Measuring Reflective Surfaces

Manufacturers are adopting automated optical inspection (AOI) systems based on phase shift profilometry (PSP) for applications in advanced packaging processes. Many of these processes use front end-like techniques to create connections among die within a package and from the packaged die to the outside world. The technique offers fast, precise measurements of the 10µm to 100µm features that are typical of these middle-end manufacturing processes. Some of the materials and features, such as silicon wafers, glass substrates, solder bumps and pillars, and more, have specularly reflective (mirror-like) surfaces that present special challenges. The first challenge is presented by multiple reflections between features, which introduce local errors in the measurement (figure 1- left). The second challenge is more fundamental. Conventional PSP measurements rely on the ability to see distortions in a pattern of varying intensity projected onto the measured surface. They assume the surface they are inspecting is diffuse, that is, when light hits the surface, it scatters in all directions. Specular surfaces do not satisfy this assumption and the projected pattern cannot be seen. Figure 1 – right illustrates the problem. The vertical fringe pattern, clearly visible on the diffuse surfaces, is invisible on the specular surfaces. To solve this problem, our latest NanoResolution MRS sensor includes an additional measurement channel design specifically for specular surfaces (figure 2).

Fig. 1: Specular reflections pose significant challenges to phase shift profilometry measurements. (Left) Multiple specular reflections among shiny features, such as solder joints, tinned leads, and metal oscillators, can cause distortions in the fringe pattern and errors in the height measurements. (Right) The projected fringe pattern is invisible on specular surfaces. The two round features are identical, except that the upper one has a diffusely reflective surface and is mounted on a similarly finished flat surface, while the lower one has specularly reflective surface and is mounted on a flat mirror. The projected vertical fringe pattern, clearly visible on the diffuse surfaces, disappears on the specular surfaces.

Fig. 2: The NanoResolution MRS sensor includes separate channels for diffuse and specular surfaces.

Suppressing Errors
Multiple local reflections may occur between components on the inspected object. If we are looking at a printed circuit board to which packaged devices are attached by specularly reflective soldered connections, we may see a glint where light from the projector reflects from one connection to another and returns to our camera. This may be only a nuisance for a visual inspection, but it can cause significant errors when trying to discern a projected pattern formed by point-to-point variations in diffusely reflected intensity. The high intensity glint originates from a different portion of the patter, and when overlaid on a neighboring area, it corrupts the pattern being measured there. Accurate measurements require the suppression of these multiple reflections, hence the name of our sensor – Multiple Reflection Suppression (MRS) sensor.

As we have discussed in previous installments of this blog, the sensor uses multiple patterns at different spatial frequencies to provide high sensitivity, extend dynamic range and eliminate ambiguity in measuring discontinuous fringes. It also uses multiple cameras to view features from all sides and eliminate blind spots in areas adjacent to tall features. These techniques also provide a means to suppress multiple reflections. Because the reflections are highly directional, they are unlikely to appear the same to all cameras. The reflections will also appear differently or not at all at different spatial frequencies, orientations, and compositions of the projected pattern (fig. 3). Careful analysis of the data can thus distinguish spurious multiple reflections from valid data points.

Fig. 3: Multiple reflection errors can be suppressed by analysis from multiple directions at multiple frequencies. Reflections that add coherently at lower frequencies become incoherent at higher frequencies, reducing fringe contrast but not affecting the phase measurement.

Measuring Specular Features and Surfaces
Measuring specular features and surfaces requires a different approach. Since the projected pattern intensity cannot be seen, we must treat the surface itself as an element in the optical system and look at its effect on the reflected light (fig. 4).

Fig. 4: Magnified views of the diffuse (left) and specular (right) features shown in figure 1 above. The vertical fringe pattern is projected from a direction roughly perpendicular to the mounting surface. The grid pattern on the left in the figure is printed on a flat surface that is also perpendicular to the mounting surface. The projected vertical fringes clearly reveal the shape of the diffusely reflective surfaces when viewed from a direction somewhere between the projection axis and the surface. However, the projected pattern cannot be seen on the reflective surfaces. Rather, we see the reflected image of the printed grid pattern. By analyzing the reflected light, treating the specular surfaces as elements in the optical path, we can characterize the shape, position, and dimensions of specular features and surfaces.

One of the critical measurements required for advanced packaging applications is the height of solder bumps and pillars used to make vertical connections between stacked chips or between a chip, a substrate or interposer. Both the bump and the surface around it may be specular reflectors and we need to determine the height of both to calculate the bump height above the surface. From a single measurement we cannot distinguish a change in height from a change in angle. From an analytical point of view, we require the solution of multiple equations in multiple variables. Using data from multiple cameras allows us to resolve the ambiguity.


Fig. 5: A VLSI step standard with mirror finish (left) and measurement results using the NanoResolution MRS sensor.

Engineers need to know not only the height above the surface but also the coplanarity, i.e. height relative to surrounding bumps. A bump too tall can degrade connections of nearby bumps, too short and it will fail in its own connection. Full inspection also requires detection of missing bumps and 2D measurements of size and position. With tens of millions of bumps per wafer, this is a lot of data, and keeping pace with the production process can be challenging. One approach to increasing speed is sampling, measuring only a fraction of the features and extrapolating the results to the entire population. But the cost of failure is very high when multiple know-good-die are put at risk by the failure of a connection. Moreover, manufacturers of systems with high reliability requirements, such as automotive systems where user safety is a paramount concern, often require inspection of 100% of the wafer surface. With tens of millions of bumps per wafer, measurement speed is essential. Our latest NanoResolution MRS sensor takes 2D and 3D measurements in a single pass and can inspect the full surface of 300mm wafers at rates as high as 25 wafers per hour.