Roughening kinetics of chemical vapor deposited copper films on Si ( 100 )

The roughening kinetics of copper films synthesized by low pressure chemical vapor deposition ~LPCVD! on Si~100! substrates was investigated by scanning tunneling microscopy ~STM!. By applying the dynamic scaling theory to the STM images, a steady growth roughness exponent a50.8160.05 and a dynamic growth roughness exponent b50.6260.09 were determined. The value ofa is consistent with growth model predictions incorporating surface diffusion. The value of b, while higher than expected from these models, can be related to LPCVD processing conditions favoring growth instabilities. ©1996 American Institute of Physics. @S0003-6951 ~96!01409-8#

͑Received 25 September 1995; accepted for publication 19 December 1995͒ The roughening kinetics of copper films synthesized by low pressure chemical vapor deposition ͑LPCVD͒ on Si͑100͒ substrates was investigated by scanning tunneling microscopy ͑STM͒.By applying the dynamic scaling theory to the STM images, a steady growth roughness exponent ␣ϭ0.81Ϯ0.05 and a dynamic growth roughness exponent ␤ϭ0.62Ϯ0.09were determined.The value of ␣ is consistent with growth model predictions incorporating surface diffusion.The value of ␤, while higher than expected from these models, can be related to LPCVD processing conditions favoring growth instabilities.© 1996 American Institute of Physics.͓S0003-6951͑96͒01409-8͔ Aluminum films have commonly been used in the metallization of integrated circuit devices.However, as devices shrink further into the quarter-micron regime, an aluminumbased metallization would be unable to meet the low electrical resistivity and stringent electromigration resistance requirements.Copper is a suitable alternative choice because of its low bulk electrical resistivity ͑1.67 versus 2.65 ⍀ cm for Al͒ and its superior electromigration resistance which would allow higher current densities to be sustained on smaller linewidths as well as minimize the RC delay time of the circuit.Compared to ''line of sight'' deposition techniques ͑such as sputtering and evaporation͒, low-pressure chemical vapor deposition ͑LPCVD͒ is an attractive processing method for achieving either selective or nonselective copper with excellent step coverage.Recent studies have focused on the use of ͑trimethylvinylsilyl͒ hexafluoroacetylacetonato copper I ͓͑hfac͒ Cu͑I͒ TMSV͔ as a precursor for the deposition of LPCVD Cu films at relatively low temperatures and high growth rates.Although studies on the growth kinetics and properties of such films have been reported, 1 little is known about the evolution of the interface and its effects on subsequent film growth processes.
This paper focuses on the scanning tunneling microscopy ͑STM͒ imaging of Cu films grown by LPCVD from ͑hfac͒ Cu͑I͒ TMVS on Si͑100͒ wafers under nonequilibrium conditions and over a broad range of film thicknesses.The dynamic scaling 2 analysis applied to STM images 3 allows us to evaluate the dynamic and static roughness exponents, and to derive from them the contribution of different processes to the growth mechanism of LPCVD Cu films under nonequilibrium conditions.
The Cu films were synthesized in a SPECTRUM model 211, cold wall, single wafer, cassette-to-cassette, fully automated CVD reactor.Single crystal ͑100͒ silicon wafers, 10 cm in diameter, were taken directly from sealed containers and introduced with no further cleaning treatment into the reaction chamber.The root-mean-square, rms, surface rough-ness of the virgin wafers resulted to be 0.14 nm as measured by atomic force microscopy.The wafers were exposed on the back side ͑unpolished side͒ to a dual zone radiant heater.The temperature of the wafer was measured using a thermocouple.
The ͑hfac͒ Cu͑I͒ TMVS liquid precursor used for the deposition of the Cu films was directly injected into the reaction chamber.Two vacuum pump stations were used for achieving the desirable pressures within the reaction chamber and for controlling the wafer transport system.The Cu films in this study were synthesized at a pressure of 100 mTorr, at a deposition temperature of 453 K, and average growth rate of 0.5 nm s Ϫ1 .The values for the average Cu film thicknesses, ͗h͘, were determined through profilometry to be in the range of 44 -2220 nm.
The topography of Cu films was imaged with a Nanoscope III STM operating in air.STM measurements were made using a 0.25 V bias voltage with the tip ͑ϩ͒ at 0.5 nA constant current.Different Pt-Ir tips were used to discard occasional artifacts caused by the tip geometry itself.To minimize errors STM images with 512ϫ512 pixels were taken. 3STM data were analyzed after fitting the instrument plane and applying a subtracting procedure. 4he rms roughness ͑interface width͒ for length L and average thickness ͗h͘, for a system containing N sites with single valued heights h i , is defined by the expression .

͑1͒
The dynamic scaling theory 2 predicts that when ͗h͘→0, scales as
a͒ Electronic mail: Jvara@emduam11.bitnetAtomistic-type growth models such as those of Eden, 5 ballistic deposition, 5 and restricted solid-on solid, 5 have been proposed to explain nonequilibrium growth processes.These models can be successfully described by the Kardar, Parisi, and Zhang motion equation 6 which leads to ␣ϭ0.40 and ␤ϭ0.25 in 3D growth.Alternatively, continuous models in 3D incorporating surface diffusion yield either ␣ϭ1.0 and ␤ϭ0.25 7,8 or ␣ϭ0.66 and ␤ϭ0.24. 9 Cu film growth on Si͑100͒ by the ͑hfac͒ Cu͑I͒ TMVS procedure involves a progressive roughening for increasing values of ͗h͘ due to the competitive growth of Cu grains ͑Fig.1͒.From Fig. 1͑a͒ it is important to note that the Cu film seems to start to grow quite homogeneously.This homogeneity allows one to disregard effects due to improper cleanliness of the Si͑100͒ substrate on the initial Cu film growth.It is known 2 that as ͗h͘ increases the average grain size d s increases according to d s ϰ͗h͘ ␤/␣ .From the plot ͓Fig.2͑a͔͒ of average film grain size ͑obtained from histograms of the grain size distributions͒ versus ͗h͘ a value for ␤/␣ϭ0.74 is obtained.Also, for all values of ͗h͘, the ex- perimental value of the interface width, stm , is seen to initially increase with S, the STM scan length, finally reaching a constant value stm Accurate ␣ values were obtained by using the single image dynamic scaling method. 3Accordingly, ␣ was derived from the equation stm ͑ L ͒ϰL s ␣ , ͑4͒ where L s corresponds to a segment of the STM scan of size S measured in the x direction ͑i.e., the fast scanning direc-tion͒ by plotting the log stm vs log L s in the stationary roughness regime.For Cu deposits with ͗h͘ϭ44 nm and ͗h͘ϭ2220 nm ͓Fig.3͑a͔͒ the log stm vs log L s plots are seen to exhibit a linear region exceeding three decades in log L s , and a saturation region for L s values larger than the critical value L sc which depends on ͗h͘.The value of L sc at the crossover point corresponds closely to the value of d s .
From the slope of the straight lines depicted in Fig. 3, a value of ␣ϭ0.82 was determined.After averaging more than 300 different STM images, a value of ͗␣͘ϭ0.81Ϯ0.05 was obtained.On the other hand, the exponent ␤ was evaluated from Eq. ͑2͒ ͑i.e., the variation of stm c vs ͗h͘ data͒.The log stm c vs log͗h͘ plot ͓Fig.3͑b͔͒ resulting from 15 STM images for each value of ͗h͘ yielded a straight line with ͗␤͘ϭ0.62Ϯ0.09.These values yielded a ratio of ␤/␣ϭ0.77, in agreement with the value derived from the log d s vs log͗h͘ plot ͓Fig.2͑a͔͒.
The ␣ value obtained for LPCVD Cu films, 0.81, is in agreement with other values ͑0.9-0.7͒, reported for other experimental systems.They are consistent with the ␣ values derived from growth models incorporating surface diffusion.8][9] Hence, in principle, surface diffusion appears to dominate roughness development for LPCVD Cu films grown at Tϭ453 K.The preceding explanation is also consistent with recent data on the self-affine characteristics of vapor deposited Au grown in the range 298 -673 K where surface diffusion was also found to dominate the growth process. 10It should be noted that both Au and Cu exhibit a large surface mobility even at room temperature. 11evertheless, the value ␤ϭ0.62 is considerably higher than those derived from the growth models incorporating surface diffusion which predict either ␤ϭ0.25 7,8 or ␤ϭ0.20. 9However, similar results have been reported for plasma polymer films, 12 where the roughness exponents resulted in ␣ϭ0.9 and ␤ϭ0.7, and for Si 0.4 Ge 0.6 thin films grown by molecular beam epitaxy ͑MBE͒ on Si, 13 where an experimental ␤ value of 0.71 was interpreted as a result of instabilities in MBE growth.Also it has been recently suggested 13 that the strain due to lattice mismatch in the heteroepitaxial films could be another source of growth instability.
For Cu films deposited by MBE on Cu͑100͒ at 200 K, 14 ␣Ϸ1 and ␤ϭ0.56 were obtained.In this case the high ␤ value, close to that measured in this work, was explained as a consequence of the asymmetry in the vertical and horizontal mobility of adatoms ͑Schwo ¨bel barriers͒ which tend to increase the value of ␤.For Cu electrodeposition the values ␣ϭ0.9 and ␤ϭ0.45 were obtained.This high ␤ value was assigned to the influence of the electric field 15 on the growth process.In general, experimental values of ␤ larger than those derived from growth models including local effects were observed when nonlocal effects arising from Laplacian fields were present during the interface growth.In LPCVD Cu film formation, it is possible that the pressure field created around the rough growing interface, presumably by TMVS produced in the rate determining step, 1 results in the development of growth instabilities leading to a high value of ␤.
This work was partially supported by the CSIC-CONICET cooperation program and by a grant from the NASA New Jersey Space Grant Consortium.

c
͓Fig.2͑b͔͒.The value of stm c also increases with ͗h͘.