Roland A. - PVD for microelectronics (779636), страница 62
Текст из файла (страница 62)
Also,in many advanced multilayer designs, the upper wiring layers are so-calledfat levels in which the lines and vias are both relatively thick and of lowPVD MATERIALSAND PROCESSESDamasceneprocessStandard333processMETALr~f',j,',j"j,,j,'~rj,j,',~,Deposit Blanket OxideIPattern Oxide andEtch ChannelsJDeposit Metal in Channelsand on FieldIDeposit Blanket MetalPattern Metal andEtch LinesDeposit ThickInterlayer Dielectric (ILD)&ILevel Metal Using CMPmm"//////ZDeposit Dielectricover Top SurfaceFIG. 9..rmm/1"/Level ILD Using ChemomechanicalPolishing (CMP)Comparison of conventional and damascene metal wiring processes.aspect ratio ( ~ 1"1).
For these layers, PVD barriers and liners are expectedto be applicable even beyond 0.18 ,ttm, although the process used for Cufilling could be CVD or electroplating.9 . 8 . 2 SPUTTERING AND SELF-SPUTTERING OF C uOxygen-free highly conductive (OFHC) elemental Cu targets are readilyavailable with purity of 5 nines and above and, as with A1-Cu alloy targets,can be used for high-power magnetron deposition on 200-mm Si waferswith comparably high specific deposition rates ( ~ 20 A/kW-sec). Cu has ahigher sputter yield than AI under Ar § bombardment (e.g., 2.3 Cuatoms/ion versus 1.2 A1 atoms/ion at 600 V) and a greater mass than the Arworking gas (64Cu > 4~ > 27A1). Hence, all things being equal, we wouldR.
POWELL AND S. M. ROSSNAGEL334expect about 2 times more sputtered flux from a Cu target than from an A1target, with reduced gas-atom scattering and increased directionality in thedeposited film.Probably the most interesting aspect of PVD Cu is the possibility ofusing self-sputtering to completely do away with the inert Ar workinggas and associated gas-atom scattering, residual gas impurities, and Arincorporation that are of concern at conventional mTorr-type PVD pressures [9.46-9.49].
In this self-sputtering mode, the magnetron dischargeis initiated with Ar gas but is sustained under ultralow pressure (e.g., 5 x10 -5 Torr) with sputtered Cu atoms that are ionized in the DC plasma region and accelerated by the electric field to the target. In conventionalmagnetron sputtering, the high density of secondary electrons leaving thetarget in crossed electric and magnetic fields (E • B) gives rise to a highionization rate of inert gas ions, which subsequently sputter-erodes thetarget While sputtered metal can also become ionized near the target, thisionization fraction is typically very small. On the other hand, if the production of metal ions and their self-sputter yield are sufficiently greaterthan unity, then a discharge can be sustained in the absence of any inertgasprovided that the magnetron fields are designed to redirect a largefraction of the metal ions back at the target.
These conditions are discussed in refs. 9.46 and 9.49, and the basic idea is illustrated in Fig. 9.33.The self-sputter yield Y as a function of atomic number Z and ion energy E has been given by Zalm [9.50] as"9Z!/2 )Y=Uo(Eln - O.O09U~/2)(9.4)where U0 is the sublimation energy of the element. Applying Eq.
(9.4) toCu and A1 at E = 600 V, we see that Cu is an outstanding candidate forself-sputtered PVD since it has a yield of Y ~ 2, which is several timeslarger than that of A1. On the other hand, the first ionization potential of Cu(E / = 7.68 eV) is higher than A1 (El = 5.96 eV) so it is more difficult toproduce Cu + than AI + by electron-impact ionization.
The net result, however, is that Cu can be self-sputtered at high rates (> 1 /xm/min) at lowpressure ( ~ 2 x 10 -5 Torr) in the absence of any inert working gas [9.46].However, the power density (80 W/cm 2) in this case was several timeshigher than what could be sustained with a conventional large-area planarmagnetron (e.g., 20 kW into a 12-inch-diameter target is ~ 30 W/cm2). Itremains to be seen whether self-sputtering of Cu will enter mainstream ICmanufacturing; however, the exploitation of metals ions to improve PVDperformance is an established trend. For example, as discussed in Chapter 8,PVD MATERIALSAND PROCESSES335FIG.
9.33 Comparison of conventional PVD Cu with self-sputtered PVD Cu in which Cu + ions replace Ar ~ ions, allowing the Ar working gas to be eliminated from the process.directional PVD of Cu and other metals based on plasma ionization ofsputtered-metal neutrals (albeit at a relatively high pressure of inert working gas) is under active development.9.8.3P V D TA AND T A N BARRIERSAs noted earlier, integration of Cu into IC processing will require deposition of suitable barriers to prevent its diffusion into Si and electric-field assisted drift into SiO 2. An important difference in this regard between Cu336R. POWELL AND S.
M. ROSSNAGELand A1 is that A1 can be directly deposited onto oxide with good adhesionand formation of a self-limiting interfacial A1,0, while Cu requires a barrier. Therefore, replacement of a PVD Al interconnect with PVD Cu alsorequires additional barrier layers on the sides of the metal line to preventits contact with the intermetal dielectric.
Since the resistivity of the barrieris much greater than that of Cu, the effective interconnect cross-sectionalarea is reduced, which offsets to some extent the RC gain expected fromgoing to Cu over Al. This effect is illustrated in Fig. 9.34 for a via hole anda 20-nm-thick barrier layer.
In the case of Cu, the bottom and sidewall barrier can use up 50% or more of the volume of a 0.18-pm hole. Assumingthe barrier resistivity is much greater than that of Cu, this leads to an effective conductivity of the filled feature of 2.8 p a - c m - much higherthan the 1.7 p a - c m one would obtain if the feature were filled with pureCu.
In contrast, the bottom-only barrier for A1 takes up less than 5% of thevia hole volume. From the resistivity standpoint, there is clearly a reasonto use as thin or as conductive a barrier as possible. On the other hand, itis questionable just how thin such a material can be made before its barrierproperties are no longer sufficient to prevent Cu diffusion and drift into theadjacent regions of the hole.A wide variety of materials have been investigated as diffusion barriersfor Cu.
including refractory metals (e.g., W, TiW, Ta), nitrides (e.g., W,N,TiN. TaN, Ta,N). silicides (e.g., TiSi,), and ternary amorphous alloys sichas Ta-Si-N [C):5I I. The Cu barrierladhesion layer of choice is an open issueat this time. It may be possible to simply extend older ~iiaterialsthat haveperformed well as barriers for Al metallurgy, such as TiITiN and TiW.However, two "new" materials - Ta and TaN, - have shown considerable promise as thermally stable barriers with Cu metallurgy [9.52-9.543.Ta is thermodynamically stable with respect to Cu, while Ta nitrides arechemically inert with Cu because of the absence of any compounds between Cu and Ta or Cu and N.
I t has been found that a small amount ofcontamination in a Ta film - e.g., a few atomic percent of oxygen or nitrogen - improves its ability to withstand Cu penetration. Hence, ultrapure Ta films are not preferred for this application.In analogy with the Ti/TiN system, Ta can be deposited by the sputtering of an elemental Ta target in Ar, and TaN can be deposited by the reactive sputtering of Ta in an ArIN, admixture. However, an important difference is that unlike the TiITiN case, there are two phases of PVD Ta films(a high-resistivity P-Ta phase with p = 180 p a - c m and a lower-resistivity,body-centered cubic phase of bcc-Ta with p = 40 p a - c m ) as well as twonitridation states (TaN and Ta,N). The impact of this complexity is illustrated in Fig.
9.35, where daia on reactive sputtering of Ta onto roomPVD MATERIALS AND PROCESSES337FIG. 9.34 (a) Different barrier requirements of PVD Cu and PVD AI lead to (b) different tractionalvolume of barrier in a via hole.338R. POWELL AND S. M. ROSSNAGEL50060fcc-TaN400Amorphous50tg,~300v._~>".~[3-Tat-~9 200_~"~ . . O.~ ~99.~176176"~176z.~40ro30~~o.."O:9rr~2o ~~100"bcc-Ta0L....,..:...........0510Resistivity-] 10.....
O .... N contentI15_. ~I209.125]9030N2Partial Flow (%)FIG. 9.35 Resistivity and nitrogen content of reactively sputtered TaN as a function of N~-to-Arflow ratio 19.521. (Reprinted with permission from K.-H. Min et al, J. Vac. Sci & Tech. B14(5):3263-3269 (1996). Copyright 1996 American Institute of Physics.)temperature Si(100) substrates is shown [9.52]. Based on a detailed analysis of these films, it was found that Ta sputtered in pure Ar gave predominantly/3-Ta films (with a small admixture of bcc-Ta) whereas addition ofa very small amount of N 2 to the working gas ( ~ 3% Nz/Ar flow ratio) ledto bcc-Ta.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
