Roland A. - PVD for microelectronics (779636), страница 49
Текст из файла (страница 49)
Hopwood and Qian's I-PVD dischargemodel was used to describe the same situation, as shown in Fig. 8.11[8.11]. In this case, the model uses a particle density of the sputtered AIatoms rather than a magnetron discharge power, but the general trends canbe easily seen.
The model was also used to predict the electron temperature and shows a significant cooling effect as a function of increased metaldensity (Fig. 8.12) [8.11 ].An additional indication of the cooling of the discharge is the level ofthe ion saturation current to the sample. If a sample is biased sufficientlynegatively to repel all electrons from the plasma, then the net ion flux tothe sample is simply the Bohm presheath flux, which was listed as (seeSection 3.4) [8.12]F = 0.6(ne)(8.2)where n e is the electron density in electrons/cm 3, T is the electron temperature in eV, and M is the ion mass.
The reduction of the ion saturationcurrent as the level of metal atoms and ions in the discharge increases isR. POWELLAND S. M. ROSSNAGEL2561 . 0.....'- 0.8.o"~...~...AIDensity(1/cc):---- 1 x 10112 X 1012~06~-/_.~~ 0.20.4 ~0.00.0I . . . . .
. . . .0.2r ......-,-,-,w0.4.....~-0.6-,t. . . . . .0.8Ta, i . . . . . . . . .1.01.2Electron Density (1012 cm -3)FIG. 8.11 Calculated AI ion fraction as a function of RF inductive power for several different AIatom densities [8.8].another indication of the reduction in the electron temperature in thedischarge.There may also be a subtle effect of the chamber size on the level ofmetal ionization in an I-PVD tool. Hopwood and Qian have suggested thatsmaller chambers result in increased diffusional losses of the metal ions~.51.4>1.3vI-121.11.0t J w0.09 w wtt! I I-t0.2t I ~ tt t I t-it0.4t t tv~-'-1'0.6II~9Itt',I'*WTItvI'w,0.8I , , i | , i | ,I1.01.2Aluminum Density (1013crr~~)FIG. 8.12Calculated electron temperature as a function of metal atom density [8.8].IONIZED MAGNETRON SPUTTER DEPOSITION: I-PVD257[8.11]. He has computed the relative ion flux fraction as a function of thecharacteristic diffusion length (A) for the case of A1 sputtered in Ne(Fig.
8.13) [8.11]. The implication is that if the diffusion length is on theorder of or larger than the chamber dimensions, increased losses of metalions will occur and the relative ionization will be reduced. Currentlythere are only two published reports of relative ionization in differentchamber sizes to support this conclusion.
In a relatively small chamber(8-cm-radius • 10-cm-length), Yamashita measured an A1 ion fraction of65% [8.13]. In the work of Rossnagel and Hopwood [8.5, 8.6], the measured ion flux fraction was over 80% in a chamber with a radius of 25 cmand a length of 30 cm.The primary ionization process in an I-PVD tool was originally thoughtto be electron-impact ionization. However, Hopwood has published calculations of the relative levels of both electron-impact and Penning ionization. At low electron densities, Penning ionization is seen to dominate, particularly for Ar (Fig.
8.14) [8.11 ]. The experimental work of refs. 8.5 and8.6 tend to suggest a dominant role for electron-impact ionization, particularly at high powers. This is consistent with the observation of a higherrelative ionization with Ne as compared to Ar. A Ne plasma would be expected to have a higher electron temperature (2 times that of an Ar plasma).1.00.8.Oo1.33 eV1.43c-0.6IIX_=1 . 9 ~1.1_,-- 0.4OmNe=5x2.5E1 0 ~ c m -aAI = 1 x 1012 cm-336 mTorrc-E0.2<0.0,,,,0.0~ r3.0. . .
.~..,..v,w ,,w,i6.0w w v ii,i.9.0v v v.w.~ . v r v12.0Diffusion Length (cm)FIG. 8.13 Calculated Ai ion flux fraction as a function of characteristic diffusion length for I-PVDof A1. The calculated electron temperatures are listed for each point [8.11 ].R.
POWELL AND S . M. ROSSNAGEL/-- -Penning lonlzatdon Only- Electron Impact OnfyTotal1o910'0lollI1o12Electron Density ( ~ r n - ~ )FIG. 8.14 The relalive ionization level as a function o f electron density showing the relative importance of Penning and electron-impact ionizalion procesceh [8.11].Also, the Penning ionization cross section for Ne is smaller than that forAr [8.l I ] .An interesting problem remains as a fundamental issue relating to I-PVDplasmas.
The introduction of high levels of metal atoms into the dense, inductivcly couplcd RF inert gas plasma appears to rcsult in a decrease inboth the electron temperature ltnd the plasma density. Generally the electron temperature and thc density would be expected to move in oppositedirections; i.e., lower electron temperature is compensated for by higherplasma density. A recent paper has examined this dilemma [8.14] and hasattributed the problem to gas rarefaction effects.
which have been systematically observed duri~igmagnetron sputtering, particularly at higher pressures 18.151.Gas rarefaction is a heating phenomenon. driven by the kinetic energyof the sputtered atoms. As they slow down due to gas-phase collisions. thesputtered atoms transfer some of their kinetic energy to the backgroundgas. raising its effective temperature. Since all this occurs in an open container.
the net result is a local reduction in the gas density in the region directly in front of the cathode [8.151. The recent paper by Dickson ut ul.suggests that the gas rarefaction effect may be responsible for the densityreduction a s the sputtering level is increased. The electron temperature reduction is still driven by the much lower ionization levels of the metalatoms [8.14]. Experimental measurements of the local gas density in front259IONIZED MAGNETRON SPUTTER DEPOSITION: I-PVDof the I-PVD magnetron cathode show both a significant rarefaction due tothe RF inductively coupled plasma and increasing rarefaction due to increases in the magnetron discharge power, the latter directly related to theamount of metal introduced into the discharge (Fig.
8.15).The significance of this problem is actually quite important to the largescale implementation of I-PVD. The addition of large amounts of metalatoms to the discharge is obviously necessary to attain a high depositionrate. Yet, as the metal flux is increased, the relative ionization decreases fortwo reasons: the cooling of the electron temperature and the increasedtransparency of the discharge region due to rarefaction. In effect, increasedmetal sputtering into the discharge is equivalent to reducing the operatingpressure, which results in less relative ionization because of the shorterresidence time for the sputtered atoms in the plasma.As a result of this problem, the near-term implementations of I-PVDhave tended to focus on liner and diffusion barrier applications rather thanthick film deposition.
This is convenient in that it supplants collimatedsputtering, which is used on a wide scale in the industry, but is plagued byconcerns over uniformity, rate, clogging, etc. However, I-PVD technology0.55r600 W RFe0.5E800 W RFca 0.45 ~oT-lix1000 W R Fo.4~"~0.35rcn1200 W RF0.3-1400 W RFO"~ 0 . 2 5 Uo.J0.2r10.15 0500JJ1000,1L1500l2000Magnetron Discharge Power (watts)FIG. 8.15 The local gas density in the plasma region of an I-PVD tool (300-mm cathode) as a function of magnetron discharge power for various RF inductively coupled power levels.
The startingchamber pressure was 30 mTorr, which is equivalent to 1 • 10~5/cm3. Therefore, even at the lowestRF power used (600 W) the gas is already rarefied by 50% before the magnetron is turned on.260R. POWELL AND S. M. ROSSNAGELis still constrained by this reduced relative ionization problem, which maylimit eventual applications at high rates.8.3 Deposition and Experimental ResultsThe use of I-PVD for semiconductor applications encompasses two differing regimes: thin layers inside trenches or vias and the filling of trenchesand vias.
The thin layers may be conformal or may be selectively depositedonly on the bottom or sides of features. They are not intended to be the primary current-carrying metallurgy, though. The fill metal's role is primarilylow-resistance current flow.The thin layers have numerous functions. For example, thin Ti layersoften are used at the bottoms of vias or contact holes.
The function of theTi is to provide better electrical conductivity to the underlying metal. Ti ischemically reactive and can break down the surface oxide of the metal, inessence stealing the oxygen atoms from the underlying oxide in an attemptto form TiO 2. The Ti may also be used to form a silicide with underlyingSi. Thin films within a via or a trench may also be used as a chemical diffusion barrier either between metal layers to reduce electromigration andinterdiffusion problems or to protect an underlying layer from chemical attack. An example of this is the use of TiN layers on SiO 2 sidewalls, whichprotect the SiO 2 from chemical attack by the WF 6 working gas used for WCVD deposition.
A third application is the use of Ti or TiN layers to promote growth of the desired (111) phase of AICu on subsequently depositedfilms. This phase has been shown to be the most resistant to electromigration problems. It is not clear, however, whether buried, damasceneprocessed lines will have similar microstructure requirements.At high levels of relative ionization of the metal species, the majority ofthe depositing species arrive at normal incidence to the sample surface.Three effects that might lead to a nonnormal arrival angle are (1) scattering within the sample sheath, (2) electrical deflection by the surface topography, and (3) the initial kinetic energy of the metal ion previously in theplasma.
Gas scattering is an initial concern because the operating pressureis many tens of mTorr. However, the sample sheath thickness is on theorder of a few hundred microns, which is 10 times smaller than the meanfree path for gas scattering. Second, since the sample sheath is 100 or moretimes the scale of the surface topography (typically 1 micron or less), thesurface appears flat to the incoming ions and they are not deflected electrically by ledges or via walls on the surface. Finally, even though the sputtered atoms that feed the metal-rich plasma can have several eV of kineticIONIZED MAGNETRON SPUTTER DEPOSITION: I-PVD261energy when emitted from the cathode, gas scattering at 10-40 mTorrquickly damps out this energy and the sputtered atoms are nominally "thermalized" prior to ionization and acceleration.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
