Roland A. - PVD for microelectronics (779636), страница 16
Текст из файла (страница 16)
ROSSNAGELFIG. 3.19 Reactive sputtering hysteresis curves. (a) Deposition rate, (b) discharge voltage, and (c)chamber pressure, all as a function of reactive gas flow during the sputtering of a metal target.oxygen now becomes part of the working background gas. This leads tooxidation of the A1 cathode, which has significant and unfortunate consequences. The most significant effect is that the sputter yield of the oxidizedA1 target is much, much lower (up to 25 times) than that of clean A1. Thismeans that the amount of AI atoms emitted from the cathode drops radically, and as a result, the capability for this sputtered A1 film to absorb oxygen reduces markedly.
This is a runaway type of effect. As the cathodestarts to become oxidized, the A1 emission rate drops, which leads to anPLASMA SYSTEMS79increasing amount of free oxygen in the working gas, which causes additional oxidation of the cathode.In short, at this critical flow, the system undergoes an irreversible transition from a metallic cathode operating at high rates to an oxidized cathode operating at very low rates.
The deposition rate (Fig. 3.19a) dropsrapidly and the chamber pressure (Fig. 3.19c) increases rapidly. It alsoturns out that for many gas-cathode systems, the discharge voltage changessignificantly also, due to changes in the secondary electron yield goingfrom a clean to a compound surface (Fig. 3.19b).Another way to characterize the hysteresis problem is to examine the dependence on the overall system pumping speed. In a classic experiment, itwas shown that by increasing the pumping speed, and at the same time increasing the reactive gas flow to attain the same operating pressure, thehysteresis effect became less pronounced and was virtually eliminated atthe highest flows (Fig.
3.20) [3.26]. This shows that if the pumping effectof the film can be overshadowed by the pumping speed of the systempump, then changes in the wall pumping speed (i.e., oxidation of the wall)will have little net effect.At flows above the critical flow, there are few changes in the plasma.The chamber pressure increases linearly, but the deposition rate and voltage do not change significantly. If the flow is reduced back to the criticalflow point, the system does not revert back to the clean metallic state. Thisis due to the presence of the compound layer on the cathode, which reducesFIG.
3.20 The chamber pressure hysteresis plot for reactive sputtering as a function of chamber flowrate [3.26].80R. POWELL AND S. M. ROSSNAGELthe amount of metal that can be emitted. It is necessary to go to a significantly lower reactive gas flow to get to the point where the films on thewalls and the sample can take over as the primary sink for the reactive gasspecies. The overall flow dynamics of this system show a response muchlike the hysteresis response of magnetic materials, and these reactive gasflow plots are called hysteresis curves.The fundamental problem with reactive sputter deposition is that thebest films (i.e., the most completely oxidized films) are deposited at a flowjust below the critical flow point.
In addition, since the cathode is not yetoxidized at this point, the deposition rate is high. The basic problem is thatas the critical flow is reached, the system changes rapidly into the oxidized-cathode mode, and once this transition has begun, it is very difficultto control. It should be noted, though, that in most cases it is easier to control the nitride transition than the oxide transition. There are a number ofsubtle reasons for this: The sputter yield of the cathode does not change asmuch when the cathode is nitrided (for many materials) compared to whenit is oxidized; the secondary electron yield of the cathode may not go up(which reduced the voltage) and may actually go down, which results inhigher cathode voltage and potentially increased sputter yields; and finallymany of the nitride reactions of interest (for example, TiN and AIN) areless spontaneous than their equivalent oxygen reactions.
In fact, many ofthe nitride reactions require energy to form, which is markedly differentfrom the case of oxidation. This allows better control of the process usingsimply the power supplies.Until recently, reactive sputter deposition of oxides was fairly difficulton an industrial scale due to these intrinsic control issues [3.27].
Althoughit was always possible to operate in the oxidized-cathode mode, the rateswere low enough to be impractical. In addition, since a nonconductingoxide was being deposited around the chamber, there were problems withboth arcing on the cathode edge and coating the anode with oxide, resulting in a "hidden anode" that inhibited current flow and led to more arcing.The solutions to controlling reactive deposition lie in two areas: the control of the reactive gas flow and the time-dependent control of the cathodepotential.
The gas flow problem, which is the desire to operate just belowthe critical flow, can be solved by using a fast-feedback control system forthe gas supply [3.12]. The voltage to the cathode can be pulsed in the20-200 kHz range such that for a relatively small fraction of the cycle, thecathode is operated at a positive potential. Much like the case of the RFdiode, this reduces charging and arcing effects, and allows better control ofthe deposition process. This has allowed, for example, the deposition ofhigh-quality aluminum oxide at a rate of 78% of the pure metal depositionPLASMA SYSTEMS81rate [3.13]. This is about 25 times faster than when using conventional RFpower in an oxide mode. As a parallel to this pulsing approach, it is alsopossible to use two adjacent cathodes, one of which functions as the anodefor the other.
These cathodes are then switched in polarity in the 100 kHzrange to reduce arcing and charging [3.14, 3.15].3.10 Practical Plasma Issues in PVD ToolsThe mechanical design of magnetron cathodes will be covered in Chapter4, and some aspects of the overall tool design will be discussed in Chapter5. There are some plasma issues that are relevant to introduce here, such asshielding and tooling, plasma breakdowns, gas density issues, etc.The plasma in a magnetron system, as well as in most other types ofplasma tools, expands to fill the volume of the tool: There is no real confinement or containment mechanism. This can be misleading with a magnetron system, where it appears that the plasma is all tightly connected tothe cathode surface and the rest of the tool is mostly dark.In a magnetron tool, the plasma density 'in' the etch track (E x B driftpath) can be 100-1000 times denser than the surrounding regions.
A measure of this can be to apply a small negative bias to the sample holder andsee how much current it draws. When sputtering with a few kW of discharge power on the cathode (perhaps 5 A at 500 V), substrate currents ofonly tens or a hundred milliamps are measured. This means that a strongsubstrate bias power is going to be difficult to obtain; there simply are notenough ions at the sample to make much of a difference. There is a classof magnetrons, called unbalanced magnetrons, which are designed to havea less well confined plasma near the cathode and, as a result, much highersubstrate bias capability [3.16]. However, unbalanced magnetrons have notbeen applied to semiconductor manufacturing technology and will not bediscussed here.The weak chamber plasma does result, though, in concerns about cathode currents and arcing. Most magnetron cathodes are configured with aground shield, or the same function may be designed into the mount for thecathode.
A ground shield is a grounded piece of metal located near the edgeof a magnetron cathode and serves to keep the chamber plasma away fromthe sides of the cathode. This eliminates sputtering of the edge of the cathode and perhaps the backing plate, as well as shorting or arcing across theinsulator used to hold the cathode in the chamber. This insulator is alwaysrequired, as the sputtering chambers are grounded and the metal cathodesare powered at - 3 0 0 to - 5 0 0 V DC.
The ground shield may also be usedR. POWELL AND S. M. ROSSNAGELto prevent metal deposition onto this insulator ring. This function may alsobe designed into the insulator mounting with a reentrant design such thatsputtered metal must hit one or two other surfaces before it can get to theinsulator.The spacing criteria for shields within the chamber is such that the openings be on the order of a Debye length or two. From Eq. (2.5), the Debyelength is (in cm) 7 4 3 (L/Fie)1/2, where T is the electron temperature in eVand n is the electron (plasma) density in electrons/cm 3.
The low-densityplasmas in the magnetron chamber might have densities of at most 109/cm 3and an electron temperature of 1 eV, which leads to a Debye length of0.25 mm. This is a fairly tight tolerance (1/2 mm), particularly given thatmost production cathodes are 12-14 in. (300-350 mm) in diameter andthat they are mounted on deformable o-rings.
In addition, this small spacing must hold off 500 V in the presence of a weak plasma as well as smallparticulates and flakes. Therefore, shielding tolerances are generally on theorder of a few mm, and the small stray currents that might be observed bythe very weak plasma across the insulator are ignored.RF plasmas are less well confined than DC magnetron discharges. In anRF plasma, the cathode functions as the cathode for most of the RF cycle,but then switches and becomes the anode for a small fraction of the cycle.At this point, the rest of the chamber becomes the defacto cathode eventhough the bombardment energy is low.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
