Electrical properties of a-SiO/sub x/N/sub y/:H films prepared by microwave PECVD

299

ELECTRICAL PROPERTIES OF a-SiOxNy:HF I U S PREPARED BY MICROWAVE PECVD

P. Rabiller, J.E. Klemberg-Sapieha,M.R. Wertheimer and A. Yelon

INTRODUCTION Thin dielectric layers are of importance in Si and GaAs based microelectronics applications, in macroelectronic applications such as amorphous silicon-based flat panel displays, in thin film capacitor technology, and in several other fields. For this reason, plasma enhanced chemical vapor deposition (PECVD) of silicon nitride (P-SiN) and of silicon oxide (P-SiO,) has been quite extensively studied. However, PECVD of silicon oxynitride (P-SiON) has so far received less attention. Since P-SiN and P-SiO, both have certain shortcomings, P-SiON might be more useful than either if it can combine high dielectric constant, low loss, low permeability to ions, and low internal stress. Promising results have been published on oxynitride produced by low pressure CVD [1,2]. We have recently reported microwave PECVD of P-SiON films, their characterization by various physico-chemical techniques [3], and also some preliminary electrical property determinations [ 4 ] . Microwave PECVD offers important advantages over other fabrication methods: First, the deposition temperature is relatively low compared to those which are necessary for CVD and other thermal growth methods. Second, because of the high density of very energetic electrons in the microwave plasma, it yields high deposition rates (300 to 1300 A/min) compared to those which can be obtained with lower frequency plasmas [5]. Different compositions are readily obtained by varying flow rates of silane, ammonia and nitrous oxide [3]. The objective of the present paper is to report further on electrical properties of our P-SiON films (ranging from pure nitride to oxide). These include static and high frequency permittivities, and d.c. conductivity, all of which vary systematically with film composition. EXPERIMENTAL The large volume microwave (2.45 GHz) plasma apparatus used for sample preparation has been described elsewhere [ 5 , 6 ] . In the present deposition experiments the microwave power was kept constant at 100 W and substrate temperature, Ts, was held at 280'C, while the total pressure was fixed at 0.1 Torr. The respective partial flow rates of SIH,, NH, and N,O ranged from 15 to 25, 0 to 50, and 0 to 50 sccm. Total flow rate was maintained close to 80 sccm. For electrical characterisation, dielectric layers were deposited on aluminum coated glass substrates, and A1 electrodes were evaporated after deposition of the sample. Thicknesses of P-SiON layers were between 300 and 1000 nm, while the top and bottom aluminum electrodes were about 100 nm thick. For dielectric measurements, the sample area (- 6 0 mm') was chosen so as to yield a capacitance of a few nF. For measurements of I-V characteristics, this area was further reduced (to 1 mu') to minimize the incidence of breakdown events. Chemical composition was determined using X-ray photoelectron spectroscopy (ESCA), and nuclear elastic recoil detection analysis (ERDA) [ 7 ] . The latter method is particularly important, as it yields hydrogen content; hydrogen is believed to be of great importance, as it degrades the material's chemical Groupe des Couches Minces (GCM) and Department of Engineering Physics, Ecole Polytechnique, C.P. 6 0 7 9 , Succursale "A", Montreal, Qc H3C 3A7, Canada

300

stability, and thereby its dielectric and other physical properties [E]. Good agreement between ESCA and ERDA results was obtained for samples made in the same deposition. Samples were also examined by FTIR spectroscopy. Complex permittivity measurements were performed at room temperature over a frequency range from 10' to lo5 Hz using a computer interfaced HP 4274A LCR-meter. Measurement accuracy is better than 1%, both for the real part of IE* and for the loss tangent, tan 6 , in nearly all cases. The high frequency dielectric constant n: (- nz) was obtained from the refractive index, n , 628 nm. Finally. I-V characteristics were measured by ellipsometry at X determined using a computer interfaced Keithley 230 voltage source and a Keithley 619 electrometer. Sample temperature T could be varied between 40 K and 450 K using a cryostat and temperature controller.

-

RESULTS AND DISCUSSION Samule Comuosition and Moruholocy From ESCA and ERDA, we found that we are able to reach any desired intermediate composition between stoechiometric "silicon nitride" (P-SiN) and "silicon oxide" (P-SiO,). depending upon the reagent gas flows used. "Non-equilibrium" P-SiON materials are, in fact, very stable, and their compositions are conveniently characterized by the parameter [ O / ( O + N)], where 0 and N are the atomic percentages of oxygen and nitrogzn in the film, respectively. We have found from ERDA that hydrogen content, always quite low in microwave PECVD materials [3], decreases from about 15% in P-SiN to about 1% in P-SiO,. All physical properties we have investigated, of course including electrical properties, depend systematically upon [ O / ( O + N)], as will be shown in the following sections. P-SiON samples have been examined by scanning electron microscopy (SFN), and have been found to possess morphologies ranging from "smooth" to "rough", depending on composition and fabrication conditions. "Rough" samples have many submicron-sized particles included in their structure. These protrude through the film surface, giving rise to the rough appearance. Such particles constitute unwanted defects ("weak points") in the films; they tend to occur more frequently in N-rich films, while 0-rich P-SiON tends to be "smooth". Dielectric Permittivitv Figure 1 shows the real part IE' of the complex permittivity, n * ( w ) = n ' ( w ) as a function of [ O / ( O + N)]. The circles represent the "static" value I E ~ , measured at 1 kHz; the term "static" is not inappropriate here, since no appreciable variation in IE' was observed for these samples over the frequency range (102 to 105 Hz) investigated. The square symbols represent the high frequency (or optical) permittivity I E ; , determined from Maxwell's relation, fc: = n2. The fact that IE; # U: over the entire range of compositions is symptomatic of these materials' strong dispersion at infrared frequences. We have measured infrared spectra in the frequency range 400 cm-l to 4000 cm-', as have other authors [9,10],and we find Si-N and Si-0 bond absorptions, but no evidence of mixed 0-Si-Nbonding. This has been interpreted [9,10] as evidence for a "two phase" model, according to which P-SiON consists of two separate phases, Si0,:H and SiN,.,:H, rather than being a single, homogeneous alloy with random bonding. Such a model is also supported by the permittivity data: The solid curves in Fig. 1 represent the Bruggeman mixture formula [ll] jc"(w),

with

where subscripts 0 and N refer to SiO, and Si,N,, respectively, V are their volume fractions in the oxynitride, and Os'> is the measured effective permit7.0 and tivity of the medium. We have used the accepted literature values U: 3 . 8 , and IC:4.0 and 2.0 for Si3N4 and SiOz, respectively. Both sets of measured U: and U; data are seen to fit the solid curves quite well; also, U: measurements of the present PECVD samples agree well with Remmerie's [l] data for LPCVD materials (dashed curve).

-

-

DC Conduction All samples were exposed to self healing breakdown pretreatment [12], in order to remove structural defects before I-V characteristics were measured. This has very little effect on smooth samples, but may reduce the area of rough samples by as much as 40%. We have previously reported [4] that the room temperature conductivity varies smoothly from more that S cm'l for pure nitride to less than S cm-l for pure oxide. Osenbach et a1 [13] have reported values ranging from lo-' to S cm-' for PECVD oxynitride deposited at 380 IcHz, depending upon both N,O flow and substrate temperature during deposition. They attribute this variation to the Si-H content of their samples. This is not likely to explain the smaller range of variation of our samples, as they contain relatively little total hydrogen [3], and very little of this is in the form of Si-H [14]. It should be noted, however, that in both cases it is the nitride which contains the most hydrogen, and it is also the most conductive material. In large bandgap, highly insulating materials the mechanisms of conductivity are strongly dependent upon details of structure, and thus of preparation conditions. P-SIN films, produced using the present microwave deposition system, but with substrates between 150' and 200'C. have been found 1151 to exhibit polaron conduction. The samples described here, prepared at Ts 280'C. show evidence of trap controlled conduction, at compositions between P-SIN and oxynitride with 50 atomic% oxygen. At higher oxygen concentrations we were unable to apply a wide enough electric field range to determine mechanisms since the samples were too resistive.

-

At low field, I-V plots are ohmic, and U (T) is activated, with activation energies which vary with composition as shown in Table I. In order to measure this behaviour, it is necessary to vary the voltage very slowly, as the current takes a long time to stabilize. A typical room temperature response can be fit [14] with a sum of exponentials, with time constants ranging between 0.8 and 300 seconds. Such a spread can, in turn, be explained if we assume a distribution of traps with depths ranging from 0.4 to 0.8 eV, namely the range of activation energies given in Table I. Further evidence for this hypothesis is provided by the field dependence of current density. The behaviour at three different P-SiON compositions is shown in a Poole-Frenkel (P-F) plot in Fig. 2, where it is clear that current is proportional to E ', above 0.5 MV cm-l. This is consistent with either P-F conduction or with Schottky injection. The latter mechanism should give rise to modification in U and in high field behaviour with change in electrode material; this is not the case [14]. However, when we fit the observations to the P-F

-

302

model, we obtain values of relative permittivity as shown in Table 11. These are far higher than the measured static values found from Fig. 1, and listed in the same table. A reasonable explanation of our results can be provided by Simmons' modification [16] of the PF model, which assumes that the Fermi level is determined by charged donor states and neutral traps which lie below and above this level, respectively. This model yields values, shown in Table 11, which lie between the corresponding K; and IC;values, a reasonable result. We conclude, as do Osenbach and coworkers [13,17], that conduction in P-SiON occurs via field-enhanced thermal excitation of electrons from traps (P-F effect). Unlike those authors, however, we cannot presently assert that conductivity is controlled by the Si-H content in the alloy. ACKNOWLEDGEMENTS The authors are grateful to S. Blain and S.C. Gujrathi for their contributions to this project. This work has been supported in part by the Natural Sciences and Engineering Research Council of Canada, and by the fonds "Formation de Chercheurs et Aide B la Recherche" of Quebec. REFERENCES Kuiper, A.E.T., Koo, S.W., Habraken, F.H.P.M., and Tamminga, Y., 1983, J. Vac. Sci. Technol., B l , 62-66. Remmerie, J . , Maes, H.E., De Keersmaeker, R., Habraken, F.H.P.M., OudeElferink, J . , and van der Weg, W., 1987, "Silicon Nitride and Silicon Dioxide Thin Insulating Films", The Electrochemical Society, Pennington, N.J., PV87-18, pp. 50-56. Blain, and Guirathi. S.C.. . S.. . Klembera-Sauieha. J . E . . Wertheimer. M.R.. 1989, Can. 3. Phvs., in press. Rabiller, P., Blain, S . , Klemberg-Sapieha, J.E., Wertheimer, M.R., and Yelon, A., IEEE Doc. CH 2594-0-88DE1,pp. 149.152. Wertheimer, M.R., Moisan, M., Klemberg-Sapieha,J . E . , and Claude, R., 1988, Pure and Auul. Chem., 60, 815-820. Wertheimer, M.R., Klemberg-Sapieha,J . E . , and Schreiber, H.P., 1984, Thin Solid Films, 115,109-124. Groleau, R., Gujrathi, S.C., and Martin, J.P., 1983, Nucl. Instrum. Methods Phvs. Res., 218, 11-18. Chang, C.P., Flamm, D.L., Ibbotson, D.E., and Mucha, J.A., 1987, J . Auul. 62, 1406-1415. Claassen, W.A.P., v.d.Po1, H.A.J. Th., Goernans, A.H., and Kuiper, A.E.T., 1986, 3 . Electrochem. Soc., 133, 1458-1464. Cros, Y., Jousse, D., Liu, J . , and Rostaing. R.C., 1987, J . Non-Crystal. Solids, 9 0 , 287-293. Bruggeman, D.A.G., 1935, Ann. Phvs. Leiuzig, 24, 636-642. Ramu, T.S., and Wertheimer, M.R., 1986, IEEE Trans. Electr. Insul. U, 557-563. Osenbach, J.W., Knolle, W.R., and Elia, A., 1988, "Plasma Processing", The Electrochemical Society, Pennington, N.J., PV88-22, pp. 352-361. Rabiller, P., 1988, MScA thesis, Ecole Polytechnique de Montreal. Okoniewski, A.M., Tannous, C., and Yelon, A., 1987, Phys. Rev. B, 3 5 , 64546457. Simmons, J.G., 1967, Phvs. Rev., 155,657-660. Osenbach, J . W . , and Knolle, W.R., 1986, J. Auul. Phvs., 60, 1408-1416. I

m,

.

.

303

- Activation enereies for dc conduction. from Dlots of In o ( T ) vs 1 / T

TABLE I

Composition, [O/(O + N)]

0.11

0.31

0.48

0.81

Activation energy, Ea (eV)

0.32

0.41

0.57

0.62

fi’

Composition [O/(O+N) 1

P-F model

0.11 0.31 0.48

values. and their sources

Simmons model [16]

17.1 17.6 16.8 1

I

1

I

I

measured

4.3 4.4 4.2 I

1

I

1

1

6.8 5.8 5.3

fi:

measured

fi:

3.1 3.0 2.8

1

Figure 1. Plot of permittivity IC‘ versus sample composition The round and [O/(O+N)]. square symbols represent K: and f i ; , respectively. The solid and dashed curves represent the Bruggeman mixture formula, and Remmerie‘s measurements for LPCVD samples, respectively.

K’

2I

I

I

0.0

I

0.2

I

I

0.4

I

I

I

I

0.8

I_

1.0

0.8

0 1 (O+N)

a

Y

-2’1

Figure 2. Plots of In J versus for samples having three different compositions, [O/(O+N)]: (*): 0.11; ( * ) : 0.31; ( + ) : 0.48. E’’’

-22

1

1

200

1

1

400

1

1

1

600

Evz (V/cm)vz

1

800

1

1

1000

1