The introduction of custom-designed uniformity masks can enhance lift-off processing systems, write Leszek Malaszewski, Ping Chang and Cris Kroneberger.

Physical-vapor deposition (PVD) is a technique for depositing thin metal layers on semiconductor wafers, and electron-beam evaporation is a method of PVD used more specifically for lift-off processes (Hill). In high-frequency applications, the need for metals that do not have high resistance surface oxides has led to the use of gold for interconnects, junctions and gates on compound semiconductors such as GaAs and InP (Wood). The difficulty in etching the gold (and the underlying metal layers) is avoided when patterning is generated by lift-off. In the lift-off process photoresist layers are patterned, exposed and developed, and PVD deposits metal across all of the wafer s surface. The unexposed photoresist, together with its metal covering, is then lifted off, leaving metal lines on the wafer. For a lift-off to be successful, material must not be deposited on the side-walls of the photoresist; this requires the substrate to be perpendicular to the impinging vapor stream. Metalization systems use rotating substrate holders to coat a large number of wafers during one process to maximize throughput and metal utilization. Because of the need for a perpendicular vapor stream, the substrate holder (known as a dome) rotates about the centerline of the evaporation source. To ensure perpendicular vapor deposition with respect to all substrates, this dome is a part of a sphere whose radius should equal the distance between the virtual source and the substrate (Hill and Chang). The typical system and component configuration for the lift-off process is shown in . Unfortunately the coating formed on a lift-off dome is not uniform in thickness. Flat or conical planetary systems are used for optical coatings and produce excellent uniformity with the vapor source being offset, but do not fulfill the requirement of normal vapor impingement. Consequently to ensure acceptable uniformity for all wafers, a lift-off dome arrangement must rely on a mask placed between the source and the dome. Factors affecting coating thickness The general factors that influence thickness uniformity on a lift-off dome are:

the distance between source and substrate;

the rate of evaporation and angular vapor distribution;

whether a uniformity mask intercepts some of the vapor;

the relative motion of the substrate (substrate rotation is essential in every case).

The assumptions commonly used in modeling are that the vapor source is a point source (labeled M in ), and that the vapor distribution follows an approximation of the cosine rule. Accordingly, the coating thickness Tp at point P on the wafer surface is a function of u (the emission angle of the vapor stream from the source), f (the angle of incidence at the wafer surface) and the distance R between vapor source M and point P. The relationship between these is shown in . Factor n is directly related to the material s evaporation rate and system configuration. Point M is stationary with respect to the chamber, and point P rotates around the axis of the dome. The theoretical coating distribution across a lift-off dome is plotted in . When a large flat substrate is fitted on a spherical dome, the flatness of the substrate will further reduce the uniformity. Uniformity masks Because the vapor deposition profile is related to a cosine distribution, in a single rotation almost half of the evaporant flux must be interrupted towards the center line where deposition is fastest. This requires the mask to block the vapor in the areas where normally one would expect thicker deposition layers. Achieving uniformity thus reduces "collection efficiency". , also shows a comparison of the typical coating thickness distribution for systems with and without the benefit of masks. Computer programs have been created for uniformity mask development that allow the graphical representation and analysis of measured coating thickness distribution for each row of wafers on a dome. The programs calculate uniform thickness distribution and generates shape data for mask fabrication. Mask shape outputs are provided with X and Y co-ordinates, and as a polynomial fit for mathematical and statistical analysis. The enhanced mask profiles can be generated remotely from the installed system if the user supplies accurate data. The revised mask profile can then be sent by fax to be incorporated at the user s site. shows the coating thickness distribution with and without a uniformity mask for a dome configuration with two rows of wafers. Coating uniformity without a mask for depositing Ti was determined to be 2.44% for the first row and 5.67% for the second, giving a 3s uniformity of 15.73%. After installing a standard Ti uniformity mask, the coating process produced uniformities between 0.76% and 1.11% for both rows of wafers, and a 3s uniformity between 1.74 and 1.97%. Standard uniformity masks The standard domes used in Temescal evaporation systems can handle most wafers between 2 and 6 inches in diameter. Wafers can be positioned in 14 rows depending upon dome and wafer size. The shape and location of the installed mask have been optimized for systems with and without a collar. The collar is an extension ring inserted between the source and the substrate to increase the distance between them. A larger distance reduces the angle of incidence toward the wafer edge, improving the thickness uniformity. Each source-to-substrate distance needs its own radius dome. Each of five increasingly larger systems with a lift-off dome has a standard uniformity mask made for evaporating Ti. The standard mask can generate better than 4% uniformity for wafer diameters of 26 inches, regardless of their location on the dome. Conditions affecting uniformity Substrate rotation and the rate of evaporation of selected materials massively affect the uniformity of coating thickness in lift-off processes. Additional influences, which make the coating thickness distribution deviate from the simple cosine distribution model, include the following:

the emission angle of the vapor stream;

the obstruction of the vapor by crucible wall, shields or other elements;

the type of material used;

the level of material inside the crucible;

the formation of a vapor cloud, instead of a surface, that acts as virtual source.

These effects are minimal in well designed systems where coating is performed under stable conditions. Multi-layer coating uniformity Simple III-V compound semiconductor interconnects, junctions and gates consist of a thin titanium adhesor layer and a thin platinum or palladium diffusion barrier, which are finally followed by the thick gold conductor. The thickness uniformities required for each layer and for entire stacks are additional challenges for lift-off processes. The thickness and uniformity of every layer can be adjusted to accommodate the requirements for multi-layer stacks. Each metal s rate of evaporation affects a uniform distribution, but can be used to optimize the material evaporation process. However, the adhesor and diffusion barrier layers are thin, and even an absolute uniformity figure greater than a few percent may be acceptable. At 200 thickness, even a 6% uniformity represents only 12 or four atom layers. However, given that the uniformity mask "blocks" deposition for almost half a revolution of the dome, in the case of thin coatings rotation must be fast enough to ensure that each wafer is coated equally. For example, to achieve a coating thickness of 200 at a deposition rate of 10 /s (a total time of 20 s), deposition should take place over 10 revolutions to ensure adequate averaging. Requiring 10 revolutions in 20 s equates to 30 rpm. shows that the change in evaporation (and thus deposition) rate affects uniformity. Once a uniformity mask has been developed for a set of conditions, any change in rate can alter the uniformity and may need a modified mask. Temescal s system designers recognize this problem, and use a multi-mask philosophy in the case of tight specific uniformity tolerances for multi-stacks. Thus a set of specific uniformity masks is designed for each metal with well defined coating-process parameters. Each mask is sequentially moved into position as each metal is deposited. Though this system is more expensive and complex, the performance often offsets the cost with a higher production yield and every layer having an excellent even thickness of coating. The cheaper way of handling multi-layer metal deposition is to use one mask for all coatings that has been developed for the thickest or most critical layer in the stack. While the other materials are tuned and optimized for rate and dome rotation to ensure acceptable absolute thickness uniformity, the values may be higher if expressed as a percentage. Very tight uniformity requirements Coating processes can be improved to yield very tight thickness uniformity as long as they are stable and well optimized. Although standard uniformity masks provide an even 4% thickness for wafers 26 inches in diameter, they can be adjusted for the individual wafer diameter and for the specific wafer row as shown in . Square dots represent the required mask performance in relation to the specific wafers and row locations at constant coating process conditions. The thick line represents a universal polynomial shape that can accommodate any size at any row locations. As an example, a thickness uniformity of 1% was achieved with an improved mask for gold at a specified evaporation-source pocket size and controlled material level for systems in production at Infineon Technologies in Munich. Usually only two iterations are required to provide total uniformities in the 1% range. At Alpha Industries in Woburn, MA, the "no mask" uniformity for gold on 6 inch GaAs wafers held on a 27 inch diameter dome was 15%. After the program defined the initial mask profile, a simple mask of stainless steel foil was installed. The thickness spread was again measured on new wafers, and a second mask was made and installed. After this was measured it was found to be 1.25% well within the specification. Limitations in mask design Coating thickness data for uniformity mask development is based on measuring the actual coating thickness layers deposited during the coating process. Wafers must be properly oriented on the dome so that coating thickness distribution on the wafer is measured along the imaginary line extending from the dome center to the dome edge, passing through the wafer center. Well calibrated thickness-measuring equipment is fundamental for obtaining accurate coating distribution data. It is also necessary to measure film thickness in various locations on the wafers. Temescal recommends a 0.25 inch or 6 mm gap between locations. It should be noted that the dome is a segment of a sphere and that the wafer is a flat surface attached to the dome. The angle of incidence depends on the actual position on the wafer diameter, so the source-to-substrate distance is slightly different from wafer center to wafer edge. For the sake of a mathematical exercise, one could assume that the source-to-substrate distance R is equal to the dome s spherical radius. Localized thickness distribution on a single wafer is influenced by cos(f), where f represents spherical angle and the cos(f) variation represents the thickness uniformity limit. Conclusion Electron-beam evaporation systems are widely used for lift-off processing, and can be supplied with simple coating uniformity masks or custom-designed products to enhance their capabilities. The enhanced mask profiles can be generated remotely from the installed system, provided the user supplies accurate data. The profile can then be faxed to be incorporated at the user s site. Further reading R Hill 1986 Physical Vapor Deposition (Temescal, California). R Hill and P Chang 2001 Electron Beam Evaporation Part II Vacuum & Coating Technology 51. C Wood 2000 Hetero-Structures for High Performance Devices Handbook of Thin Film Devices ed. M Francombe (Academic Press, San Diego).