BY KUOHSIUNG LI AND PHIL GREENE FROM FERROTEC
SMARTPHONES have freed us from the confines of physical location, allowing us to expand communication and commerce with many distant people or enterprises simultaneously or in rapid succession. Fifth Generation (5G) mobile networks to be rolled out by 2020 will offer tremendous improvement, increasing spectral efficiency 1000 fold, compared to today’s widely deployed 4G LTE network. This feat of mobile communication gives us instantaneous ‘virtual presence’ anytime, anywhere even as we travel the globe. The 5G implementation will place an ever higher premium on spectrum utilization, promoting a broader deployment of acoustic wave filters and resonators in mobile devices to minimize interference among the expanding range of tightly packed frequency bands. For example, 15 to 50 acoustic wave devices are now commonly utilized in a high-end smartphone.
Surface acoustic wave (SAW) devices, key components in most smartphones, often utilize electrodes made of metal alloy to achieve a composite of desirable characteristics. For example, AlCu is a common material for the interdigital transducer in a SAW device. Rich in aluminium, this alloy is hence endowed with the desirable characteristics of low electrical resistivity and low density for optimal frequency response. The role of copper, which accounts for only about 2 percent of the weight of the AlCu alloy, is to serve as a ‘glue’ between the aluminium grains to reduce hillock and void formation. This improves device reliability by inhibiting electromigration, which can move grain boundaries at high temperatures or high powers. Two percent is the sweet spot in copper concentration. Adding more copper will result in a rise in electrical resistance and too little (<1 percent) copper will render the copper ineffective.
Ferrotec’s new generation Temescal evaporators, featuring HULA (High Uniformity Lift-off Assembly) planetary domes, present the solution to the historical challenges on the deposition of metal alloys by either sputtering or co-evaporation. The double-axis planetary rotating wafer domes result in exceptional thickness uniformity within wafers and from wafer to wafer (see Figure 1). Reliable process recipes are further developed for precise stoichiometry control for the single-source electron-beam evaporation of metal alloys.
Figure 1. Ferrotec's Temescal UEFC-4900 electron-beam evaporator with HULA Planetary domes
Single-source evaporation, the approach we advocate for the deposition of many metal alloys, has been historically overshadowed by sputtering and co-evaporation. In the silicon IC industry, sputtering of AlCu is especially popular. However, unlike electron-beam evaporation, the non-directionality of sputtering makes it an inherently poor choice for the metal lift-off process commonly adopted in the compound semiconductor and acoustic wave industry. One of the biggest drawbacks of non-directionality is that it gives rise to sidewall coverage of metal on photoresist. After the lift-off process with the removal of photoresist, metal ‘wings’ are left loosely attached to metal features. The resulting yield loss from shorting and defects caused by these metal ‘wings’ is a particularly onerous problem for the tightly packed interdigital transducer fingers in acoustic wave devices.
Target erosion is another challenge engineers must contend with throughout the lifetime of the sputter target. Since the shape of the target changes over time, the deposition rate for a given process recipe will drift over time as well. As a consequence, sputter process engineers resort to continuously adjusting the process time and/or target power in order to control the deposition rate (and thickness) from wafer lot to wafer lot. Automated adjustment to the changing deposition rate has been recently developed to reduce the significant down time of sputtering systems and the expenditure of engineering resource (as well as test wafers) caused by the manual adjustment process. However, implementing the automated adjustment requires first monitoring the drift in thickness throughout the lifetimes of several targets, and must be done independently for each target type.
In contrast, Ferrotec’s Temescal evaporators actively and continuously monitor the deposition rate to facilitate efficient production. This enables the user to simply set the deposition rate in the recipe. The execution of an automatic process parameter feedback loop will maintain the rate at the set point. As an option, our electron-beam source could also be replenished by wire feeding under high vacuum in the source chamber. The frequent wire feeding or source replenishment maintains the source in the same state and gives rise to very consistent run-to-run repeatability.
Co-evaporation is sometimes employed to simultaneously deposit different materials from two separate electron-beam sources. However, this automatically necessitates the placement of either one or both sources somewhat off the optimal central axis of the dome sphere. The displacement from the central axis will then distort the orthogonal directionality from the source to the substrates and may cause yield loss in the metal lift-off process.
Another drawback of co-evaporation is the difficulty in process control. In the example of AlCu (98 percent/2 percent by weight), the large ratio of aluminium versus copper means that for a total film thickness of 200 nm the effective copper thickness needed is only 1.2 nm. If the aluminium is deposited at 1 nm/s, the copper needs to be evaporated at an exceedingly low rate of 0.006 nm/s. This makes the repeatability of AlCu co-evaporation very challenging to control. Furthermore, in order to ensure a consistent stoichiometry throughout the film, growth rates must not change significantly during the deposition process. Otherwise, rate variation in one source would necessitate the rate alteration of the second source to mirror the change.
The new generation of Temescal evaporators avoids all the above-mentioned issues from sputtering and co-evaporation. Ferrotec’s HULA evaporators enable uniform deposition from a single AlCu source located at the very centre of the dome sphere for orthogonal directionality and excellent lift-off yield.
A set of planetary rotating wafer carrier domes change the location of each wafer relative to the evaporants in the source crucible continuously to ensure the utmost within-wafer and wafer-to-wafer uniformity of both thickness and composition via a contactless magnetic drive that generates no particles. The planetary domes are both self rotating and rotating around the central axis to provide each wafer with the same exposure to the source in distance and angles over the evaporation period.
All of these benefits are highly affordable. At about a quarter of the cost of a cluster sputtering tool, Ferrotec’s UEFC-4900 evaporators deliver better lift-off and deposition control. Furthermore, they offer a far higher throughput than a single-wafer, six-chamber sputtering tool. In the case of an AlCu metal alloy with a high concentration ratio, single-source evaporation is better for the stoichiometry control and overall cost of ownership compared to co-evaporation (see table 1 for an overview of the pros and cons of sputtering, co-evaporation and single-source evaporation).
Single-source evaporation has been used for many years for the deposition of NiCr and AuGe. One of the challenges of this approach is that the elements in the alloys can have distinctly different vapour pressures at a given temperature, leading to the formation of films with a notably different composition ratio from the starting source material. This difference is resulted from the difference in the partial pressures of the materials when mixed together at the evaporation temperature – these pressures can even differ from the partial pressure of the elements in their pure forms.
Fortunately for commonly evaporated alloys, either the partial pressures are not too disparate from one another, or there is a wide tolerance on the film’s acceptable composition ratio.
The good news for AlCu is that the vapour pressures of the individual elements are relatively close over a wide range of temperatures (see Figure 2). These curves suggest that an AlCu source consisting of 98 percent aluminium and 2 percent copper (by weight) would initially produce a film of about 99.3 percent aluminium and 0.7 percent copper. That’s not far from the composition obtained in practice, which has a typical copper concentration of 0.4 percent.
Figure 2. The vapour pressures of aluminium and copper are plotted in log scale as a function of temperature.
Obviously, in order to deposit a film with 2 percent copper by weight, the source must have a significantly higher copper concentration. It is quite feasible to determine the exact composition – modelling source behaviour using vapour pressure curves with a couple of empirical fitting terms gives a good indication of the relationship between the composition of the source and the resulting film.
Selecting the source’s right initial composition is only the starting point for ensuring the desired film composition. That’s because during evaporation, aluminium evaporates more readily than copper, leading to a gradual change in source stoichiometry. This deviation is exacerbated if the crucible is replenished after partial depletion with material of the same copper concentration as used initially (see Figure 3, which shows the modelled results).
Figure 3. The undesirable effect of replenishing the crucible pocket with material of the same copper concentration as used initially.
The solution is to replenish the crucible pocket with an alloy composition that maintains the intended copper composition in the films. It is possible to understand the relationship between the weight of the source and the copper content in films as the source is regularly replenished with the alloy of correct concentration (see Figure 4).
Figure 4. Frequent source replenishment with the AlCu of the correct composition minimises variations in the weight percentage of copper in AlCu films.
Although it is not possible to eliminate the variation in the copper content completely as material is evaporated from the source, these deviations can be controlled to within a relatively tight range by frequent source replenishments.
Table 1. Comparative analysis on the deposition approaches of metal alloys in a lift-off process for 6 chamber single-wafer sputtering and Ferrotec’s UEFC-4900 (for either co-evaporation or single-source evaporation).Location, location, location
Ferrotec is not the first company to launch a planetary evaporator. A number of them were already on the market in 2011, when we launched our new generation of evaporators that feature the HULA planetary domes.
However, one critical difference is that the source crucible is placed at the centre of Knudsen sphere in HULA evaporators instead of the edge. A Knudsen sphere is characterized by uniform thickness across any point on the sphere from an evaporating source placed on any fixed point on the sphere. The central location of the source in the HULA evaporators facilitates orthogonal deposition and higher lift-off yield.
With Ferrotec’s HULA systems, process engineers can typically achieve within-wafer uniformities for elemental metals of 2 percent, and wafer-to-wafer uniformities of 1 percent. The planetary HULA evaporators particularly excel at the deposition of ultra-thin film (5 nm or less) and typically the uniformities for platinum and nickel are better than 0.5 percent. In the case of AlCu (98 percent/2 percent by weight), the standard deviation of the run-to-run copper concentration is 0.2 percent (see Figure 5), and the within-wafer and run-to-run thickness uniformity are about 1 percent.
Figure 5. AlCu run-to-run stoichiometry control with a refill of the source crucible before the 11th run.
The high degree of uniformity, in both composition and thickness, results from the planetary motion of the HULA domes. The pair of elements that make up an alloy source evaporate with different angular distributions, and in the case of AlCu, copper varies more rapidly with angle than aluminium. With single-source evaporation of an alloy, it is not possible to use a different uniformity mask for each element. Consequently, a traditional single-axis system could only maintain a specified aluminium-to-copper ratio over a limited range of deposition angles (see Figure 6 for details). This greatly limits the number of wafers that can be coated at a time.
Figure 6. The copper concentration in an AlCu alloy (98%/2%) decreases with evaporation angle in a single-axis dome.
A great strength of the HULA system is its planetary motion. Wafers that could only reside in the richer or poorer copper regions in a single-axis system can now share time in both regions. So advantageous is this location-averaging approach that it minimizes the deviation in within-wafer composition to typically below the limit of detection. Simulation from vapour cloud models further confers that the variation is less than 0.1 percent.
The capability of Ferrotec’s Temescal evaporators is also underlined by calculations of the variation in composition of the AuSn eutectic alloy. According to our calculation, the standard deviation in uniformity of AuSn (80 percent/20 percent) is as low as 0.02 percent within-wafer, even without the use of a uniformity mask. The thickness variation is also impressive, with wafer-to-wafer variations for both gold and tin of 0.3 percent, which ensures a highly consistent stoichiometry. This attribute is valued highly, because a 0.75 percent reduction in tin concentration could increase the eutectic melting temperature by 30 °C. With Ferrotec’s HULA system, minimising variations in composition ensures uniformity of the eutectic melting temperature, and ultimately plays a significant role in realising high bonding quality.
In summary, two key innovations are incorporated into Ferrotec’s Temescal electron-beam evaporators. The first patented technology is double-axis planetary rotating domes with the source crucible at the centre of the Knudsen sphere for orthogonal, directional evaporation.
The second patented technology is the conic chamber that takes into account the shape of evaporated vapour cloud and minimizes the volume and surface area of the product chamber to reduce the pump down time. The typical pump down time to 1 x 10-6 Torr for a Temescal UEFC-4900 with 25 6-inch wafer batch size is 18 minutes and for a Temescal UEFC-5700 with 42 6-inch wafer batch size is 11 minutes. Ferrotec’s Temescal systems are therefore the best choice for the deposition of a metal alloy in a lift-off process.