Before Thin Goes In: Inspecting Flip Chips

Where small size seizes the day—whether to balance SWaP for military, aerospace, and avionics applications or to pack more electronics in a smartphone or in IoT and wearable devices where space is at a premium, flip chips come in handy. Here’s how acoustic micro imaging tools get flip chips carpe diem ready.

Flip chip assemblies have taken advantage of the transparency of semiconductor silicon to ultrasound ever since flip chips became popular—and needed to be inspected nondestructively—in the late 1990s. These flip chips used capillary under-fill. Overmolding, if required, was added in a separate step.

Figure 1: Two small flip chips imaged acoustically through bare silicon (left) and through molded underfill (right).

Figure 1: Two small flip chips imaged acoustically through bare silicon (left) and through molded underfill (right).

Many flip chips still use this process. But in recent years, especially for very thin assemblies, a molded underfill has replaced the capillary underfill. The same material is used for the underfill and the overmolding, which are applied in the same step. Molded underfill is widely used, for example, in very thin flip chip assemblies used in mobile phones. Three advantages make molded underfill attractive in this and other applications: fewer overall processing steps, shorter assembly time, and cost reduction.

An acoustic micro imaging tool, such as those in Sonoscan’s C-SAM line, performs nondestructive inspection. The tool’s ultrasonic transducer scans the surface of a sample at speeds that may exceed 1 m/s. In Time Domain Imaging, the most frequently used of several imaging modes, the moving transducer sends a pulse of ultrasound into the sample several thousand times each second and, a few millionths of a second later, receives echoes from a single x-y location at materials within the sample. Each echo provides one pixel in the acoustic image of the sample. The highest-amplitude echoes come from the interfaces between a solid and air or another gas, and produce the brightest pixels in the acoustic image. These bright features identify the voids, cracks, delaminations, and other gap-type defects that inspection is looking for. Solid-to-solid interfaces without gaps are some shade of gray, while homogenous materials are black (no return signal).

Standard underfill materials and mold compounds are polymers that contain particles. When the material has cured, ultrasound pulsed into it is to some degree absorbed by the viscoelastic polymer and scattered by the particles, but the losses are generally compatible with successful imaging.

Molded underfills also consist of a polymer and small particles, but with a significant difference: the particles are much smaller, more numerous and are better at scattering ultrasound. Together, the absorption and scattering in molded underfill materials mean that a lower ultrasonic frequency may be needed to produce a good acoustic image.

To make the internal features of molded underfill flip chips more visible, Sonoscan has developed a family of transducers specifically designed for molded under-fill work, along with a variety of application techniques.

Sonoscan’s acoustic micro imaging tools use transducers that range from 5 MHz to 400 MHz in frequency output. Low frequencies penetrate deeper into materials, but create low-resolution images. High frequencies have less penetration but better resolution. Flip chips with no overmolding are typically imaged at 230 MHz to see tiny internal defects. If even smaller defects—minuscule voids in a solder bump, for example—are a concern, a 300 MHz transducer may be used, or even a 400 MHz transducer.

Direct Launch

Molded underfill, though, is at the other end of the scale. High frequencies won’t penetrate far into a molded underfill assembly. If a molded underfill assembly can’t be imaged with a 230 MHz transducer, a 100 MHz may have enough penetration to produce an image, or perhaps a 50 MHz transducer. What is needed is a transducer whose ultrasound can undergo scattering and absorption both traveling to a feature and returning as an echo. As long as the image displays internal structural defects as bright white features, the degree of resolution is relatively unimportant. Even lower frequencies, such as 20 MHz, might be used, but in flip chips the significant features are so small that a 20 MHz transducer might return an echo with no meaningful details.

Time Domain Imaging produced the images of the two flip chips in Figure 1, which are nearly the same size (length 1+ mm), but differ in their packaging. On the chip at left the back side of the die is exposed. There is no overmolding, so ultrasound can be launched directly into the silicon, which absorbs or scatters little ultrasound. The flip chip at right was packaged with molded underfill, so ultra-sound must be launched through the molded underfill on its route to the depth of interest between the die face and the substrate. The point of interest for both flip chips was the presence or absence of voids, delaminations and cracks. Return echoes were gated only on the depth between the die face and the top of the substrate; echoes from other depths were ignored.

The flip chip at left was imaged at the high resolution/low penetration frequency of 230 MHz. The launched ultrasound traveled with little loss or scattering through the silicon of the die, was reflected by features in the depth of interest, and returned as echoes through the silicon. The image shows that all solder bump bonds are intact, i.e., there are no ‘white bumps’ and there are no other defects.

Molded underfill was used to package the flip chip at right. This meant that a 230 MHz transducer would have no chance of penetrating to the depth of interest, and that a lower frequency was needed. From among the family of transducers that Sonoscan has developed to image through molded underfill, several were selected. Only one transducer succeeded. Surprisingly, its frequency was 120 MHz. One would expect that a lower frequency—likely between 50 and 100 MHz—would perform best, but none of the transducers below 120 MHz could make as clear an image.

This unexpected outcome may have resulted from factors other than frequency that go into the design of an ultrasonic transducer. A 120 MHz transducer, may have a focal length (the distance from the transducer to the depth of interest) ranging from 4mm to 12mm. A small focal length means that a pulse launched by the transducer will be absorbed less when passing through a thin layer of water. The design of a transducer also determines the transducer’s F-number (F#), which is a function of the transducer’s focal length and the diameter of the transducer’s piezoelectric element. A low F# does not improve penetration, but it does produce a small spot size and thus finer spatial resolution.

Figure 2 is the 75 MHz Time Domain image of a larger flip chip. The return echoes were gated, as is usually the case with flip chips, on the depth from the die face to the top of the substrate. The packaging material was molded underfill.

Figure 2: A larger flip chip imaged through molded underfill. The white feature is a series of voids, but all bumps are bonded.

Figure 2: A larger flip chip imaged through molded underfill. The white feature is a series of voids, but all bumps are bonded.

A Slump into the Void

The spatial resolution in this image is mediocre, but it is sufficient to demonstrate that there are no “white bumps” whose color in the image reveals that are disbonded from either the die face or the substrate. Some of the bumps are hardly discernible, but none approaches the brightness of the long white series of voids just to the right of the center of the image. These voids lie between the die face and the substrate, and are in contact with, or nearly in contact with, several bumps. The risk is that one or more of the bumps will gradually slump into the void and lose its electrical connection.

Figure 3 is the 100 MHz Time Domain image of one of many molded underfill-packaged flip chips still in strip form. A 100 MHz transducer with a 0.5 inch (1.27 cm) focal length was used. Gating was at the interface between the chip and the bumps. Three defects are visible:

• A bright white void just left of the center of the image
• A somewhat fainter void at the center of the top row of bumps
• An indistinct dark area just to the right of the void at the center of the image.

The dark indistinct area is hard to diagnose. It has the appearance of an acoustic shadow, which is typically caused by a feature such as a void that lies above the gated depth. Echoes returning from the gated depth are locally blocked by a void above the gate, with the result that the void is imaged as a black acoustic shadow. But this shadow transmits some of the ultrasound reflected from the gate. It is most likely a dense grouping of the small MUF filler particles, and might consist of particles displaced by the nearby void. Other flip chips from the same lot, imaged in the same way, showed similar features. This phenomenon does not seem to occur in conventional underfill, where the particle count is lower.

Figure 3: The two bright features in this molded underfill flip chip are voids. The nearby gray feature may be an inhomogeneity in the underfill material.

Figure 3: The two bright features in this molded underfill flip chip are voids. The nearby gray feature may be an inhomogeneity in the underfill material.

The images above were all made using the Time Domain Imaging Mode, but there several other imaging modes that can be used. One such mode is Frequency Domain Imaging (FDI). A pulse launched from a 100 MHz transducer will consist of a number of different frequencies, ranging roughly from 70 MHz to 120 MHz. As a given frequency travels through a material, it tends to be downshifted slightly. At each of the thousands or millions of x-y scan locations, Time Domain Imaging simply selects the single return echo having the highest amplitude and uses the amplitude to determine pixel brightness.

FDI uses a different approach. When it receives the echoes at a given x-y location, it selects perhaps 15 or 20 short frequency ranges (80-82 MHz, for example) and assigns a pixel to each one. This means, of course, that a single scan of a sample produces 15 to 20 images of the sample. The same material interface may react very differently to two different frequencies, and two images from adjacent frequency groups may differ strikingly in their appearance. Among the 15 or 20 images of a molded underfill flip chip there may be frequencies that display the desired internal features.

The methods described here make it possible to acquire, nondestructively, sufficient data about internal structural anomalies in molded underfill flip chips. The images may lack high spatial resolution, but they succeed in identifying flip chips likely to become field failures.

Tom Adams is a consultant to Sonoscan, Inc.

Share and Enjoy:
  • Digg
  • Sphinn
  • Facebook
  • Mixx
  • Google
  • TwitThis