Ionized physical vapor deposition of integrated circuit interconnects*

PHYSICS OF PLASMAS VOLUME 5, NUMBER 5 MAY 1998Ionized physical vapor deposition of integrated circuit interconnects*J. Hopwood †,a)Northeastern University, Boston, Massachusetts 02115Received 19 November 1997; accepted 25 November 1997Interconnects, once the technological backwater of integrated circuit technology, now dominateintegrated circuit cost and performance. As much as 90 percent of the signal delay time in futureintegrated circuit designs will be due to the interconnection of semiconductor devices while theremaining 10 percent is due to transistor-related delay. This shifts the thrust of critical researchtoward an improved understanding of interconnect science and technology. Shrinking circuitgeometries will require high aspect ratio AR vias to interconnect adjacent metal layers. By the year2007 it is predicted that logic circuits will use 6 to 7 interconnected metal layers with via ARs of5.2:1. Memory will need fewer layers, but ARs as high as 9:1. In this paper, the demands ofinterconnect technology will be reviewed and the opportunities for plasma-based deposition of viaswill be discussed. One promising new method of fabricating high-aspect ratio vias is ionizedphysical vapor deposition I-PVD. The technique economically creates a unidirectional flux ofmetal which is uniform over 200–300 mm diameter wafers. Since metal ejected by conventionalsputtering is primarily neutral and exhibits a cosine angular velocity distribution, sputtered metalatoms do not reach the bottom of high AR vias. By sputtering these atoms into a moderate pressure4Pa, high-density Ar plasma, however, the metal atoms are first thermalized and then ionized. Theions are then readily collimated by the plasma sheath and directionally deposited into narrow, deepvia structures. Experiments have consistently shown that over 80% of the metal species are ionizedusing I-PVD. The physical mechanisms responsible for ionization will be discussed from both anexperimental and modeling perspective and the spatial variation of metal ionization isexperimentally determined. © 1998 American Institute of Physics. S1070-664X9890405-0I. BACKGROUND: INTEGRATED CIRCUITINTERCONNECT TECHNOLOGYIn the microelectronics revolution much attention hasbeen focused on the rapidly increasing number of devices perchip and the shrinking critical dimensions of these electroniccomponents. The explosion in the number of transistors fabricatedon a single integrated circuit IC, however, hasplaced extreme demands on electrically interconnectingthese devices in the manner necessary to perform the logicaloperations of a modern microprocessor. Figure 1 shows asimplified, partial cross section of a typical IC. A singlemetal-oxide-semiconductor field effect transistor MOSFETis shown in the figure with gate G, source S, and drainD connections labeled. Interconnection of this MOSFETwith other devices on the chip not shown is accomplishedthrough the polysilicon/metal alloy silicide gate level andseveral metal-SiO 2 interlayer dielectric ILD levels joinedtogether by vertical vias. Past generations of microprocessorsused three metal levels as shown in Fig. 1, but designs usingmore transistors demand additional levels of metal. The NationalTechnology Roadmap for Semiconductors 1 forecaststhat 7 to 8 levels of interconnect will be required by 2010.With such a large number of layers it is clear that the fabricationof interconnects will dominate the process of manufacturingfuture ICs.*Paper pThpT-1 Bull. Am. Phys. Soc. 42, 2026 1997.† Tutorial speaker.a Electronic mail: jhopwood@lynx.neu.eduThe electrical performance of an IC is most easilygauged by the maximum clock speed at which the chip operatesreliably. The clock speed is determined by howquickly the transistors can be switched on and off as well ashow quickly the signal propagates from one device to thenext. As the interconnect scheme becomes more complex,the signal delay due to longer, thinner interconnect linesdominates the total switching time and, therefore, the maximumclock rate. In the most simple sense, the interconnectdelay is determined by the product of the interconnect resistanceR and the parasitic capacitance C of the interconnectmetal to the adjacent layers. 2 The thickness of the ILDmust remain constant in future ICs such that the interconnectcapacitance does not become any larger. At the same time,however, the widths of the metal lines and vias will shrink toaccommodate the greater density of transistors. The result isthat the aspect ratio AR of interconnect vias may increaseto approximately 9:1 as the lateral critical dimensions decreaseto 0.11 m during the coming decade.The fabrication of the interconnect structure uses the‘‘damascene’’ method. First a blanket layer of ILD is depositedand vias are anisotropically plasma etched. Titanium isdeposited into the via. The titanium acts as an adhesionlayer, but also improves the electrical contact resistance byconsuming residual oxides from the surface of the underlyinglayer. It is important to deposit material all the way to thebottom of the structure during this step. The success of thedeposition is measured by the bottom coverage which is definedas the ratio of Ti film thickness at the bottom of the via1070-664X/98/5(5)/1624/8/$15.001624© 1998 American Institute of Physics

Phys. Plasmas, Vol. 5, No. 5, May 1998J. Hopwood1625FIG. 1. A partial cross section of a simplified integrated circuit. The expandedview shows a via that is lined with Ti and TiN and subsequentlyfilled with W or Al. Future materials are indicated in parentheses.to the thickness deposited on the top surface of the ILD.Next the via receives a barrier layer of TiN that protects theILD from corrosive effects of WF 6 during chemical vapordeposition of the tungsten via plug. Alternately, if aluminumis used as the plug material, the TiN acts as a wetting layerthat allows the sputtered aluminum to be reflowed 3 into thevia at elevated temperatures. In both cases, the TiN acts as adiffusion barrier between the plug and the ILD. The diffusionbarrier must be conformal to the via and it should be asthin as possible so that very little cross sectional area is occupiedby the somewhat resistive barrier material. Finally thelow-resistivity via plug material is deposited. This material isprogressing from W to Al, and eventually Cu. The overburdenof metal present on the ILD after deposition of the plugis removed by chemical-mechanical polishing. In the singledamascene process, these step are repeated for the depositionof the metal lines above the vias. A future alternative is thedual damascene process 4 in which the via and metal levelsare etched simultaneously. Then both the via and the metalline are filled with conducting materials and planarized byCMP in the final step. The dual damascene process reducesthe number of processing steps needed to fabricate a metallevel, but the requirements on the deposition process are considerablymore stringent.Aluminum-0.5% Cu metallization is projected to be replacedby pure copper. 5 Copper has a lower bulk electricalresistivity than AlCu 1.7 vs 3.2 -cm so that the interconnectdelay time RC may be decreased. Equally importantis copper’s ability to conduct higher current densitiesthan AlCu without failure of the thin lines due to electromigrationof the metal atoms. Since copper does not form aself-limiting oxide, copper interconnects will need to becompletely enveloped in a passivating diffusion barrier, mostlikely consisting of Ta and TaN layers.Although many technological challenges exist in the areasof lithography, low-permittivity dielectric materials andvia etching, this paper will focus on the directional depositionof barrier and plug materials. There are three potentialmethods of interconnect deposition: Physical vapor depositionPVD, chemical vapor deposition CVD, and electroplating.The CVD methods use elevated temperatures orplasma 6 to dissociate precursor species. Filling or lining ofhigh aspect ratio features is accomplished by fragmented precursorspecies with low sticking coefficients that are capableof reaching and depositing at the bottom of deep, narrowopenings. While highly conformal films can be depositedusing CVD, the cost and toxicity of the precursor materialsremains an issue. Plating is also an effective way of fillinghigh AR features. The plating baths, however, can be difficultto maintain, present a liquid waste hazard, and representa departure from the semiconductor industry’s preference fordry processing. PVD, and in particular sputtering, has longbeen the workhorse of the industry. The fundamental problemwith sputtering, however, is the sputtered species areejected from the solid target with a cos n () angulardistribution 7,8 which limits bottom coverage to approximately0.2 in structures with 2:1 ARs. Physical collimation 9of the sputtered flux by a honeycomb-shaped filter betweenthe target and wafer improves film conformality by trappingsputtered species with large angle trajectories. Bottomcoverage 10 of 2:1 features increases to 0.3–0.4, but the collimatorsignificantly reduces the deposition rate and may createparticles that reduce the yield of functional ICs.II. IONIZED PHYSICAL VAPOR DEPOSITION „I-PVD…I-PVD is a recent advance in PVD technology thatachieves directional deposition of metals by ionizing thesputtered or evaporated metal atoms and collimating theseions with the plasma sheath adjacent to the wafer as shownin Fig. 2. A high electron density (n e 10 11 cm 3 ), inert gasplasma between the target and the wafer is needed to ionizethe metal vapor. Strong ionization of the metal occurs sincethe electron temperature depends primarily on the ionizationpotential of the inert gas 15.7 eV for argon which is muchgreater than that of metals 6.0 and 7.7 eV for Al and Cu.The plasma source is commonly an electron cyclotron resonanceplasma 11–13 or inductively coupled plasma. 14–17The average distance that a sputtered neutral will travelbefore being ionized dictates the design of an I-PVD system.Simple analysis of a sputtered neutral traversing a high densityplasma gives the ionization mean free path 18,19 as iz s /K i n e where s is the velocity of the sputtered neutraland K i is the ionization rate constant see below. Atomssputtered from the target exhibit a Thomson distribution 20where the most probable energy is one half the surface bindingenergy (1.5 eV for Al. Therefore, iz 60 cm in anAr plasma where T e 3 eV and n e 510 11 cm 3 . Thisanalysis suggests two methods of generating a highly ionized,directional metal flux. First, if the inert gas pressure isquite low, the target-to-wafer distance must be quite long.

Phys. Plasmas, Vol. 5, No. 5, May 1998J. Hopwood1629FIG. 5. Measurement of the aluminum ion and neutral density along the axisof an I-PVD reactor demonstrates that high ion fractions are achievedthrough he decay of the neutral metal density and sustained volume generationof metal ions. Solid curves are diffusion models described in the text.The target diameter for this experiment was 75 mm.FIG. 6. The radial variation of Al and Al suggests diffusive transport ofboth species. Solid curves are diffusion models described in the text.fusive losses of metal to the chamber walls, and 2 approximatelyuniform volume generation of metal ions extendingdown to the region near the wafer.Using the same measurement technique, the radial distributionsof Al and Al adjacent to the surface of a 200 mmwafer were determined and are plotted in Fig. 6. Notice thatalthough the bulk ionization fraction Al /AlAl isonly 0.3, the ionized flux fraction is about 0.8 as shown inFig. 3. The aluminum ion density is peaked at the center ofthe wafer and decreases at larger radii. This is a typicaldiffusion-dominated distribution of ions as discussed below.To aid in the understanding of ionization in an I-PVDreactor, a simple model is proposed. Since the sputteredmetal atoms are thermalized in the first few centimeters nearthe target, transport of metal atoms to the wafer is diffusiverather than ballistic. From this observation it is reasonable toattempt a solution to the diffusion equation for metal atoms 2 n M (r,z)0 in a cylinder R, L with the boundary conditionthat M 0 at the chamber walls. The sputteringsource of M is modeled with a boundary adjacent to thetarget that is the sum of k disks, each of constant Al densityand concentric about the central axis at z 0,kn M r,z0N 0 A i 1urb i ,i118where u is the unit step function and the constants b i and A iare chosen to match the erosion profile of the target. Thediffusion solution isn M r,z2N 0i1kb iA J1 x 0 j b i /RiR j0 x 0 j J 2 1 x 0 j 0 J x 0 jrR expk zzexpk z z2L1exp2k z L , 19where J n is the n th -order Bessel function of the first kind,k z x 0 j /R, and J 0 (x 0 j )0. The modeled aluminum neutraldensity Eq. 19 is plotted along with the experimentalmeasurements in Figs. 5 and 6. Only the magnitude of thedensity N 0 ) was scaled to fit the data. The agreement of thediffusion model for the Al distribution is within the accuracyof the experiment. The right hand side of Fig. 7 shows acontour plot of the neutral density Al superimposed on asketch of an I-PVD reactor with R22 cm and L15 cm.It has previously been shown that sputtered atoms mustbe thermalized if ionization is to occur within the confines ofthe I-PVD reactor. Since the metal ions are thermalized, it isreasonable to assume that the ion density distribution will bediffusive as well. The Klyarfeld parabolic approximation 25has been shown to predict the distribution of ions in inductivelycoupled plasmas. 45 By combining the Klyarfeld solutionfor a long tube n T (r) with the solution for an infiniteslab geometry n S (z), an estimate of the ion density within acylinder is n M (r,z)n T (r)n S (z) wheren T rn io11h R n S zn io11h l zL/2Rr 2, 202L/2 ,21h l 0.86,223L/2 iand h R is given in Eq. 6. Figures 5 and 6 show that thisapproximate solution for aluminum ion density agrees wellwith respect to z but predicts a metal ion density that is moreuniform along the radial direction than experimentally observed.The left hand side of Fig. 7 shows a contour plot ofthe Klyarfeld parabolic approximation within a typical Al I-PVD reactor at 10 mTorr. Finally, although the uniformityof the deposition rate over 200 mm wafers was typically7%–13%, the fraction of ionized flux is much more uniformsince the diffusion dominated metal and metal ion densitiesdecrease radially at similar rates. There are several unifor-

1630 Phys. Plasmas, Vol. 5, No. 5, May 1998 J. HopwoodFIG. 7. Contour plots of aluminum neutral and ion density in a 44 cm diameter I-PVD reactor.mity issues yet to be resolved, but bottom coverage which isproportional to ionization fraction should be quite consistentacross large wafers.V. CONCLUSIONIonized physical vapor deposition is a viable techniquefor the fabrication of high aspect ratio interconnects for futuregenerations of integrated circuits. The advantages of I-PVD over the alternative methods are numerous: The metalsource is a nontoxic low-cost solid, the process does notproduce chemical wastes, sputtering is a very clean process,and sputtering has a well-established technology base in theIC fabrication industry. The most promising applications forI-PVD are for the deposition of liners and barriers—the refillingof high aspect ratio vias with metal is an area whereadditional research is required.Ionization of sputtered species is most effective whenthe ionization potential of the inert gas is much greater thanthe metal. Metals with low ion mobility will also exhibithigher fractional ionization. Reactor geometry is shown toinfluence ionization. These factors include designing the reactorwith sufficient target-to-wafer spacing so that thermalizationof sputtered atoms occurs within the high densityplasma. Uniformity of the deposition rate and bottom coveragedepend on the distribution of ions and the fraction ofionized flux. The ion flux fraction is inherently uniform sincediffusive transport applies to both metal neutrals and ions.Further research and development are needed to control theuniformity of ions at the wafer. Finally, while the body ofknowledge for directional metal deposition using inert gasI-PVD is maturing, the physics, gas-phase chemistry, andprocess technology of nitride deposition TiN, TaN, andWN is not yet well-understood for I-PVD processes.ACKNOWLEDGMENTSThe author wishes to acknowledge the assistance of M.Dickson and G. Zhong.This work is currently supported by the National ScienceFoundation and U.S. Department of Energy under Grant No.DMR-9712988.1 National Technology Roadmap for Semiconductors Semiconductor IndustryAssociation, San Jose, CA, 1994.2 S. P. Murarka, Metallization: Theory and Practice for VLSI and ULSIButterworth-Heinemann, Boston, 1993.3 H. Liao and T. S. Cale, J. Vac. Sci. Technol. B 14, 2615 1996, andreferences therein.4 S. M. Rossnagel, J. Vac. Sci. Technol. B 13, 125 1995.5 P. F. Cheung, S. M. Rossnagel, and D. N. Ruzic, J. Vac. Sci. Technol. B13, 203 1995.6 J. Faltermeier, A. Knorr, R. Talevi, H. Gundlach, K. A. Kumar, G. G.Peterson, A. E. Kaloyeros, J. J. Sullivan, and J. Loan, J. Vac. Sci. Technol.B 15, 1758 1997.7 H. H. Andersen, B. Stenum, T. Sorensen, and H. Whitlow, Nucl. Instrum.Methods Phys. Res. B 6, 459 1985.8 Y. Yamamura and M. Ishida, J. Vac. Sci. Technol. A 13, 101 1995.9 S. M. Rossnagel, D. Mikalsen, H. Kinsohita, and J. J. Cuomo, J. Vac. Sci.Technol. A 9, 261 1991.10 R. Powell private communication, 1996.11 T. Ono, C. Takahashi, and S. Matsuo, Jpn. J. Appl. Phys. 23, L534 1984.12 W. M. Holber, J. S. Logan, J. J. Grabarz, J. T. C. Yeh, J. B. O. Caughman,A. Sugerman, and F. E. Turene, J. Vac. Sci. Technol. A 11, 2903 1993.13 C. Doughty, S. M. Gorbatkin, T. Y. Tsui, G. M. Pharr, and D. L. Medlin,J. Vac. Sci. Technol. A 15, 2623 1997.14 M. Yamashita, J. Vac. Sci. Technol. A 7, 151 1989.15 M. Yamashita, J. Vac. Sci. Technol. A 7, 2752 1989.16 S. M. Rossnagel and J. Hopwood, Appl. Phys. Lett. 63, 3285 1993.17 S. M. Rossnagel and J. Hopwood, J. Vac. Sci. Technol. B 12, 449 1994.18 S. M. Rossnagel, Thin Solid Films 263, 11995.19 M. Dickson and J. Hopwood, J. Vac. Sci. Technol. A 15, 2307 1997.20 M. W. Thompson, Philos. Mag. 18, 377 1968.21 S. Hamaguchi and S. M. Rossnagel, J. Vac. Sci. Technol. B 14, 26031996.22 C. A. Nichols, S. M. Rossnagel, J. Vac. Sci. Technol. B 14, 3270 1996.23 J. Hopwood and F. Qian, J. Appl. Phys. 78, 758 1995.24 M. A. Lieberman and A. J. Lichtenberg, Principles of Plasma Discharges

Phys. Plasmas, Vol. 5, No. 5, May 1998J. Hopwood1631and Materials Processing Wiley, New York, 1994, p. 140.25 V. A. Godyak, Soviet Radio Frequency Discharge Research Delphic,Falls Church, VA, 1986, p. 79ff.26 B. E. Cherrington, Gaseous Electronics and Gas Lasers Pergamon Press,Oxford, 1979, p. 119.27 L. M. Chanin and M. A. Biondi, Phys. Rev. 107, 1219 1957.28 A. H. Futch and F. A. Grant, Phys. Rev. 104, 356 1956.29 T. Holstein, Phys. Rev. 83, 1159 1951.30 L. A. Riseberg, W. F. Parks, and L. D. Schearer, Phys. Rev. A 8, 19621973.31 M. A. Biondi and L. M. Chanin, Phys. Rev. 94, 910 1954.32 K. B. McAfee, D. Siple, and D. Edelson, Phys. Rev. 160, 130 1967.33 F. J. de Heer, R. H. Jansen, and W. van der Kaay, J. Phys. B: Atom.Molec. Phys. 12, 979 1979.34 L. Vriens, Phys. Lett. 8, 260 1964.35 D. Rapp and P. Englander-Golden, J. Chem. Phys. 43, 1464 1965.36 L. L. Shimon, E. I. Nepiipov, and I. P. Zapesochnyi, Sov. Phys. Tech.Phys. 20, 434 1975.37 M. A. Lennon, K. L. Bell, H. B. Gilbody, J. G. Hughes, A. E. Kingston,M. J. Murray, and F. J. Smith, J. Phys. Chem. Ref. Data 17, 1285 1988.38 W. Lotz, Z. Phys. 232, 101 1970, and references therein.39 M. Dickson, F. Qian, and J. Hopwood, J. Vac. Sci. Technol. A 15, 3401997.40 S. M. Rossnagel, J. Vac. Sci. Technol. A 6, 191988.41 J. W. Coburn and E. Kay, Appl. Phys. Lett. 18, 435 1971.42 B. E. Cherrington, Gaseous Electronic and Gas Lasers Pergamon Press,Oxford, 1979, p. 169.43 A. Gras-Marti and J. A. Valles-Abarca, J. Appl. Phys. 54, 1071 1983.44 L. T. Ball, I. S. Falconer, D. R. McKenzie, and J. M. Smelt, J. Appl. Phys.59, 720 1986.45 N. Forgotson, V. Khemka, and J. Hopwood, J. Vac. Sci. Technol. B 14,732 1996.