Diamond and Related Materials, 3 (1994) 951-956
Correlation between breakdown voltage and structural properties of polycrystalline and heteroepitaxial CVD diamond films R. Hessmer, M. Schreck, S. Geier and B. Stritzker Universitdt Augsburg, Lehrstuhl fiir Experimentalphysik I V, Memminger Str. 6, D-86135 Augsburg (German),)
Abstract The dielectric strength of both non-oriented polycrystalline(grain size less than 100 nm) and highly textured diamond films (grain size about 2 ~tm) which have been epitaxially nucleated on silicon(100) has been investigated by the ramping method at room temperature. We show that annealing in air or etching in CrO3 + HzSO. which drastically increases the resistivityof the samples in comparison with the as-grown state, has no significant effecton the breakdown voltage. Besides this, the dependence of the dielectric strength on the film thickness has been studied. We observe a pronounced reduction of the dielectric strength from 4 × l06 V cm ~to about 1 x 10 6 V cm 1for film thicknesses ranging from 150 nm to 2 p.m respectively.
1. Introduction Because of its very high thermal conductivity and dielectric strength, diamond is a promising material for heat sinks with a high breakdown voltage for highpower, high-voltage electronics. For low-voltage highpower devices, e.g. laser diodes, heat sinks (or more accurately, heat spreaders)made of natural IIa diamond have already been used in industry for some years [ 1]. For high voltage devices like insulated-gate bipolartransistors (IGBT), heat sinks have to meet the additional requirement of a high dielectric strength. It has been shown [2,3] that polycrystalline chemical vapor deposited (CVD) diamond films can be grown with thermal conductivities up to 20-24 W cm -1 K 1 close to the values of type IIa natural diamond. Thus polycrystalline films are already an attractive alternative to single-crystalline heat spreaders for low-voltage applications. In contrast, only limited data about the dielectric strength of diamond films exist which allow the estimation of the potential of these films as passive dielectric layers. A few values for the dielectric strength have been deduced from the breakdown voltages of diamond diodes in the reverse direction [4,5]. Szmidt et al. [ 6 ] investigated the dielectric strength of thin (up to 150nm) passivating diamond-like films. Joshi et al.  studied the high-field conduction in polycrystalline CVD diamond films. After the deposition process, CVD diamond films display relatively low values of resistivity, of the order of 106 ~ cm in comparison with 1016 if2 cm reported for natural diamond. In order to achieve higher resistivities, as necessary for electrically insulating heat sinks, the films have to be annealed or chemically etched in CrO3 + H2SO4 [8 10].
Because of the technological importance, systematic studies are needed on the correlation between the dielectric strength and the preparation procedure used after deposition or the thickness of the films. The above-cited investigations refer to relatively thin films (thickness, 1 ~tm or less) with correspondingly low breakdown voltages. These results cannot be extrapolated for thicker films, as it is known from extended studies on many different dielectric materials, e.g. SiO 2, A1203, that the breakdown field decreases with increasing film thickness . In this paper we present our investigations on the dielectric strength of two types of diamond films. The first set of samples consisted of non-oriented polycrystalline films with high nucleation density and small grains (less than 100 nm), which are henceforth simply referred to as "polycrystalline". For the second set of samples, a heteroepitaxial nucleation step [ 12] was followed by a textured growth step [ 13] resulting in a high proportion of oriented grains. In what follows we refer to these samples as "heteroepitaxial". The films have been grown on ( 100)-oriented silicon. The influence of c o m m o n postdeposition treatments on the dielectric strength as well as the dependence on the film thickness have been studied.
2. Sample preparation and experimental details The samples were grown on (001).-oriented borondoped silicon wafers with a resistivity of 0.01-0.02 D cm in a commercial microwave plasma CVD reactor from ASTeX TM. The reactor was slightly modified to allow for in situ biasing of the substrates at - 2 0 0 V against the
f5 1994 - Elsevier Sequoia. All rights reserved SSDI 0925-9635t 93 )05052-E
R. Hessmer et al. / Breakdown voltage and structural properties of diamond films
plasma, as necessary for the nucleation step [13,14]. Table 1 summarizes the growth parameters for both polycrystalline and heteroepitaxial samples. For nonoriented polycrystalline growth the substrates were nucleated for 30 min at a bias voltage of - 2 0 0 V and a methane content of 2% in hydrogen in order to get a very high nucleation density (more than 101° c m - 2 ) as determined by atomic force microscopy. The nucleation step was followed by a standard growth process with 0.6°/0 C H 4 q- 0.6o/o C O 2 in hydrogen. Because of the high nucleation density, the growth resulted in very homogeneous and smooth diamond films, even for film thicknesses as low as 150 nm. In order to achieve heteroepitaxial films we used a nucleation time of 15 min at a methane content of 0.3%, resulting in a nucleation density of about 5 x 107 c m -2. A two-step process as indicated in Table 1 is necessary to grow the heteroepitaxial samples. A scanning electron micrograph of a typical film is shown in Fig. 1(a). Figure 2 displays typical Raman spectra of both polycrystalline and heteroepitaxial samples. The high nucleation density for the polycrystalline samples results in diamond films with small grains and consequently a large number of grain boundaries. This leads to a broadening of the diamond Raman line at about 1330 cm -1 and a high signal from non-diamond carbon phases (broad peak centered at about 1500cm -t) as compared to the Raman signal of the high-quality heteroepitaxial diamond films. After the deposition process, four different kinds of sample preparation were used to investigate their influence on the breakdown behaviour: a. annealing in a vacuum at 800 °C for 4 h, b. etching for 30 min in CrO3 + H2SO4 at 200 °C, c. annealing in air at 500 °C for 15 min, d. s t e p c f o l l o w e d b y
in h y d r o g e n
plasma at 400 °C for 2 h.
~ ~ " " " ' ' " ~
i ~ ~
Fig. 1. Scanning electron micrographs of a "heteroepitaxial" sample 10.7 gm thick. (a) Undisturbed surface with well-aligned diamond crystals. (b) Surface after breakdown has occurred: a crater is formed
and moltensiliconcoversthe diamondsurface.
TABLE 1. Process parameters for the growth of "polycrystalline" and "heteroepitaxial" diamond films "Polycrystalline.
Time (h) Microwave power (W) Pressure (mbar) Temperature (°C) Total gas flow (sccm) Hz (%) CH 4 (%)
CO 2 (%) Bias voltage (V)
First growth step
Second growth step
0.5 600 20 870 500 98 2 --200
20 60 900 20 740 500 98.8 0.6 0.6 --
0.25 600 20 870 500 99.7 0.3 --200
12 900 20 700 500 99.5 0.5 ---
20-60 1300 45 780 200 90 8.7 1.3 --
R. Hessmer et al. / Breakdown voltage and structural properties of diamond,films
%-. .-e_ r-
samples under measurements.
3. Results and discussion
z 7< :~ < rv
1000 1100 1200 1300 1400 1500 1600 1700 1800
RAMAN SHIFT ( 1 / c m ) Fig. 2. Typical Raman spectra of the investigated "polycrystalline" and "heteroepitaxial" films. The spectra were taken with the 514 nm argon
laser line. In the last case the sample was cooled down to 100 °C in the hydrogen plasma in order to avoid the removal of the hydrogen from the diamond film. The plasma was formed in the deposition system at a microwave power of 600 W and 20 mbar hydrogen pressure. For reference, a part of each sample was left in the as-grown state, We measured the film thickness at the edge of the cracked samples using a surface texture profiler. This works well for film thicknesses above 1 pm. Since for thinner diamond films the cracked edges did not show a step between diamond film and silicon substrate, we used laser reflective interferometry  and scanning electron microscopy to determine the film thickness. The former method was calibrated on thicker samples by comparison with measurements obtained by the surface texture profiler, Nine electrical contacts to the diamond side of each sample were formed by evaporating titanium circular dots of 1 mm diameter and 200 nm thickness through contact shadow masks. The contacts were protected by gold overlayers 20 nm thick. These nine electrical contacts allowed for some statistics on the breakdown voltage of each sample. Some of the contacts were equipped with a guard ring. In order to get an ohmic contact on the silicon side of the samples, we etched off the S i O 2 with hydrofluoric acid and then deposited a large-area contact with silver paint, The electrical breakdown of the samples was characterized using a computerized current-voltage (1-1/) measurement set-up by ramping a stabilized power supply at a rate of about 2 V s-1 up to 2500 V. The dark current was measured with a Keithley model 195A digital multimeter with a resolution of 1 nA. The measurements were conducted at room temperature. In order to avoid surface conductivity due to humidity, we kept the
3.1. Influence of post-deposition treatments In the voltage range far below the breakdown voltage, all samples show ohmic behaviour. The resistance strongly depends on the preparation procedure which has been performed after the deposition process, as is known from the literature [8,9]. In the following we discuss the influence of these post-deposition procedures by means of the example of a heteroepitaxial diamond film with a thickness of 29 gm. Etching in CrO3 + H2NO 4 or annealing in air increases the resistance from about 10T fl for the as-grown state to more than 10 it ~ . After hydrogenation in a microwave plasma the resistance decreased to about 106 f~. Annealing the samples in a vacuum did not result in any significant change of the resistance. Figure 3(a) shows the I Vcharacteristic for the sample first annealed in air and then hydrogenated. The curve was obtained by using a guard ring which is turned on and off during the measurement. The "on" state means that the top titanium contact and the guard ring are at the same potential. In the "off" state the potential of the guard ring floats. When the guard ring is switched off, the current increases by about three orders of magnitude, indicating that almost the whole current is conducted at the surface of the diamond film. Similar results have been obtained by Geis et al. . Figure 3(b) displays the 1-V characteristic for another contact on the same sample. No guard ring was used for this measurement. The solid line represents the first measurement at this contact. Especially from the linear plot (inset of Fig. 3(b)) it can be seen that an ohmic region (about 106 f~) is followed by a region with sharp spikes. At about 1900 V the current abruptly decreases by almost three orders of magnitude. Some spikes and the abrupt current decrease were accompanied by sparks at the top electrode. A second measurement at the same contact yielded the dashed curve. Compared with the first measurement, the current is drastically lowered. We attribute this effect to self-annealing of the sample surface during the measurement due to the ohmic heating by the surface current. Similar behaviour could be observed for all samples with low resistivity (as-grown, or annealed in a vacuum, or first annealed in air and then hydrogenated) and film thicknesses greater than about 3 lam. Samples with film thicknesses below 3 pm did not show self-annealing. We think that the breakdown voltage is too low to allow for large enough surface currents. However, besides this, the dielectric strength of all
R. Hessmer et al. / Breakdown voltage and structural properties of diamond films
, , ,,,,,
Z Iz, I FY t'Y
D (_) 10 -8
10 . 9
(b) 1o ~
'on' I 1500
i I llil
0 L,J J L,.J
Fig. 4. Dielectric strength as a function of the film thickness for "polycrystalline" and "heteroepitaxial" samples.
second measurement ,... . . . . . . (.f~A ~ " / 1
Fig. 3. I V characteristics of a "heteroepitaxial" sample first annealed in air at 500 °C for 15 min and then hydrogenated in microwave plasma: the film thickness is 29 ~rn. (a) l-Vcurve obtained by using a guard ring: the connection between power supply and guard ring was turned on and off during the measurement. (b) Two successive measurements at the same contact point of the sample, the solid line representing the first measurement: a linear plot of this measurement
is shown in the inset,
samples under investigation did not depend on the postdeposition treatment of the films. This is quite reasonable since the post-deposition treatments only influence the surface conductivity, whereas the breakdown depends on the bulk properties,
3.2. Dependence of the dielectric strength on thefilm thickness In Fig. 4 the dielectric strength of 12 samples is shown as a function of the film thickness. Open and filled circles represent polycrystalline and heteroepitaxial samples, respectively. The value of the dielectric strength was obtained by measuring the breakdown voltage at more than four different points for each sample and then averaging the three highest values. Thus we reduced the influence of pinholes and other weak links on the extracted breakdown values and obtained a reasonable
data set for a comparison of differently prepared samples. For each sample the scatter of the three values is clearly below 10%. For the polycrystalline samples we chose a nucleation density more than 101° c m -2, corresponding to grain sizes of about 100 nm. Thus the diamond films did not show holes for film thicknesses down to 150 nm. A clear reduction of the dielectric strength from about 4 X 10 6 V cm -1 to about 1 x 10 6 V cm -1 with increasing film thickness can be seen from the figure. The dielectric strengths of the heteroepitaxial samples seem to be slightly lower than those of the polycrystalline o n e s in spite of the significantly higher film quality w h i c h c a n be d e d u c e d f r o m t h e R a m a n s p e c t r a in Fig. 2. W h e r e a s in t h e p o l y c r y s t a l l i n e s a m p l e s the b r e a k d o w n
nearly always occurs beneath the titanium top electrode, t h e s i t u a t i o n for t h e h e t e r o e p i t a x i a l s a m p l e s is q u i t e
different. Here the electrical breakdown point very often is several hundred micrometres away from the contact dot. This indicates that the breakdown is dominated by weak links shifting the overall dielectric strength to lower values. Consistently the scatter of the values for the electric breakdown field for different points of the same sample is higher for the heteroepitaxial samples than for the more homogeneous polycrystalline films. We speculate that these weak links are correlated with deep hollows between the crystallites at the surface which can clearly be seen in Fig. 1. Future work is necessary to clarify whether these hollows can be closed by growing thicker films, thus improving the dielectric strength. A typical picture of the point where the breakdown has occurred is shown in Fig. l(b) for a heteroepitaxial diamond film of 10.7 I~m thickness. The micrograph does not show the titanium dot which has been contacted during the measurement. It is located about 50 Ixm to the left of the breakdown crater. At the breakdown point
R. Hessmer et al. / Breakdown voltage and structural properties ofdiamond films
the diamond film is removed down to the silicon wafer and a crater of about 60 ~tm diameter is formed, Molten and resolidified silicon, as verified by energy dispersive X-ray analysis, covers the film surface at the edge of the crater, The above-mentioned weak links give the impression that the polycrystalline samples investigated are better candidates for heat sinks for high-voltage applications. However, first one has to keep in mind that, owing to the large number of grain boundaries accompanied by non-diamond phases, the thermal conductivity of these samples should be much lower than that of the heteroepitaxial films (studies are in progress). Second, the resistivity of the polycrystalline films is always lower than that of the heteroepitaxial ones. Figure 5 shows the I V characteristics normalized to the breakdown voltage for two samples of comparable film thickness, both etched in CrO3 + H 2 S O 4 t o remove the surface conductive layer. Throughout the plotted voltage region, the current through the polycrystalline sample is higher by about three orders of magnitude. This difference is still present after annealing both samples in air. Moreover we want to point out that both samples show the same exponential current-voltage dependence in spite of the drastically 1 0- 1 I 1 0 -2
"10 - 3 <
1 0 -4
different film quality. Since samples with top electrodes of aluminium, forming Schottky instead of ohmic contacts, show the same behaviour we can exclude the possibility that this effect results from the specific contact material.
4. Conclusion We have investigated the dielectric strength of polycrystalline and heteroepitaxial diamond films with grain sizes of less than 100nm and about 2 btm. Different treatments have been used after the deposition. By employing guard-ring experiments it could be shown that these treatments mainly influence the surface conductivity. Samples with low surface resistivity (as-grown or hydrogenated) show a self-annealing effect during the l-Vmeasurements owing to the ohmic losses generated by the current. This leads to sharp structures in the I V curve and abrupt current decreases. None of the investigated post-deposition treatments showed a significant effect on the breakdown voltages of the films. The dielectric strength of the investigated samples shows a pronounced dependence on the film thickness. For polycrystalline fihns grown with identical deposition conditions, the dielectric strength decreases from a maximum value of 4.0 x 10 6 V cm 1 for film thicknesses of 155 nm and 300 nm to 1.5 x 10~' V cm 1 for film thicknesses between 2 gm and 7 gin. The values for the dielectric strenglh of the heteroepitaxial samples are slightly lower than the values for the non-oriented films. We attribute this to weak links which may reduce the dielectric strength. Both the non-oriented and the heteroepitaxial samples show the same exponential current--,,oltage dependence in the high-field range below breakdown. This is surprising since both the film quality as determined by Raman spectroscopy and the growth mode are quite different. Thus we conclude that, in spite of the drastically different values for the conductivity, the same conduction mechanisms are valid in the high-field range.
I' 1 0-7
1 0 -8 0.7
I 1 .0
u/u B re o kd o w n
Fig. 5. Semilogarithmic plots of the I-Vcharacteristics of "polycrystalline" and "heteroepita×ial" samples, both etched in CrO 3 + n z s o 4 to remove the conductive layer at the surface. The current is plotted against the voltage normalized to the breakdown voltage.
This work has been supported in part by the Deutsche Forschungsgemeinschaft (key programme "Synthesis of super-hard materials") and by the Bayerisches Staatsministerium far Unterricht, Kultus, Wissenschaft und Kunst (Bavarian rapid programme "New materials").
References 1 M. Seal, Interdisc. Sci. Rev., I4 (1989) 64. 2 J.E. Graebner, S. Jin, G.W. Kammlott, Y.-H. Wong, J.A. Herb and C.F. Gardinier, Diamond Relat, Ma~er., 2 (1993) 1059.
R. Hessmer et al. / Breakdown voltage and structural properties of diamond films
3 S. Jin, L.H. Chen, J.E. Graebner, M. McCormack and M.E. Reiss, Appl. Phys. Lett., 63 (1993) 622. 4 M.I. Landstrass, M.A. Piano, M.A. Moreno, S. McWilliams, L.S. Pan, D.R. Kania and S. Han, Diamond Relat. Mater., 2 (1993) 1033. 5 D.G. Jeng, H.S. Tuan, R.F. Salat and G.J. Fricano, J. Appl. Phys., 68 (1990)5902. 6 J. Szmidt, T. Brozek, R.H. Gantz and A. Olszyna, Surf. Coat. Technol., 47 (1991) 496. 7 R.P. Joshi, M.K. Kennedy, K.H. Schoenbach and W.W. Hofer, J. Appl. Phys., 72 (1992) 4781. 8 H. Nakahata, T. Imai and N. Fujimori, Proc. 2nd Int. Syrup. on Diamond Materials, 1991, The Electrochemical Society, Pennington, NJ, 1991, p. 487, and references cited therein. 9 T. Sugino, Y. Muto, J. Shirafuji and K. Kobashi, Diamond Relat. Mater., 2 (1993) 797, and references cited therein.
10 Y. Mori, Y. Show, M. Deguchi, H. Yagi, H. Yagyu, N. Eimori, T. Okada, A. Hatta, K. Nishimura, N. Kitabatake, T. Ito, T. Hirao, T. Izumi, T. Sasaki and A. Hiraki, Jpn. J. Appl. Phys., 32 (1993) L987, and references cited therein. 11 J.J. O'Dwyer, The Theory of Electrical Conduction and Breakdown in Solid Dielectrics, Oxford University Press, London, 1973. 12 X. Jiang and C.-P. Klages, Diamond Relat. Mater., 2 (1993) 1112. 13 B.R. Stoner, S.R. Sahaida, J.P. Bade, P. Southworth and P.J. Ellis, J. Mater. Res., 8 (1993) 1334. 14 S. Yugo, T. Kanai, T. Kimura and T. Muto, Appl. Phys. Lett., 58 (1991) 1036. 15 C.-H. Wu, W.H. Weber, T.J. Potter and M.A.Tanor, J. Appl. Phys., 73 (1993) 2977. 16 M.W. Geis, N.N. Efremow and J.A. von Windheim, Appl. Phys. Lett., 63 (1993) 952.