To form the best films, users of electron beam evaporators must understand how uniformity is influenced by the number of axes for substrate rotation, the insertion of shadow masks and the distance between source and substrate.

BY PHILIP GREENE FROM FERROTEC, TEMESCAL DIVISION

In very general terms, compound semiconductor devices are all fabricated in the same way. Whether it is LEDs, lasers, solar cells or electronics, production always begins with deposition of epitaxial layers onto a substrate, followed by processes that add metal films and contact wires to construct a chip that is housed in a suitable package.

A standard technique for depositing a metal on a compound semiconductor is electron beam evaporation. It may be performed as part of a lift-off process, which involves deposition on a patterned sacrificial layer that is then removed toleave a metallic film on part of the surface.

Strengths of electron-beam evaporation include a high effective deposition rate, which enables acceptable deposition times, and a highly directional source that limits detrimental sidewall coverage. Very high-quality films are possible by carrying out electron beam evaporation under a high vacuum: This leads to very low levels of incorporation of background gas contaminants, while scattering of the evaporated materials is minimised, thereby maintaining the highly directional nature of the evaporated species.

The directional nature limits photoresist via sidewall coverage when the substrate is oriented perpedicularly to the evaporant flux. However, a substrate oriented perpendicularly to the evaporant flux only receives truly normal incident flux at its centre, with some angular variation of the incident flux across the width of the substrate.  This can be kept within a desired range by placing it at an appropriate distance from the source. The greater the distance, the smaller the deviation from normal incidence, with a lower deposition rate as one price to pay. Although the rate can be cranked up by increasing the source power density, there are material-based limits to what is practical. Unfortunately, the nature of the process is such that when an array of perpendicularly oriented substrates are placed at a fixed distance from the source, the deposition thickness will not be inherently uniform, but have some variation in thickness across and between substrates at different locations.

At Temescal, a division of Ferrotec based in Livermore, California, we have devoted much effort to exploring the critical interrelationships between electron beam system architecture, material behaviour and process methodology in order to devise approaches that lead to higher uniformities for electron beam evaporation. Our success includes the development of a process methodology that could reduce losses from uniformity masks and offer significant improvements to collection efficiency (see Compound SemiconductorMarch 2013, p. 29). We have also explored the more general issue of attainable film uniformity, and here we consider what can be achieved with the historical techniques and the newest methodologies available.

Realising a highly uniform film is not just an end in itself – it is critical to several processes, including the formation of eutectic films used for bonding.  In these films, small variations in the ratio of materials can lead to a rise in melting temperature, and ultimately be the source of bonding issues across the substrate. That’s because composition variations have to be addressed by turning to a higher temperature to ensure melting across the entire film. The composition variation also has the unwanted side effect of possible changes in film morphology upon cooling.

Users of electron-beam evaporators face many potential sources of non-uniformity, and it is not possible to cover all of them here. So we will focus on the main factors influencing uniformity, and assume that film deposition is carried out using a well-controlled evaporation source, housed in a tool with tight mechanical tolerances relating to its geometry.

One factor having a big influence on film uniformity is the shape of the vapour cloud. Its impact on the uniformity of films deposited on 150 mm wafers is quite different in two of our tools designed for high-volume production: the Temescal FC-4400, which features a wafer carrier with a single axis of rotation that holds up to 30 wafers in three radial tiers; and the Temescal UEFC-5700, a tool with dual axes of rotation that either accommodates: 36 wafers in six domes, each holding six wafers; or 42 wafers in three domes, each holding 14 wafers.

Both tools have similar distances between the electron-beam source and the substrates of about 42 inches.

Vapour clouds

Many factors influence the distribution of the evaporant flux, including the inclination angle and azimuthal angle, the power provided to the source material and its evaporation rate, beam cross section and density (size) and the height and shape of the source.

We have worked hard to minimise the influence of all these parameters by designing the hardware and refining the process so that it provides the most stable, consistent conditions.  One of our insights is that it is critical to get high levels of uniformity at the source, because any variation here can carry over into non-uniformities for the vapour cloud. Manufacturers of electron-beam deposition tools often mount the substrates on a rotating dome-shaped carrier to minimize the influence of asymmetries in the azimuthal angle. By taking this approach, substrates experience an average of the deposition rate across all azimuthal angles, and thus have a smooth film thickness profile, with high values in the centre and lower ones towards the outer edges of the dome.

Engineers will often curve-fit the shape of the thickness profile observed across the wafers and carrier to better predict and understand the influence of measurement errors and small random fluctuations in conditions. One curve function offering a close fit to the observed thickness variation is the cosine power law – using this, one can predict thickness resulting from changes to the hardware, not including measurement error. Here we report the results of the application of that methodology to predict the best uniformity that can be achieved under realistic conditions.

The shape of an azimuthally symmetric cloud of gold vapour emanating from a 6 kW source is shown in Figure 1. This plot reveals the effective deposition rate across a substrate carrier placed normal to the flux. Regardless of source-substrate distance, the deposition rate varies with angle and is highest directly above the source. Given this, one could say that evaporation from the source is ‘directed upwards’, with a degree that changes significantly with different evaporants − and may even change significantly via modifications to the evaporation rate of a material.



Figure 1. Growth rates of metals in an electron-beam evaporator decrease with distance from the source to the substrate, and vary with angle

Rotation about one axis…

Over the years, we have expended considerable effort in documenting vapour cloud profiles for numerous materials over a range of process conditions. In this article, we draw on a small fraction of this database to highlight the sensitivity of the system. For example, the choice of material has a big impact on film thickness variation across all substrates: it can range from +/-12.8 percent to +/-24.0 percent in a tool employing a standard single rotation axis, and featuring a lift-off dome configuration (see Figure 2).



Figure 2. In a single-axis rotating dome, deposition without a mask produces variations in metal film thickness of +/-12.8 percent to +/-24 percent, depending on material

 

Adding one or more ‘shadow masks’ to block some of the evaporant flux is a conventional approach for increasing the uniformity along the radius of the dome (see Figure 3 for an illustration of this approach). The evaporant flux is greatest towards the centre of the rotating dome, so shadow masks are carefully shaped to block more of the flux heading towards the inner portion of the dome, compared to that travelling to the outer regions.

With a fixed-position shadow mask, each and every material, regardless of deposition conditions, gets the same correction for relative thickness.  The improvement can be far from modest – see Figure 4 to gauge the benefit wrought by the addition of a shadow mask designed for the average thickness profiles from the data in Figure 2.

The improved thickness uniformity with the addition of one shadow mask varies from material to material, from a low of +/-0.9 percent to a high of +/-6.9 percent.  These results highlight that awell designed mask can tune the uniformity for a given process condition to exceedingly tight standards. However, although these gains are significant, there is a limit to what a single mask can produce, in terms of tuning the uniformity for multiple processes. The obvious solution – and one that has been applied when there is a select set of processes with critical uniformity tolerances – is to work with multiple shadow masks, which are moved into and out of the evaporant cloud at appropriate times. However, it is often impractical to do this for every material, so process engineers only select mask designs for the most critical materials.



Figure 3. Inserting a shadow mask can improve thickness uniformity in an electron-beam evaporator



Figure 4. In a single-axis rotating dome, adding a shadow mask improves thickness uniformity.

Note that the deposition conditions used to generate the plot shown in figure 2 are repeated, but with the addition of an optimised shadow mask. The vertical axis has been changed in order to better display the data

Complicating matters, associated with every process is a normal variation in the shape of the evaporant cloud. The shadow mask is a fixed component in the system, designed for a particular process (material and rate), and so its correction cannot always match actively changing conditions. The degree of variation strongly depends on the material, although it can be restrained by employing identical source conditions.

So far, we have only presented plots of film uniformity based on representative deposition conditions for each material. When these deposition conditions vary, thickness profiles change. An example of the extent of this change is illustrated in the variations in growth of titanium films deposited in six identically configured evaporation systems (see Figure 5 for changes in radial thickness profile.) By repeating the same deposition process across six tools we explore the influence that variations in deposition conditions have on the uniformity. These results demonstrate that repeatability of the deposition profile can be kept in the

+/-1.5 percent range for this practical range of deposition conditions.  The range includes some variation caused

by source depletion from within the crucible over the course of multiple runs, as well as other small, unintended differences.

 

Figure 5. Deposition conditions vary from run to run and evaporator to evaporator. These variations, which impact the thickness profile, are considered in this plot that shows deviations in thickness profiles over multiple runs from six evaporators. Note that to remove the slope across the dome – an inherent feature from a mask that was designed to balance the results from two materials – data were normalized to remove that slope from the average profile

Changes to the height of the evaporant surface also produce variations in the system, but in this case they are independent of material type. To some extent these changes are unavoidable, because as material evaporates from a source, the height of the evaporant will fall. The impact on uniformity can be significant when uniformity tolerances are tight. For example, changes in uniformity of +/-1 percent and a change in the outer position thickness of nearly 2 percent can occur when the source height shifts from 0.5-inch above to 0.5 inch below the nominal evaporant surface (see Figure 6). This is a realistic scenario for several common sources.  Processes that require a tighter range on uniformity cannot allow that large a change in source height, so must break more frequently for source replenishment to maintain an even height.



Figure 6. Even variations in the height of the source produce differences in film thickness. This plot show the variations in a single axis rotating dome with a mask, with source heights of + 0.5 inch (blue line) and -0.5 inch (red line)

While repeatability in the +/-1.5 percent range is representative of the exceptional performance capabilities of a well-tuned precision system like the FC-4400, it serves as a noteworthy benchmark for conventional methodologies when compared to the levels of uniformity achievable in the UEFC-5700.

... and two axes

To minimize the inclination angle non-uniformity, we have developed the high uniformity lift-off apparatus (HULA).  These patent-pending tools feature a second axis of rotation, which greatly reduces the range in non-uniformity prior to the addition of a shadow mask. 

A tool that we have already mentioned in this article, the UEFC-5700, has a HULA configuration (see Figure 7). This tool features six small domes, which each spin about their own axes while rotating about the tool’s central axis. Rotating the substrates in this manner has several major benefits: it increases collection efficiency by reducing the need to turn to a shadow mask to increase uniformity; and it greatly reduces sensitivity to material type and process conditions, thereby enhancing uniformity. 

 

Figure 7. Higher levels of film thickness uniformity are possible by switching from a single-axis tool to one that provides dual axes of rotation. An example of the latter is the Temescal UEFC-5700, which accommodates 36 wafers in six domes, each holding six wafers

With the addition of the HULA motion, calculations indicate that the ranges for unmasked thickness profiles of different metals plummet to just  +/-1.74 percent to +/-2.36 percent (see Figure 8). The tremendous reduction in overall range for each material and the differences in ranges between material types go hand-in-hand with similar reductions in sensitivity of the uniformity profile to run-to-run variation and source height variation.  Those variations that caused shifts in uniformity in a single-axis tool of +/-1.5 percent are reduced in their influence to under 0.2 percent. The upshot of all of this is a system that delivers very robust uniformity characteristics across a wide range of materials.

Further improvements to thickness variation are possible with the addition of a very small shadow mask (see Figure 9). Using one designed for platinum film deposition, thickness variations are of the order of +/-0.5 percent – values so small that the metrology used to measure film thickness becomes very challenging, and the variations in film thickness may exceed the practical measurement accuracy commonly available.



Figure 8.  With the high uniformity lift-off apparatus (HULA), which incorporates a second axis of rotation, unmasked thickness profiles are just  +/-1.74 percent to +/-2.36 percent. Note that this plot displays the uniformity across the wafers in the radial direction for the same set of material conditions as shown in Figure 2. The scale of the vertical axis has been reduced to better display the range in thickness.  Since the wafers in this design are held in a single tier, the uniformity is representative of the shape for all of the wafers

We have also compared the calculated results for titanium (shown in Figure 8) with measurements on 20 nm-thick titanium films. Comparisons were made by plotting deviations from the average thickness for each substrate, and measuring the titanium film thickness with an optical technique that is sensitive to tiny variations in average thickness over the measurement area. There is good agreement between calculated and measured values, but some discrepancy at levels near +/-0.2 percent (see Figure 10). In this thickness range a 1 percent variation in thickness is comparable to the diameter of an atom, and this level of uniformity indicates a remarkably uniform distribution of atoms across the substrate.



Figure 9. Adding a mask for platinum deposition to a HULA apparatus trims the thickness variations to around +/-0.5 percent



Figure 10. The calculated values can be compared to experimental values

Some of these measurements reveal a small decline in thickness between the inner and outer portions of the wafer. However, there is also other structure apparent, which varies from substrate to substrate. Although these features will partly be caused by measurement uncertainty, it is not possible to rule out substrate mechanical positioning as a contributing factor behind the thickness variation. When films are this uniform, variations in thickness that result from an angular tilt to the substrate or a radial displacement of just 0.04” become apparent.

Wafer-related limitations

Another impediment to realising a perfectly uniform film is the flatness of the substrate. For lift-off, the substrate is oriented in such a manner that at its centre, the direction of the impinging incidence flux is normal. Consequently, deposition at the edges of the substrate occurs at a small angle from normal. Making matters worse, the distance to the source is larger at the edges of the substrate than its centre. This difference is magnified as wafer diameter increases, but can be reduced by increasing the distance between the electron-beam source and the substrate – although this action has downsides, such as reduced deposition rates.

Inserting a shadow mask into the electron-beam evaporator can address variations in film thickness in the radial direction. However, in the direction perpendicular to this – the cross-wafer direction – the mask has a limited influence on reductions in film thickness.

Variation in film thickness due to flat wafers is also influenced by the number of axes of rotation in the system: the HULA motion without a shadow mask gives some slight improvement over the single axis motion without a mask, but the single axis and HULA results are effectively identical after the addition of a shadow mask (see Figure 11).



Figure 11. Without a mask, the addition of a second axes of rotation leads to a significant improvement in cross-wafer uniformity. But when the mask is added, this difference is minimal

These uniformities, and those revealed in other plots presented in this feature, indicate that users of electron-beam evaporators have entered a new era for process tolerances and expectations in a production tool. When a single-axis tool is built to tight mechanical tolerances and runs a well-controlled and stable process, select processes can produce a thickness uniformity within a range of +/-1.5 percent on 150 mm wafers. The range of processes capable of delivering this level of uniformity expands with the addition of a second axis of rotation, which is found in the HULA tools.

For the deposition of eutectic films, this level of uniformity can mean the difference between an experimental process and a production wafer. But that’s not all, as thislevel of uniformity can also drive further manufacturing efficiencies, thanks to reduced product binning − the sorting of finished devices based on variations in performance characteristics, such asvoltage output, operational frequency and thermal tolerance. Such variations result from process variables and the inherent limits of each process, with a wider spectrum of product binning increasing parametric yield loss, which is the cost-per-wafer after sorting.

To increase uniformity to inside the +/-1 percent range requires the addressing of additional impairments to perfect film growth. This is hampered by wafer flatness, with success demanding even tighter tolerances on the mechanical systems that support and move the substrates. Programmes directed at uniformities of better than +/-1 percent must also consider whether the metrology is capable of accurately and consistently reporting the correct thickness.  This is particularly challenging when the goal is to achieve differences in thickness that are smaller than that of the typical substrate surface roughness, or in the case of thin films, where the difference can be less than the width of an atom.