3-D X-ray Inspection has Arrived BY DAVID BERNARD, Ph.D.,Dage Precision Industries
Advanced 3-D packages are replacing leadframe packages with the emergence of more subsystem integration, strengthened by advances in wafer-level technology. System designers prefer using completely tested subsystems and have encouraged the development of system-in-package (SiP) device integration. Board-level assembly requires surface mount technology, which means that packages need to interface with solder land pads or solder bumped pads. Multiple stacked-die packages, such as package-in-package (PiP) and package-on-package (PoP) (Figure 1) meet this demand for greater circuit density and improved electrical performance.
These complex 3-D package types are entering into mainstream production, and typically contain multiple stacked die with multi-level wire bonding, or wafer bumping, internal to the device.
 Figure 1. Variations of advanced 3-D package technologies. (Courtesy of STATS ChipPAC) |
null
Package Complexity
3-D packages place a greater demand on any required X-ray inspection and provide unique challenges for package inspection and process qualification during the package, assembly, and test operation. Traditional analysis using 2-D X-ray imaging is often limited with these new package types since all layers within the device are seen at the same time. This can be analytically confusing with the multiple dies and multiple layers of wire bonds appearing to overlap each other in the image, as shown in Figure 2.
Computerized Tomography
Computerized tomography (CT) is an imaging method whereby computational geometric processing is used to generate a 3-D model of an object from a large series of individual 2-D X-ray images taken around a single axis of rotation. Since its invention in 1972, CT has aided medical diagnosticians around the world to predict, diagnose, and treat disease. This same technique is being applied to inspecting advanced 3-D packages, because of 2-D X-ray imaging systems’ limitations. Specifically, CT is increasingly being used to inspect for die-attach quality, and the quality and effectiveness of wire-bond integrity within complex 3-D packages. Until recently, the use of CT for semiconductor and 3-D package inspection has been hindered by its slow computer processing times, low resolution, and expense. While some CT systems have been used for semiconductor inspection, they typically have not delivered the analytical performance required for such critical, high-density applications. However, advancements in CT have improved imaging speed, increased resolution to allow complete analysis of detailed package features, and decreased the price. Therefore, computerized tomography is now an ideal inspection methodology for complex 3-D packages. The resulting 3-D model can be viewed with real-time manipulation so that interconnections normally obscured by other joints or components within the package can be diagnosed, assuring complete package inspection (Figure 3).
 Figure 2. Complex 3-layer die and wire-bonded 3-D package. |
A CT model is developed computationally from a series of 2-D X-ray images taken as the sample, or semiconductor device, is rotated in an X-ray beam. The density variations within those images, and how their relative locations change as the sample rotates, are evaluated in a computer to reconstruct a 3-D model of the sample which can then be viewed and manipulated – providing analytical images, or slices, through any 2-D plane in the object CT model.
Imaging Requirements
The critical elements of producing a CT model for analysis include acquisition of the necessary 2-D X-ray images, computational reconstruction from those images into a CT model, and, finally, visualization and manipulation of the final 3-D model. A series of 2-D X-ray images are taken over a 360° rotation; taking a greater number of steps for each degree of rotation, and therefore a greater number of 2-D images, will provide a better CT model.
 Figure 3. Two-layer stacked-die device. |
Image acquisition is the most time-consuming phase of the CT model production and is dependent upon the number of rotation steps used. For example, taking 720 X-ray images within a full rotation of the sample (2 images per degree of rotation) will take twice as long to complete as 360 images per rotation (1 image per degree). In addition, each 2-D X-ray image is produced from an average of a number of frames from the real-time image capture of the X-ray system. The number of frames averaged at each image step, and the time the X-ray system can acquire those frames, further governs the speed of the acquisition process. This can be particularly significant depending upon the type of image-capture device being used in the X-ray system. For example, taking an average of 32 frames per image (a reasonable trade-off between good final image quality and throughput speed), a digital image intensifier acquires 32 frames in 1.3 seconds as it operates at 25 frames per second (fps). In contrast, 32 frames at an acquisition rate typical for the alternative to a digital image intensifier, the flat-panel detector at 4 fps requires 8.0 seconds to complete. Therefore, a flat-panel detector will take 8.0 sec., or six times longer per step. Reducing the number of steps and/or reducing the frame average per step will speed up the acquisition process but will reduce the eventual reconstruction quality of the 3-D model.
 Figure 4. CT image of 3D packaged device. |
Precision of the sample rotation during the image acquisition process is crucial for achieving good reconstructions at high resolution. Without this, the ability of the CT reconstruction algorithm is severely compromised because the math requires that density difference within each 2-D image be tracked precisely as the sample rotates. The increased availability of digital X-ray detectors in modern X-ray systems provides enhanced greyscale sensitivity in the 2-D X-ray images compared to earlier systems, which, in turn, provides a greater amount of subtle density-difference information that can be used during the reconstruction process.
For many CT systems, the CT volume reconstruction process is also a substantially time-consuming element. Vast ‘number crunching’ and computer processing time is needed to calculate the density variation of each pixel in each step image and work out how that density variation moves as the sample rotates. In this way the sample volume can be mapped out within a 3-D information matrix.
The resolution of the reconstructed 3-D matrix is generally defined in terms of volume pixels, or voxels. The more voxels, the more required processing, and the better the quality of the final 3-D model. Typically CT systems can have either 256 × 256 × 256 (2563) voxels, 512 × 512 × 512 (5123) voxels or 1024 × 1024 × 1024 (10243) voxels. It should be remembered that a 2563 voxel array contains 8× less information than a 5123 array and 64× less information than a 10243 array. For many CT systems substantial additional time is needed to produce the CT reconstruction because it is computationally hungry and is usually undertaken after completing image acquisition. With a single PC processor, necessary calculations are undertaken serially. If the model requires more voxels, there is a substantial and dramatic increase in the time required to achieve it. However, new CT systems are available that minimize the time required for the reconstruction process by feeding the acquired 2-D images into a dedicated reconstruction server as soon as they are produced. Such systems also have dedicated parallel processor boards to manipulate huge quantities of data as fast as possible. Therefore, these new systems have no real additional time penalty for the CT process when a 5123 voxel reconstruction model is required compared to a 2563 model. With a serial approach this would not be the case. In this manner, high-resolution CT models are available for viewing within moments of the image acquisition being completed. (Figure 4).
In addition to analysis of die-attach quality and wire-bond integrity within complex 3-D packages, many of these CT systems can also perform high-resolution 2-D inspection functions. Therefore, the user has a common inspection platform that has the flexibility to rapidly and easily convert their X-ray system from 3- to 2-D mode to satisfy manufacturing inspection needs. The latest X-ray tube advances means that 2- and 3-D analysis is possible with the greatest grey scale sensitivity and feature recognition down to as little as 250 nm (0.25 µm).
Inspection Capabilities and Model Visualization
The final, and vital, element of the CT process is the advanced processing software necessary to manipulate the reconstructed 3-D model and facilitate visualization of the necessary slices for the correct analytical view. This critical step allows the user to see the density contours within the sample CT model and change the viewed slice. The latest CT systems provide hardware accelerator graphics cards in the PC to specifically handle the manipulation and rendering of the CT model in real time. All of this can be viewed off-line of the image acquisition; allowing analysis of one model while the 2-D images of the next sample are being acquired. In this way individual slices can be viewed through any plane in the model to provide complete visualization of all stacked die, individual bumps, and multiple wire bonds within an advanced 3-D packaged device as shown in Figure 5.
 Figure 5. Individual layers of stacked die device. |
The visualization software functions include image manipulation – rotate, pan, and zoom; clip to eliminate unwanted detail; volume rendering for creation of solid surfaces and maximum intensity, and to vary the opacity and color of volumes; slices – multiple cross-section at any orientation and the ability to vary the opacity and color of individual slices; illumination models – a single light source or two light sources with shadow computation for optimal 3-D perception; image save for saving 2-D images in JPG or TIF formats; creation of movie clips allowing a ‘flying trip’ through the model; and measurement functionality – for example when needing to inspect wire-bond loop heights.
Conclusion
Recent advancements in computerized tomography make it an ideal technique for inspection of advanced 3-D packaging because it allows complete viewing of interconnections within the package, which otherwise may be obscured by other joints when seen in 2-D X-ray inspection images. Fast CT volume reconstruction carried out at the same time as image acquisition allows complete package inspection, assuring die-attach quality, wire-bond integrity, and improved package performance.
DAVID BERNARD, Ph.D., product manager X-ray systems, may be contacted at Dage Precision Industries, Inc., 48065 Fremont Boulevard, Fremont, CA 94538; 510/683-3930; E-mail: [email protected].