during the initial stages of Ta2O5reactive sputtering, which
was then dissociated when the ®lms were subjected to N2
annealing at 8008C for 30 s.
One of the most important features for a material to be used as an alternative storage dielectric in DRAM is the low leakage current density. Fig. 3 shows the I±V characteristics of the Ta2O5 thin ®lm MIM capacitors as deposited and
RTA processed at 8008C for 30 s in N2ambient. It is clear
that the leakage current of as-deposited amorphous Ta2O5
®lm is larger than that of RTA processed polycrystalline ®lm. The leakage current density of the as-deposited ®lm is ,1024A/cm2 at 100 kV/cm which is brought down by
nearly 4 orders to 1028 A/cm2 by RTA processing. This
value falls in the middle of the leakage current densities
recently reported for Ta2O5 ®lms, which range from 1027
A/cm2[13] to 10211A/cm2[14]. The reduction of leakage
current after the 8008C 30 s N2 RTA process might be
closely related to the decrease in defects such as broken bonds and improvement in the ®lm microstructure by way of higher densi®cation. The as-deposited ®lm could form an oxidized layer at the bottom electrode/Ta2O5 interface
during initial stages of Ta2O5 reactive sputtering, which
might be reduced during the 30 s 8008C N2RTA processing,
as evident from X-ray and XPS results, yet preserving lower oxygen vacancy in the ®lm, thereby restoring the leakage current density to ,1028A/cm2.
The oxidation of Cu bottom electrodes can be explained in terms of oxygen de®ciency as a result of the reaction between the substrates and the adsorbed oxygen as follows. If the adsorbed oxygen molecules or radicals react with the substrate fast enough to remove the possibility of oxidation from the Ta source, Ta2O5cannot be formed. As the surface
oxide becomes thicker the diffusion ¯ux of oxygen into the substrate becomes smaller, which alters the sputtered Ta ions to react with oxygen to form Ta2O5. Once the Ta2O5
®lm starts to grow, the oxidation of the substrate is actually stopped and the thickness of the interfacial oxide layer remains constant, because Ta2O5®lm is an effective
diffu-sion barrier material against oxygen [15].
The leakage current in a dielectric ®lm can be owing to several conduction mechanisms including Schottky emis-sion, Poole±Frenkel emisemis-sion, Fowler±Nordheim tunneling and a space charge limited current. Herein, the leakage current mechanisms of Cu/Ta2O5/Cu/n-Si capacitors are
investigated as well. Fig. 4 shows the Schottky emission (SE) plot for the as-deposited and the RTA processed Ta2O5 ®lms. Two distinct regions may be observed in the
I±V characteristic plotted in the form log10(J) versus E1/2. At
Fig. 3. I±V characteristics of the Ta2O5 ®lms, as-deposited and RTA
processed at 8008C for 30 s in N2ambient.
Fig. 2. XPS spectra of Cu bottom electrode: (a) Cu (2p3/2) and (b) O (1s); (i)
very low electric ®eld the current density increases approxi-mately linearly with the electric ®led displaying nearly Ohmic behavior. Also a plot of I versus V (Fig. 5) indicates that for very low electric ®elds, 0±100 kV/cm, the relation-ship is Ohmic [16]. At higher electric ®elds and higher current densities, the I±V relationship is no longer Ohmic. It shows a non-linearity and the currents become quadratic with voltage [16]. Fig. 6 shows the Arrhenius plot of temperature dependent leakage current for RTA processed Ta2O5 ®lm. It demonstrates two distinct slopes, i.e. a low
activation energy process at low temperatures (suggesting electron hopping from one trap to the other with low mobi-lity [17,18]), and a high activation energy mechanism at higher temperatures. The activation energies calculated from the measured slopes at low and higher temperatures are 0.08 eV and 0.39 eV respectively. It is also to be noticed that the current density of the ®lm varies with temperature nearly in the form of J / 1=T1=4, at lower temperatures and low ®elds, thus demonstrating the existence of hopping conduction [16±19]. Therefore, the current at lower electric ®elds (,100 kV/cm) in Fig. 4 could be due to hopping conduction, because the thermal excitation of the trapped electrons from one site to the other dominate transport in the ®lm; this is given by [19]
J sEexp 2Eÿ a=KT 1
where Eais the activation energy of hopping electrons. But
at higher electric ®elds .100 kV/cm (Fig. 4), the current densities are proportional to the square root of the applied electric ®eld which extend further with the ®eld. Further-more, we were able to ®t the current density variation at higher electric ®eld in the form of a straight line, as indi-cated in Fig. 4, which satis®es the SE process. The linear relation at higher electric ®eld demonstrates the dominance of SE process across the interface between the dielectric
®lm and the electrode as a result of barrier lowering due to the applied ®eld and the image force [10,14,16]. The current (JSE) governed by the SE mechanism is described
as [19] JSE AT2exp 2q f b2 qV=4ÿ peid1=2=KT h i 2 where A denotes a constant,fbthe Schottky barrier height,
ei the dielectric constant of the insulator, V the applied
voltage and d the insulator thickness. The comparison of calculated dielectric constant from the slope of the straight line portion of the SE plot with the experimentally deter-mined (from C±V measurement at 100 kHz) value further con®rmed the existence of the SE process for the present ®lms. In the Cu/Ta2O5/Cu/n-Si structure, electrons are
injected from n-Si into Cu when the top electrode is posi-tively biased. Since the work functions of Ta2O5and Cu are
4.05 eV [20] and 4.7 eV [20] respectively, the barrier height at the Cu/Ta2O5interface is smaller than that of the normally
used Pt/Ta2O5interfaces (because of the higher work
func-tion of Pt ,5.65 eV [20]). Therefore, for the same applied ®elds $100 kV/cm, the number of electrons injected from the Cu/Ta2O5interface into the dielectric ®lm is higher than
that from the Pt/Ta2O5interface. In other words, the leakage
current in the Ta2O5 ®lm increases with decreased work
function of the top electrode. This result indicates that the current ¯owing through the Cu/Ta2O5interface is limited by
the SE process. Thus the I±V characteristics clearly demon-strate the existence of two possible dominant conduction mechanisms for Cu/Ta2O5/Cu/n-Si MIM capacitors. The
problem of higher leakage current density in Ta2O5 ®lms
with the use of Cu as an electrode can be improved by carrying out the N2 RTA processing at 8008C for 30 s
after the deposition of the top electrode. It has been reported that annealing the dielectric ®lm (Ba,Sr)TiO3with both top
and bottom electrodes in reducing atmosphere (N2) creates
an n-type conductivity in the dielectric ®lm and produces a high interface energy barrier, which plays a major role in reducing the current density when the bias voltage is applied
Fig. 5. Current±voltage plot of leakage current in RTA processed Cu/ Ta2O5/Cu/n-Si capacitor.
[21]. Hence we can imagine that the decreased work func-tion of the top electrode may not be a disadvantage in using Cu as an electrode in dielectric ®lms.
Fig. 7a,b displays the results of the dielectric studies performed on the Cu/Ta2O5/Cu/n-Si MIM con®guration. It
shows the variation of accumulation capacitance as a func-tion of logarithmic frequency (a) for the as-deposited Ta2O5
®lm and (b) for a Ta2O5®lm RTA processed at 8008C for
30 s in N2ambient, for frequencies ranging from 100 Hz to
1 MHz. The capacitance of the as-deposited ®lm decreased from 1.0 F/m2at 100 Hz to 2 £ 1023F/m2at 1 MHz and the
dielectric loss tangent falls from a high value of 12 at 100 Hz to 1 at 1 MHz. The capacitance of the RTA processed ®lm, however, shows less variation with frequency, i.e. it falls from 4:6 £ 1023 F/m2at 100 Hz to
3:1 £ 1023 F/m2at 1 MHz. The loss tangent is a fairly low
value and it varies from 0.06 to 0.01 in the above measured frequency range. The dielectric constants of the as-depos-ited and the RTA processed Ta2O5®lms calculated from the
capacitance measured at 100 kHz are 30 and 40, respec-tively. The presented dielectric constant is higher than that reported in the literature [14,16]. This difference may be attributed to the differences in the processing methods, processing temperature and ambient and also in the resultant structure-phase modi®cations. The large capacitance varia-tion and the associated higher dielectric loss tangent at low frequencies for the as-deposited ®lms may be attributed to the higher leakage current density. The RTA processing at 8008C for 30 s leads to complete crystallization and higher densi®cation resulting in lower leakage current density, which may be the probable reason for the low dielectric loss tangent of the polycrystalline ®lm. Furthermore, RTA processing in N2ambient provides a reducing interface for
the oxidized surface layer of the Cu electrode, yet preser-ving a lower oxygen vacancy concentration in the dielectric ®lm. The reason is that the value of the formation energy of Ta2O5at 8008C (2597 kJ/mol) [20] is much more negative
than that of Cu2O (260 kJ/mol) [20], therefore the
dissocia-tion of oxygen from the Ta2O5 ®lm is less probable than
from Cu2O during the short duration (30 s) of the RTA
processing in N2.
Time-dependent dielectric breakdown (TDDB) is a char-acteristic of the intrinsic materials, the method of processing and electrode materials. Fig. 8 shows the lifetime extrapola-tion from the dependence of the cumulative failure on TDDB stress time for RTA processed Cu/Ta2O5/Cu/n-Si
thin ®lms. The TDDB lifetime for Ta2O5 ®lm with the
conventional Pt electrode MIM structure is also shown for comparison [14]. Present studies demonstrate that Ta2O5
MIM ®lms with Cu as the top and bottom electrode can also survive the 10 years lifetime at a stress ®eld of $700 kV/cm. However, we believe that more optimal conditions of RTA processing in N2are required to improve
the Cu/Ta2O5/Cu/n-Si capacitor performance. The details
have to be separately worked out. 4. Conclusion
We have successfully demonstrated the effective use of Cu as a possible electrode material replacing the conven-tional precious metal electrodes for Ta2O5®lm storage
capa-Fig. 7. Capacitance as a function of logarithmic frequency for the Ta2O5
®lm: (a) as-deposited and (b) RTA processed ®lm. Fig. 6. Arrhenius plot of temperature dependent leakage current for Ta2O5
citors. Usage of Cu as an electrode will signi®cantly reduce the production cost of future high density DRAMs. Acknowledgements
The authors gratefully appreciate the ®nancial support from the National Science Council of R.O.C under project no. NSC 87-2218-E 009-008.
References
[1] S. Kamiyama, H. Suzuki, H. Watanabe, et al., IEDM Tech. Dig. (1993) 49.
[2] H. Shinriki, M. Nakata, IEEE Trans. Electron Devices 38 (1991) 455. [3] F.C. Chiu, J.J. Wang, J.Y. Lee, S.C. Wu, J. Appl. Phys. 81 (1997)
6911.
[4] H. Matsuhashi, S. Nishikawa, Jpn. J. Appl. Phys. 33 (1994) 1293. [5] I. Kim, J.S. Chun, W.J. Lee, Mater. Chem. Phys. 44 (1996) 288. [6] S. Shinriki, Y. Nishioka, Y. Ohji, K. Mukai, IEEE Trans. Electron.
Devices 36 (1989) 328.
[7] M.B. Lee, H.D. Lee, B.L. Park, U.I. Chung, Y.B. Koh, M.Y. Lee, IEDM Tech. Dig. (1996) 683.
[8] N. Awaya, H. Inokawa, E. Yamamoto, et al., IEEE Trans. Electron Devices 43 (1996) 1206.
[9] A.R. Sethuraman, J.F. Wang, L.M. Cook, Semicond. Int. 6 (1996) 177.
[10] S. Ezhilvalavan, T.Y. Tseng, J. Appl. Phys. 83 (1998) 4797. [11] S. Ezhilvalavan, T.Y. Tseng, J. Am. Ceram. Soc. 82 (1999) 600. [12] C.P. Smyth, Dielectric: Behaviour and Structure, McGraw-Hill, New
York, 1955.
[13] G.Q. Lo, D.L. Kwong, P.C. Fazan, V.K. Mathews, N. Sandler, Elec-tron Devices Lett. 14 (1993) 216.
[14] S. Zaima, T. Furuta, Y. Koide, Y. Yasuda, J. Electrochem. Soc. 137 (1990) 2876.
[15] T. Kato, T. Ito, J. Electrochem. Soc. 135 (1988) 2586.
[16] S. Banerjee, B. Shen, I. Shen, J. Bohlman, G. Brown, R. Doering, J. Appl. Phys. 65 (1989) 1140.
[17] P.L. Young, J. Appl. Phys. 47 (1976) 235. [18] C.A. Mead, Phys. Rev. 128 (1962) 2088.
[19] J. O'Dwyer, Theory of Electrical Conduction and Breakdown in Solid Dielectrics, Clarendon, Oxford, 1973.
[20] D.R. Lide, CRC Handbook of Chemistry and Physics, CRC Press, Boston, MA, 1991.
[21] C.S. Hwang, S.O. Park, H.J. Cho, C.S. Kang, H.K. Kang, S.I. Lee, M.Y. Lee, Appl. Phys. Lett. 67 (1995) 2819.
