Reactive Sputtering and Process Control

Reactive versus Nonreactive Processes. Both reactive and nonreactive processes may be used in the formation of sputter-deposited films. For nonreactive sputtering, an inert gas, which does not participate directly in the formation of compounds on either the target or the substrate, is used to generate a plasma and sputter material from the target. The inert gas is also sometimes termed the working gas. Argon is used in most cases because its mass is high enough to ensure adequate sputtering yields and it is less expensive than xenon or krypton. Although inert gas ions and atoms are not incorporated as primary constituents, incorporation even in very small concentrations can have deleterious effects on film properties. For example, argon incorporation can cause lattice expansions, increasing the internal stress of the films (Ref 17). Additionally, inert-gas ion bombardment of the substrate/film couple using an applied negative-substrate bias can increase inert gas incorporation as well as alter the growth mode, stoichiometry, and properties of deposited films. Nonreactive sputtering processes are common in the deposition of thin metallic overlayers for electron microscopy, industrial-scale deposition of metals and metallic alloys, industrial-scale deposition of some insulators, and research-scale deposition of insulators and compounds. The primary advantage of this type of thin-film deposition is its simplicity.

Nonreactive processes can be used to directly sputter compound targets such as TiN, which is used in many decorative and wear-resistant applications. However, this use presents several difficulties: (a) the rate at which The pure metal (titanium) can be sputtered is about an order of magnitude lower than the rate at which pure titanium can be sputtered; (b) off-stoichiometry of the films can occur during deposition; and (c) the thermal conductivity of the compound is often much lower than that of the pure metallic species, and thus the target power must be reduced accordingly due to the heating and fracture of the target. These difficulties often preclude the cost-effective use of nonreactive processes in many applications. In these cases, use of reactive sputtering becomes a necessity.

Reactively sputtered films can be deposited using a variety of methods including dc diode, rf diode, triode, magnetron, and modified rf magnetron sputtering. In any case, there are only two basic reactive sputtering modes: compound-coated cathode and metallic cathode. Sputtering in the compound-coated cathode mode is straightforward: sufficient reactive gas is bled into the chamber during sputtering to form the desired compound on the target surface; this compound is then sputtered off and redeposited on the substrate. In many ways, there is little difference between reactive sputtering in the compound-coated cathode mode and nonreactive sputtering from a compound target. The sputtering rates are usually much lower for compounds because of a reduction in the sputtering yield and an increase in the secondary electron emission that is observed with most compound targets. Additionally, depending on the sputtering technique, materials, and deposition conditions, the film may not possess the same chemical composition as the target material. For these reasons, sputtering in the metallic cathode mode is often preferable.

In the metallic cathode mode, the target is maintained as a clean metallic surface and compound formation is limited to the deposited material. Although simple in concept, careful process control is necessary to avoid contamination of the target or deposition of substoichiometric films. Control of the reactive gas species is often costly, requiring at a minimum an automated feedback control and a sensor system to measure partial pressure of the reactive species. However, this is often the only cost-effective means of depositing compound thin-film materials on an industrial scale.

Process Control. The process control necessary for successful reactive sputtering in the metallic cathode mode is often quite difficult to achieve. When flow control is used, the reactive gas is bled into the chamber until there is sufficient gas to form the desired compound at the substrate. However, in most cases this also means that there is sufficient reactive gas present to form the compound on the target surface as well. This phenomenon is known as poisoning of the target and generally results in a several-fold decrease in the sputtering rate and, hence, the deposition rate.

During the early 1980s a number of gas-control methods were proposed based on timed or pulsed gas flows. Although they provided significant improvements in the film deposition rates, truly homogenous films are unlikely to result from this type of pulsing technique (Ref 18). The problem is that simple gas flow control does not permit direct control of the partial pressure of the reactive gas species in the chamber. This is illustrated in the hysteresis behavior observed in measuring the reactive gas partial pressure as a function of gas flow, shown for the case of TiN deposition in Fig. 4(a). In this case, stoichiometric TiN is formed under the partial-pressure conditions of point "B." Clearly this exact condition is difficult to maintain by manually pulsing the gas flow, and in the case of slow pulsing, it is likely that the full hysteresis is traveled with each pulse, creating nonstoichiometric layered films.

Fig. 4 (a) Nitrogen partial pressure vs. reactive gas flow in a mixed Ar-N2 discharge under mass flow control, at a target power of 10 kW. (b) Deposition rate vs. flow hysteresis behavior for TiNx deposition, at a target power of 10 kW, in a mixed Ar-N2 discharge. Source: Ref 18

As discussed, as the target surface becomes compound coated the deposition rate drops precipitously. This effect can be illustrated by plotting the deposition rate as a function of the reactive gas flow, as shown in Fig. 4(b); again the desired operating conditions are given by point "B." However, the problem of achieving stable operation at point "B," in both Fig. 4(a) and 4(b), is nontrivial because a relatively small increase in flow results in a large decrease in the deposition rate, which is then accompanied by a rapid increase in the partial pressure of the reactive species in front of the target.

This instability is influenced by several other factors. During film deposition at the optimum flow rate, the target can very quickly become completely poisoned if the partial pressure or flow increases slightly, as commonly occurs during slight arcing at the target surface. The consequent decrease in the sputtering rate in turn results in an excess partial pressure of the reactive gas species in front of the target, creating a circular chain of events that amplifies the initial instability. Another factor fueling this instability is the target power, because the current-voltage requirements of the target change as the target becomes poisoned. In order to avoid driving the process up or down the hysteresis curve, the input current and/or voltage must be adjusted to maintain a constant target power constant.

Associated with this hysteresis effect is the problem of maintaining film stoichiometry. It has been shown that for TixN1-x film microhardness increases monotonically with increasing nitrogen flow rate until stoichiometric TiN is formed (i.e., point "B" in Fig. 4 a and b). Once this optimal nitrogen flow rate is exceeded, the microhardness drops precipitously and the target becomes poisoned (Ref 19). Before film growth can be resumed, intensive "presputtering" is required to return the target surface to a purely metallic state.

A significant advance in process control was the development of automated flow-control systems using a feedback control loop. Sproul and Tomashek introduced the first closed-loop feedback control system in 1984, which monitored the nitrogen peak height obtained from mass spectrometer analysis and generated a feedback signal for a gas flow controller (Ref 20). Because the peak height obtained from the mass spectrometer can be correlated to a certain gas partial pressure within the chamber, this type of automated flow control is, in effect, a partial pressure control system. Similarly, Affinito and Parsons developed a microprocessor-based system to monitor a number of discharge parameters and provide feedback control of the reactive gas. In this case, control of the reactive gas was possible for nitrogen, but not for oxygen (Ref 21). Both of these control systems work well, but in most cases it is necessary to differentially pump the mass spectrometer head, which adds to the cost and complexity of these systems and can also generate undesirable delays in the control process and signal distortions.

Schiller et al. developed a somewhat different control system, the plasma emission monitor (PEM), that uses the optical emission spectra from the target material to produce a feedback signal to control the reactive gas flow and more directly monitor the conditions at the target surface (Ref 22). The optical emission spectra from the plasma near the target is collected using a collimator connected to a quartz fiber. This signal is fed into a monochromator that is linked to a photomultiplier. The photomultiplier outputs an amplified electrical signal to a control unit where the incoming signal level is compared to a preset "optimum" signal level. A control signal is then sent from the control unit to a piezoelectric gas-control valve that is opened or closed in response to a change in the spectral signal. Typically, the optimum reactive gas level is set as some fraction of the spectral peak height of the metallic species. The spectral peak height of pure metal must be measured for each run just prior to the introduction of the reactive gas, and a thin metallic layer may be deposited before the reactive process attains stoichiometry. In some applications this thin metallic underlayer can create a problem; however, this difficulty can be overcome. The primary advantage of the PEM technique is that neither a mass spectrometer nor differential pumping of the sensor is required, so it is less expensive. This and several similar systems are now commercially available.

Another commercially available device for partial-pressure control is the optical gas controller (OGC), which uses the optical emission generated within the sensor from a gas sample drawn directly from the chamber to determine the partial pressure and drive the closed-loop control system at sputtering pressures (Ref 23, 24). Like the PEM system, no mass spectrometer or differential pumping is required. However, unlike the PEM system, the OGC is based on electron-impact ionization spectroscopy. This permits the process gases to be ionized under reproducible conditions within the sensor head, but it also adds the limitation that the sensor must be located very close to the target to provide accurate control (Ref 24, 25, 26, 27).

Alternatively, research by Penfold (Ref 28), Kadlec et al. (Ref 29), and others has shown that the hysteresis effect can be eliminated entirely by increasing the pumping speed of the system beyond a critical level. While successful, this technique is rather expensive in practice, resulting in reduced deposition rates, inefficient gas use, and very high pumping throughput requirements. Whatever the method, effective process control of sputtering is critical in the production of reproducible high-quality thin films at rates that allow them to be commercially competitive with other thin-film processes.

0 0

Post a comment