Anda di halaman 1dari 6

Home

Search

Collections

Journals

About

Contact us

My IOPscience

Dielectric breakdown in AlOx tunnelling barriers

This article has been downloaded from IOPscience. Please scroll down to see the full text article. 2011 J. Phys. D: Appl. Phys. 44 135403 (http://iopscience.iop.org/0022-3727/44/13/135403) View the table of contents for this issue, or go to the journal homepage for more

Download details: IP Address: 200.130.19.175 The article was downloaded on 04/07/2012 at 20:45

Please note that terms and conditions apply.

IOP PUBLISHING J. Phys. D: Appl. Phys. 44 (2011) 135403 (5pp)

JOURNAL OF PHYSICS D: APPLIED PHYSICS doi:10.1088/0022-3727/44/13/135403

Dielectric breakdown in AlOx tunnelling barriers


D M Schaefer1,2 , P F P Fichtner2 , M Carara1 , L F Schelp1 and L S Dorneles1
Universidade Federal de Santa Maria, Departamento de F sica, Av. Roraima, 1000, Santa Maria 97105-900, RS, Brazil 2 Universidade Federal do Rio Grande do Sul, Instituto de F sica, Av. Bento Gonc alves, 9500, Caixa Postal 15051, Porto Alegre 91501-970, RS, Brazil E-mail: lsdorneles@gmail.com
1

Received 22 September 2010, in nal form 24 January 2011 Published 16 March 2011 Online at stacks.iop.org/JPhysD/44/135403 Abstract We studied the dielectric breakdown in tunnelling barriers produced by plasma-assisted oxidation of an aluminium surface. The barrier mean height, thickness and the effective tunnelling area were extracted from current versus voltage curves measured at room temperature. The effective tunnelling area ranged from 1010 to 105 cm2 , corresponding to less than 1% of the geometrical surface of the samples. The estimated electrical eld to breakdown agreed with predictions from thermochemical models, and decreased exponentially with the effective tunnelling area.

1. Introduction
The destruction of thin insulators by dielectric breakdown (BD) has been a key issue in microelectronics for decades, where its statistical occurrence determines the lifetime of metaloxide semiconductor components. As a result, several relations between the chemical quality/morphology of barriers and the breakdown mechanisms were established, specially for SiO2 layers [15]. More recently, the interest in developing non-volatile random access memories has attracted considerable attention to the study of magnetic tunnel junctions. These devices consist of two magnetic electrodes connected by a very thin tunnelling barrier, usually amorphous Al2 O3 or crystalline MgO [616]. Although presenting quite similar breakdown phenomena, gate capacitors and tunnel junctions have some intrinsic differences that make unclear, up to now, to what extent the knowledge accumulated in the rst system can be applied to the second one. In several nanometre thick SiO2 layers, the leak currents are relatively small and just in part due to hole/electron tunnelling. In contrast, in magnetic junctions a high tunnelling probability is mandatory to allow large magnetoresistance and to maintain the device resistance inside practical limits. As a consequence, the insulating barrier should not be more than a few nanometres thick, meaning electrical elds up to 1 GV m1 with 2 V applied to the electrodes. As shown by scanning
0022-3727/11/135403+05$33.00

tunnelling microscopy and transport measurements in Al2 O3 [1722] and more recently in MgO [23, 24], uctuations in the insulator thickness concentrate the tunnelling current in hot spots. In these small portions of the insulating barrier, the high current densities certainly alter parameters relevant for the breakdown, e.g. charge to breakdown and local temperature. Another parameter sensible to inhomogeneous current distributions is the stress variable, either the time or the electrical eld to breakdown. As long as the density of intrinsic defects (due to thickness uctuations, barrier structural defects, impurities, etc) is constant over the junction surface and extrinsic defects (such as dust inclusions) are absent, the stress variable is expected to decrease as the junction area increases [2527]. But when the current is not homogeneously distributed over the insulating barrier surface, we have a less predictable situation and the stress variable can, for example, scale with the junction perimeter [27]. As will be shown here for a group of samples with the same geometrical area, but presenting hot spots, the electrical eld to breakdown scales with the effective tunnelling area.

2. Experimental details
Al/AlOx /Al tunnel junctions were deposited by magnetron sputtering with a typical base pressure of 107 mbar (or lower) using masks to dene 200 m wide electrodes in the crossed stripe geometry, producing samples with a 4 104 cm2
1
2011 IOP Publishing Ltd Printed in the UK & the USA

J. Phys. D: Appl. Phys. 44 (2011) 135403

D M Schaefer et al

Figure 1. I versus V curves, measured at room temperature in the current drive mode (symbols), and calculated (line). Inset shows the measurement up to higher currents, past the breakdown voltage.

Figure 2. Weibull plot for a group of Al/AlOx /Al samples subject to the same stress.

junction area. This system has symmetric interfaces, and no over/under oxidation of the insulating barrier is possible. Typically 14 out of 16 samples survive the deposition process, and more than half survive the connection to the measurement system. In a rst group of samples the insulating barrier was created by plasma-assisted oxidation of the freshly deposited bottom Al electrode, in a 100 mTorr O2 atmosphere with time of plasma discharge (Tox ) ranging from 10 to 50 s. In a second group, the insulating barrier was created by exposing the Al bottom electrode to the same oxygen atmosphere for 12 h, but with no discharge (Tox = 0). I versusV curves were measured at room temperature, and for each sample Simmons equation [28, 29] was tted to the experimental data in a voltage window below 0.8 V, following the modications introduced in [22]. In the non-linear least square tting algorithm, the barrier mean height ( ), the barrier thickness (tAlOx ) and also the effective tunnelling area (Aeff ) have been left as free parameters. The iteration procedure was made until a predetermined chi-square is not further reduced. Local minima were dismissed by testing different sets of initial values. The breakdown process was followed using current ramps and as the breakdown voltage (VBD ) criterion, the rst drop in voltage was considered.

Figure 3. Breakdown voltage VBD as a function of stress ramp speed dV /dt . Oxidation times (Tox ) are indicated.

3. Results and discussion


Figure 1 presents a typical I versus V curve obtained with a constant dI /dt . A linear dependence in the low voltage region is followed by an exponential growth of I with V , as expected from tunnelling models [2831]. As exemplied in the inset of gure 1, as VBD is reached there is a drastic reduction in the voltage, followed by a linear I versus V behaviour. For most of the studied samples the breakdown was a one-step process, characterized by a sudden drop in the voltage needed to maintain the set current. 2

Figure 2 presents the Weibull plot for a group of samples prepared with Tox > 0, where F is the fraction of samples that failed in V < VBD . It follows reasonably well a straight line, indicating that intrinsic defects drive the breakdown process. It is expected that VBD increases with the stress speed [32, 33], which is observed in gure 3. But, this behaviour is not observed in the samples prepared with Tox = 0, suggesting that the chemical quality or the density of defects is signicantly altered when natural oxidation is used. It is important to note the values observed for samples prepared with Tox > 0. They range from 1.4 to 1.9 V, and are not consistent if the geometrical area of the junctions is relevant to the breakdown description. From extrapolations of the reported behaviour of VBD versus area [2527] obtained from devices with much smaller areas, the studied samples should have VBD values smaller than 0.5 V, even if we admit differences in density of defects due to sample production particularities. An open possibility is to admit that the insulating barrier is not at and thickness uctuations are

J. Phys. D: Appl. Phys. 44 (2011) 135403

D M Schaefer et al

Figure 4. RAeff versus tAlOx for the studied samples. Oxidation times (Tox ) are indicated.

present, as is the case for our samples [22, 34] and also for samples from other groups [35] and other materials, such as MgO [23, 24, 36] or MgB2 native oxide [37]. In this case electronic transport would be restricted to a smaller area of the sample, as the tunnelling current would be concentrated in regions less thick. To include this effective area in the analysis, we consider the area as an additional free parameter while tting the calculated I versus V curves to the experimental data. As discussed below, we have independent evidence that this approach is justied and gives meaningful gures to the barrier parameters. First, if we consider that the current ows through the whole geometrical junction area, there is no clear exponential increase in the junctions resistance and the extracted tAlOx , which is true not only here but also in most of the works found in the literature. Exceptions are found for samples with extremely high structural quality and/or small lateral size. If we consider the effective tunnelling area Aeff obtained from the tting process, a clear exponential dependence of R versus Aeff is observed, as shown in gure 4 for more than 20 samples studied, prepared under different conditions. The samples prepared using Tox = 0 (without plasma-assisted oxidation) are in the lower thickness range as should be expected. The values of for all the samples show a trend to increase with tAlOx (gure not shown) and lie within the range from 1.1 to 1.6 eV, smaller than the reported values for massive Al2 O3 probably due to imperfections in the oxide layer. The values of Aeff are within the range from 1010 to 105 cm2 , around 1% of the geometrical area, due to thickness uctuations of the insulating barrier [1, 5, 1622]. These values are consistent with those obtained by direct measurement, where simultaneous topographic and tunnelling current images were obtained from an Al2 O3 insulating barrier [19, 38, 39]. As will be discussed below we were able to test this procedure in samples where one of the barrier parameters, the barrier height, was known. Second, if the breakdown occurs at voltages higher than the barrier height, a plot of the logarithmic derivative of conductivity versus V will present a maximum at a voltage of about 1.2 times the barrier height [40, 41]. In other words, it is 3

Figure 5. I versus V curve measured at room temperature (symbols). Lines are ttings of Simmons equation to the data using the parameters presented in table 1. Solid black line is from routine (a), dashed red line is from routine (b) and dotted blue line is from routine (c), as explained in the text. Inset: the arrow indicates the maximum in the logarithmic derivative of conductivity versus V curve. (This gure is in colour only in the electronic version) Table 1. Parameters obtained by tting Simmons equation to the experimental data from gure 5. (a), (b) and (c) are the parameters used to generate the solid, dashed and dotted lines in the gure, respectively. Fitting routine (a) (b) (c) Aeff (107 cm2 ) 31 5 0.3 4000 (xed) tAlOx () 18 2 18.3 0.1 25.6 0.9 (eV) 1.4 0.2 1.485 (xed) 1.43 0.06

possible to obtain one of the barrier parameters independently by direct measurement. This plot, for a representative sample, is shown in the inset of gure 5. In order to compare the results from our tting procedure (three tting parameters) with the results from the traditional procedures (two tting parameters), gure 5 also shows a measurement of I versus V extracted from this sample (open circles), and three curves obtained tting Simmons equation to the experimental data using three different routines (in a voltage window below 1.5 V). In routine (a) (solid line) tting was performed with , tAlOx and Aeff as free parameters; in routine (b) (dashed line) tAlOx and Aeff were free parameters, and the measured value for was used; in routine (c) (dotted line) and tAlOx were free parameters. As can be seen in the gure, both solid and dashed lines t very well the measured curve, as the dotted line hardly matches it. In table 1 the values obtained for the barrier parameters extracted from one representative sample, using the three different tting routines, are given. The values for tAlOx obtained from routine (b) is much smaller than that from routine (c). This result is also observed in other works where ttings to STM I versus V curves, where the probe size is known with nanometre resolution, are compared with ttings

J. Phys. D: Appl. Phys. 44 (2011) 135403

D M Schaefer et al

proper tting of Simmons equation to it, one can estimate the breakdown eld or the time to breakdown.

4. Conclusions
We have analysed the breakdown process in tunnelling barriers where, due to thickness uctuations in the insulator, the current concentrates in hot spots. The barrier parameters, particularly the barrier thickness and the tunnelling effective area, were obtained for each sample from ttings of Simmons equation to the measured I versus V curves. For samples prepared with the same geometrical area, the expected behaviour of the breakdown eld with the area is only observed if the effective tunnelling area and thickness are taken into account.

Figure 6. Breakdown eld EBD as a function of the effective area Aeff . The vertical dashed line indicates the geometrical area (4 104 cm2 ). Oxidation times (Tox ) are indicated.

Acknowledgments
This work was partially supported by the Brazilian agencies, CNPq (Conselho Nacional de Desenvolvimento Cient co o de Aperfeic e Tecnol ogico), CAPES (Coodenac a oamento o de de Pessoal de N vel Superior) and FAPERGS (Fundac a ` Pesquisa do Estado do Rio Grande do Sul). Amparo a

to curves extracted from the tunnelling junction with the same oxide barrier [19]. The values for tAlOx , and Aeff (for this particular sample Aeff is less than 1% of the geometrical area) obtained from routine (a) are in very good agreement with the values obtained from routine (b). This was observed for all the samples where it was possible to measure directly. At the stress speed used BD occurs for VBD above , in the FowlerNordheim regime [4] and in contrast with some of the reported results [33]. The transport measurements do not allow us to observe the exact geometrical location of the breakdown; however, as the time to breakdown is shorter when the current density is larger [3], one can infer that the BD occurs within the area determined by Aeff . Considering that the breakdown occurs within Aeff where the thickness is tAlOx , in a rst approximation the breakdown eld can be dened as EBD = VBD /tAlOx . A very clear behaviour of EBD versus Aeff can be seen in gure 6, where the values obtained from all samples (the tting routine is used to obtain ts of the same quality as the one presented in gure 5(a )) are plotted. This plot includes samples with Aeff within a range of ve orders of magnitude from 1010 to 105 cm2 , indicating that EBD behaves as indicated by the solid line independent of Tox or stress speed. This can be explained by the Poisson model for randomly distributed defects (see, e.g., [42]), and we can estimate the defect density in the barrier by the slope of the EBD versus ln(Aeff ) plot. Using the data presented in gure 6 we obtain a defect density = 2.5 107 cm2 . This value is very close to that estimated by other authors such as Stathis [43] and DiMaria [44] for SiO2 barriers within the same thickness range. This percolation model applied to these SiO2 barriers states that insulating barriers in the very small thickness range (<3 nm) need a minor but almost constant 0.5 107 cm2 to break down. For the samples with a smaller effective area, EBD approaches 1.3 GV m1 , a value that is larger than those usually reported in the literature. But this value is closer to the prediction based in the thermochemical model for crystalline Al2 O3 [45, 46]. Another important result is the fact that by the measurement of an I versus V curve, associated with the 4

References
[1] Lee J C, Chen I-C and Hu C 1988 IEEE Trans. Electron Devices 35 2268 [2] Suehle J S 2002 IEEE Trans. Electron Devices 49 958 [3] Su ne J, Placena I, Barniol N, Farr es E, Mart n F and Aymerich X 1990 Thin Solid Films 185 347 [4] Degraeve R, Kaczer B and Groeseneken G 1999 Microeletron. Reliab. 39 1445 [5] Wu E Y, Nowak E J, Vollertsen R-P and Han L-K 2000 IEEE Trans. Electron Devices 47 2301 [6] Engel B N, Rizzo N D, Janesky J, Slaughter J M, Dave R, DeHerrera M, Durlam M and Tehrani S 2002 IEEE Trans. Nanotechnol. 1 32 [7] Butler W H, Zhang X-G and Schulthess T C 2001 Phys. Rev. B 63 054416 [8] Mathon J and Umerski A 2001 Phys. Rev. B 63 220403(R) [9] Parkin S P P, Kaiser C, Panchula A, Rice P M, Hughes B, Samant M and Yang S-H 2004 Nature Mater. 3 862 [10] Yuasa S, Nagahama T, Fukushima A, Suzuki Y and Ando K 2004 Nature Mater. 3 868 [11] Guerrero R, Herranz D, Aliev F G, Greullet F, Tiusan C, Hehn M and Montaigne F 2007 Appl. Phys. Lett. 91 132504 [12] Wang S G, Ward R C C, Du G X, Han X F, Wang C and Kohn A 2008 Phys. Rev. B 78 180411(R) [13] Ikeda S, Hayakawa J, Ashizawa Y, Lee Y M, Miura K, Hasegawa H, Tsunoda M, Matsukura F and Ohno H 2008 Appl. Phys. Lett. 93 082508 [14] Ma Q L, Wang S G, Zhang J, Wang Y, Ward R C C, Wang C, Kohn A, Zhang X-G and Han X F 2009 Appl. Phys. Lett. 95 052506 [15] Du G X, Wang S G, Ma Q L, Wang Y, Ward R C C, Zhang X-G, Wang C, Kohn A and Han X F 2010 Phys. Rev. B 81 064438 [16] Khan A A, Schmalhorst J, Thomas A, Schebaum O and Reiss G 2008 J. Appl. Phys. 103 123705 [17] Niu C, Magtoto N P and Kelber J A 2001 J. Vac. Sci. Technol. A 19 1947 [18] Ando Y, Kameda J, Kubota H and Miyazaki T 2000 J. Appl. Phys. 87 5206

J. Phys. D: Appl. Phys. 44 (2011) 135403

D M Schaefer et al

[19] da Costa V, Tiusan C, Dimopoulos T and Ounadjela K 2000 Phys. Rev. Lett. 85 876 [20] Olbrich A, Ebersberger B, Boit C, Vancea J, Hoffmann H, Altmann H, Gieres G and Wecker J 2001 Appl. Phys. Lett. 78 2934 [21] da Costa V, Romeo M and Bardou F 2003 J. Magn. Magn. Mater. 258259 90 [22] Dorneles L S, Schaefer D M, Carara M and Schelp L F 2003 Appl. Phys. Lett. 82 2832 [23] Thomas A, Drewello V, Sch afers M, Weddermann A, Reiss G, Eilers G, M unzenberg M, Thiel K and Seibt M 2008 Appl. Phys. Lett. 93 152508 [24] Yoshida C, Kurasawa M, Lee Y M, Tsunoda K, Aoki M and Sugiyama Y 2009 IEEE CFP09RPS-CDR 47th Annual Int. Reliability Physics Symp. (Montreal) p 139 [25] Girgis E, Boeve H, de Boeck J, Schelten J, Rottl ander P, Kohlstedt H and Gr unberg P 2000 J. Magn. Magn. Mater. 222 133 [26] Das J, Degraeve R, Roussel P, Groeseneken G, Borghs G and de Boeck J 2002 J. Appl. Phys. 91 7712 [27] Nakajima K, Asao Y and Saito Y 2003 J. Appl. Phys. 93 9316 [28] Simmons J G 1963 J. Appl. Phys. 34 1793 [29] Simmons J G 1964 J. Appl. Phys. 35 2655 [30] Stratton R 1962 J. Phys. Chem. Solids 23 1177 [31] Brinkman W F, Dynes R C and Rowell J M 1970 J. Appl. Phys. 41 1915

[32] Oepts W, Verhagen H J, Coehoorn R and de Jonge W J M 1999 J. Appl. Phys. 86 3863 [33] Oliver B, Tuttle G, He Q, Tang X and Nowak J 2004 J. Appl. Phys. 95 1315 [34] Cruz de Gracia E S, Dorneles L S, Schelp L F, Teixeira S R and Baibich M N 2007 Phys. Rev. B 76 214426 [35] Buchanan J D R, Hase T P A, Tanner B K, Hughes N D and Hicken R J 2002 Appl. Phys. Lett. 81 751 [36] Miller C W, Li Z-P, Schuller I K, Dave R W, Slaughter J M and kerman J 2006 Phys. Rev. B 74 212404 [37] Singh R K, Gandikota R, Kim J, Newman N and Rowell J M 2006 Appl. Phys. Lett. 89 042512 [38] Da Costa V, Bardou F, B eal C, Henry Y, Bucher J P and Ounadjela K 1998 J. Appl. Phys. 83 6703 [39] Da Costa V, Henry Y, Bardou F, Romeo M and Ounadjela K 2000 Eur. Phys. J. B 13 297 [40] Halbritter J 1985 Surf. Sci. 159 509 [41] Rottl ander P, Hehn M and Schuhl A 2002 Phys. Rev. B 65 054422 [42] Wu E Y and Vollerstsen R-P 2002 IEEE Trans. Electron Devices 49 2131 [43] Stathis J H 1999 J. Appl. Phys. 86 5757 [44] DiMaria D J 1997 Microelectron. Eng. 36 317 [45] MccPherson J W and Mogul H C 1998 J. Appl. Phys. 84 1513 [46] MccPherson J W, Kim J-Y, Shanware A and Mogul H C 2003 Appl. Phys. Lett. 82 2121

Anda mungkin juga menyukai