Arthur Sherman - Chemical Vapor Deposition for Microelectronics, страница 10

Above 10kHz, there is not enough time for the electronsin the positive column to be lost to the walls, so the discharge remains lit con-Fundamentals of Plasma-Assisted CVD49tinuously. Depending on the discharge geometry and gas involved, the startingvoltage for an AC discharge can depend on the RF frequency and pressureYFinally, when an AC discharge is operated with a blocking capacitor between the power supply and one of the electrodes, that electrode will assumenegative self bias.

Such an average negative voltage on this electrode can serveto accelerate ions toward it with considerable energies.It is interesting to note that this is the ion bombardment that is used inplasma ("dry") etching to promote the anisotropic character of that process.It is also important in the understanding of sputtering phenomena. 13CROOKESjASTON}ARADAY~CATHODE:NEGATIVEGLOWPOSITIVECOLUMN-~LIGHT INTENSITYPOTENTIALDISTRIBUTION-~ ~FIELD STRENGTH~++J+=-f\NET SPACE CHARGE'6lJ+-If\NEGATIVECHARGESJ+-~POSITIVECHARGESJ+Figure 4: A DC glow discharge at low pressure. 1250Chemical Vapor Deposition for MicroelectronicsWhy the negative self bias forms has been described by Butler and Kino 14and can be seen from Figure 5.

In the first figure, an alternating potential isapplied to a conducting probe in a plasma. When the probe sees a positivevoltage, a large electron current flows. Reversing the voltage produces only asmall current flow due to the immobility of heavy ions. Thus when the probeis conducting, the self bias is zero and large net currents flow on average. Onthe other hand, when the probe is nonconducting (i.e., electrode attached tothe blocking capacitor), the behavior is shown in the second figure.

In orderfor the average current to be zero, the average applied voltage (i.e., RF signal)must become negative.;{!.tzw0::0::::>02.82.82.42.42.02.01.61.61.2:i0.80.4,0-0.4-0.8-1.20.8tzw0.40::0:::JIONCURRENT0o -0.4-0.8RF SIGNAL-1.2-1.6-101.2!-1.6010VOLTAGE-10010VOLTAGEFigure 5: Creation of negative self bias in AC discharge. 142.3.2 AC Discharges with Unequal Area ElectrodesNext, it will be valuable to consider the discharge behavior when the electrodes are not of equal size.

A simpl ified analysis of this situation can be made 15if we make a number of approximations. Consider a geometry such as shown inFigure 6, where a blocking capacitor is used between the power supply andel ectrode 1. The function of the blocki ng capacitor is to allow a DC bias to existbetween the DC plasma potential and the electrode adjacent to the capacitor./NONCONDUCTING CHAMBERBLOCKINGCAPACITORPOWER SUPPLYFigure 6: Reactor with unequal size electrodes.Fundamentals of Plasma-Assisted CVD51We begin by assuming that the ion current density to all internal surfaces isequal.

Then, if we assume a collision-free sheath with the field being zero atthe edge of the positive column, we can express the space charge-limited ioncu rrent as 16,j.0.:where jj is the ion current, V is the potential difference across the sheath, Lis the sheath thickness, and mj is the ion mass.Since the ion currents on each electrode have been taken equal, we canwrite(9)For capacitances in series, we can write(10)where C 1 and C2 are the capacitances of the two sheaths.

Again, we can assumethe capacitances of a dark space to be aA/L, where A is the electrode area.Then, we have(11 )Eliminating L1 !L 2 between Equations (9) and (11) gives(12)Then, replacing C2 !C 1 in Equation (12) from Equation (10) results in the desired relationship between Vl/V 2 and AI!A2 .Clearly, there are many approximations in this model, so close agreementwith experimental data should not be expected. It is interesting, however, toconsider some relevant experimental data. 17 In these experiments, peak-to-peakand DC bias voltages were measured for a 13.56-M Hz RF glow discharge confined within a reactor chamber with two electrodes of unequal area.

Using a52Chemical Vapor Deposition for Microelectronicssimpl ified theoretical model of the discharge, the authors derived values forthe voltage on each electrode relative to the plasma potential, and could thencalculate the voltage ratios from the measured data. It was found that the voltage ratios depended not only on the reactor area ratio, but also on the electrode material, the peak-to-peak voltage across the electrodes, the gas and itspressure. Results are shown in Figure 7 for Argon gas in a stainless steel systemat 50 mTorr with a 600 peak-to-peak voltage.20KOENIG 8 MAISSELCOBURN 8 KAYORIGINALMODIFIED.. /.............../'" ,.~10tI~II~\-5"\V"V' ( R/L tv'ODEL\"IIA..Jl.~2Ar GASSTAINLESS STEEL SYSTEM6.7 Po, 600 Vpp0.10.2CORRECTED AREA RATIOFigure 7: Voltage ratio versus area ratio for argon plasma.

17For these particular conditions, it can be seen that V I /V 2 = (A I /A 2 )n isa reasonable model, but that n == 4 only between A 1 /A 2 = 0.6 and 1.0. Forsmaller area ratios, we find n == 1. The data reported by Coburn and Kay I8were modified by the author by correcting the areas they used in their calculation.From these results, we see that under certain conditions it is possible tohave substantial ion bombardment on the powered electrode while almost noneon the remainder of the chamber. The magnitude of the DC bias on each electrode will, in general, depend on peak-to-peak voltage, gas chosen, pressureand chamber materials, as noted earlier. For a given experimental setup, wecan change the DC bias on the grounded electrode of Figure 6 by placing a variable LC circuit in between it and ground.

19 Such an arrangement where thesecond electrode is isolated from the grounded outer chamber is shown inFigure 8. With this circuit, it was possible to develop a 30V DC bias on the substrate holder.Fundamentals of Plasma-Assisted CVD53CathodeRF po\vcr in --"'---1(13.56 MI-Iz)C:=========~E----tt Target Si0 2r-nsu) a tedsubstrateholder~----+t-IDC voltage meterFigure 8: System for substrate electrode tuning to decrease DC bias. 192.3.3 Frequency Effects on RF Plasma Reactor BehaviorWhen the RF discharges are used to create plasmas in PECVD reactors,the infl uence of the frequency at which the discharge is operated is anotherquestion that must be explored. It is generally recognized that a lower frequency discharge (50 to 100 kHz) will produce a CVD film with greater compressive stress than a film created in a higher frequency discharge (13.56 MHz).The speculation has been that the ion bombardment is more intense at lowerfrequencies, and this bombardment causes the film stress to be compressive.In fact, recent careful experimental work has verified the more intense ionbombardment at low frequencies.

2o Studies were carried out in a plasma etching apparatus where a hole in one electrode allowed ions to be examined in amass spectrometer while different retarding voltages were applied. The retarding voltage necessary to cut off beam current was recorded as a function ofboth power level and frequency. The results are shown in Figure 9, whereearlier results 21 are shown in parenthesis.There are two potential explanations of why the ion bombardment is moreintense at low frequencies. First, the sheath potential drop, on average, willbe higher at the lower frequencies. The el ectrons are Iighter than ions, so they54Chemical Vapor Deposition for Microelectronics600.------r----y---........---~-,A(6)400200(0)(e)oFigure 9: CI 2 plasma beam maximum ion energy.

Circles = 27 MHz;triangles =100 kHz; solid = CI 2+; open = CI+.2° Reprinted by permission of the publisher,The Electrochemical Society, Inc.tend to preferentially diffuse out of the plasma, and the electrode assumes anegative bias. As frequency is increased, there is increasingly less time available for charged particles to diffuse to the reactor walls between cycles. Therefore, if there is less opportunity for electrons to diffuse out at higher frequencies, there will be less need for a strong bias to form.

If we recognize that thenegative bias formed has the effect of accelerating ions toward the surface,then more bias (lower frequencies) means more ion bombardment on the forming film.Secondly, the plasma potential will vary with time, depending on the cycleof the appl ied potential.

22 If the ion can transit the sheath before the appl iedelectric field reverses, it can experience the maximum sheath potential. Asthe frequency is raised, the ion cannot cross the sheath before the field reverses,s-o it experiences the average sheath potential which is approximately one-thirdthe maxirnum. Therefore, ion bombardment will be less intense at the highestfrequencies.2.3.4 Influence of Applied Magnetic Fields on RF Plasma ReactorsThe final aspect of reactor influence on a plasma that we wish to discussis the use of an applied magnetic field. This is one more parameter that canbe adjusted to modify the plasma and the reactive species produced.

For theFundamentals of Plasma-Assisted CVD55mean-free paths typical of plasmas used in PECVD equipment (~1 mm), amagnetic field on the order of 100 gauss is sufficient to cause the electrons togyrate many times between collisions. This has the effect of immobilizing theelectrons. Since there can be very little charge separation in a plasma, the ionswill be attracted by the immobilized electrons and be immobilized as well. Allneutrals (free radicals, excited species, etc.) will be uneffected by the magneticfield. Therefore, the magnetic field provides a means for possibly confiningthe plasma to one region of the reactor (over the wafer, for example) or modifying the spatial distribution of reactive species.Such effects were demonstrated in an experiment where a discharge in CF 4gas at 10 mTorr pressure was established along the axis of acyl indrical tube,and an axial magnetic field of 30 gauss was applied. 23 The plasma was confinedto the central 1-inch diameter core within the 5-inch diam.eter tube.

Two newphenomena are created by this arrangemement. First, the species that causeetching (i ,e., of Si or Si0 2 ) are neutrals (i n this case, F atoms created byCF4 + e-~ CF 3 + + F + 2e-), and they are present throughout the region. Ifions enhance the etch rate, they are only present in the plasma, so the etch rateon a wafer will depend on whether it is in the plasma or not. Again, when thewafer is in the plasma, biasing it to increase ion bombardment will also increaseetch rates.