Improved copper wire bonding with non-contact metrology

03/07/2012

MATT NOVAK, Bruker Nano Surfaces Division, Tucson, AZ,

Copper wire is fast gaining popularity for chip bonding, but it is a less understood and less mature process than bonding with gold. A 3D optical microscope can help provide data needed for imprint studies aimed at process optimization and sampling.

Due to the ever increasing cost of gold, combined with several performance advantages, copper has fast been gaining popularity as an alternative to gold for wire bonding in semiconductor packaging. However, the higher stiffness of copper means that the bonding process must use higher force, which can lead to pad damage problems and potentially reduced reliability. Until now, our industry has used the scanning electron microscope (SEM) to visualize these issues and thereby identify process problems and enable process optimization. But this is a very slow measurement methodology that only delivers a single cross sectional view of the bonded pad. Conversely, the optical microscope is often used to supplement these measurements with areal views of the entire imprint, but it does not deliver quantitative surface metrology data. This article shows why some companies are now switching instead to the 3D optical microscope to obtain the entire 3D profile of a bond pad in seconds, with a non-contact areal measurement that can be readily automated.

Copper bonding ??? the need for imprint studies

For the past 30 years, gold wire has predominated in the back end packaging semiconductor industry. Copper wire is fast gaining popularity for chip bonding for several reasons. First and foremost is the savings in material costs as the price of gold continues to increase at a dramatic rate. In addition, copper exhibits higher thermal and electrical conductivity, making copper bonded products more thermally tolerant than their gold counterparts, and hence more attractive for higher power applications in particular. Plus, these higher conductivities enable thinner wires to be used.

Another advantage of copper is that it is much less susceptible to forming intermetallic compounds (IMCs) during ball bonding to the aluminum pads. In a "good" bond, ideally over 50% of the ball's footprint should feature formation of a thin IMC interface layer. If the IMC coverage area is under 30%, the bond may be too weak and susceptible to lifting. But gold and aluminum very easily co-diffuse, which can cause excessive production of IMCs. This is undesirable as these are usually more brittle than Al or Au alone. Over production of IMCs (deeper co-diffusion) can thus further reduce bond strength in terms of resistance to thermo-mechanical stresses.

Finally, copper is more resistant to damage during looping for two reasons. First, its higher thermal conductivity limits the spatial spread of any phase transformations (i.e., changes in grain structure) during bonding. Second, copper is mechanically stronger (stiffer) than gold.

However, the fact that copper is less ductile leads to a major potential pitfall with using it for wire bonding; it takes more energy to create the bonds, i.e., higher ultrasonic power. It is better to store and bond the copper wire under a nitrogen atmosphere since it oxidizes much more easily than gold and the presence of oxide layers predicates even higher bonding forces. Unfortunately, the use of higher bonding force can lead to two potential negative consequences that can impact the quality, function and longevity of the bond and hence packaged chip: cratering and aluminum splash (or splash-out). Cratering refers to damage that penetrates through the pad and into the underlying dielectric layers. Aluminum splash occurs when the higher force used for copper causes the top aluminum layer of the pad to splash out beyond the ball footprint and potentially even beyond the edges of the pad.

In spite of these potential limitations, tremendous advances in understanding and optimizing copper wire bonding means that chip bonding yields with copper are approaching those with gold. But because of these limitations, copper wire bonding is a less understood and less mature process than bonding with gold. As a result, it requires more effort in process optimization and a higher sampling rate in order to maximize yield of high reliability packages. This is driving a fast growing need for a tool to quantitatively evaluate pad imprints. Moreover, a tool that can support higher sampling rates than those that have been traditionally needed with gold wire bonding is required.

Limitations of imprint evaluation methods

There are several tools that have been used to visualize and/or quantify bond imprints. Until now, the most commonly used have been the optical microscope, sometimes in a stereoscopic implementation, and the scanning electron microscope (SEM).

For optical microscopy, the pad imprint is exposed using a combination of sawing, polishing and chemical decapsulation. Using various types of illumination and contrast enhancements, high-resolution optical microscopy is excellent for generating color images of the imprint. Moreover, it can give very good quantitative estimates of the size and area of any damage sites or splash zones. But it does not provide quantitative data in the z axis and thus cannot be used to measure the depth/height of any damage. Therefore it cannot be automated to measure the volume of the splash, the depth of any craters, damage holes or voids, and cannot compute key statistical profiling of the pad. It can only be used to estimate some of these parameters in a time-consuming and subjective manner.

The SEM provides gage-capable quantitative data, but is limited in its ability to support not only copper but all wire bonding because it is very time consuming (over one hour for the sample preparation and measurement) and can only deliver a single cross-section view, i.e., 2D surface analysis. This measurement requires encasing the package in polymer and grinding the hardened polymer down to expose a single cross sectional view of the bonded pad. This is then placed in the SEM where an electron beam is scanned across the exposed pad to acquire a detailed image and thus an accurate transect. In turn, this can be used to compute a Ra average roughness value based on the single transect.

Instead, the packaging industry needs a tool that can provide areal analysis with true three-dimensional gage-capable measurements, to enable visualization and spatial analysis of the entire pad. Ideally, such data can then be used to automatically compute key statistical parameters including splash volume, maximum depth of any craters and other damage, and ISO-approved statistical roughness parameters such as S_a and S_q. In addition, the tool should provide high speed to enable dense sampling.

The 3D optical microscope

Based on a depth sensing technique called white light interferometry, the 3D optical microscope is a non-contact surface metrology tool. Equipped with a digital camera for image detection, this instrument looks very similar to a conventional optical microscope, and the device under test is placed on the sample stage. In operation, the 3D optical microscope then uses light waves as a high-resolution ruler or depth gauge. This allows the camera's computer to obtain a full 3D profile of the entire field of view defined by the camera and the microscope's objective as the sample is quickly and automatically stepped through focus.

In terms of performance, the z axis (vertical) resolution can be as high as 0.01 nanometers, whereas the resolution (and field of view) in the xy plane are determined by the magnification of the interchangeable objectives, as well as the size and number of pixels in the microscope's digital camera. The ability to optimize the field of view is an important benefit as it enables this to be matched to the size of the pad. The need to survey the entire pad is one of the reasons that this application cannot use an atomic force microscope (AFM), which is the highest resolution surface analysis tool and the instrument of choice for many industries and applications.

Figure 1. Examples of 3D images of pad imprints captured by a 3D microscope (Bruker ContourGT-X8) with height/depth represented here as false color. Image data courtesy of ON Semiconductor.

Two distinguishing features of this type of 3D optical microscope are its ability to acquire a comprehensive high-resolution 3D data set of the entire pad (Fig. 1) and its high speed. Because hundreds of thousands of pixels of data are recorded simultaneously, it takes only seconds to acquire the entire low-noise data set (Fig. 2), thus including complete spatial details of any cratering area and depth as well as aluminum splash volume.

Figure 2. Compared to the SEM, overall measurement time for the 3D microscope is much less.

Conclusion

Wire bonding is increasingly moving to copper instead of gold because of the relative cost of these materials. The migration has begun in earnest in logic and will soon be followed by MEMS and analog applications. But while copper avoids some of the drawbacks of gold, including its rapidly escalating cost, it comes with its own challenges and is considerably less mature. High-density pad imprint sampling is necessary to achieve the high reliability the packaging industry needs, which thankfully can approach that of gold. This need is well met by the 3D optical microscope.

Matt Novak, Bruker Nano Surfaces Division, Tucson, AZ, email: [email protected] www.bruker-nano.com

Solid State Technology, Volume 55, Issue 2, March 2012

More Solid State Technology Current Issue Articles

More Solid State Technology Archives Issue Articles