3-D X-Ray Inspection Looks Into the Void

Has 3-D X-ray PCB assembly inspection finally become a mainstream, inline production tool? The answer ranges from “it has been for some time” to “almost” and clearly depends on whom you ask.

There are many disparate factors shaping the most recent X-ray inspection developments including faster computing, better X-ray sources and detectors, more emphasis on cancer screening, and the miniaturization associated with personal communications devices. Designing consumer electronics that use multilayer fine-line PCBs with blind vias and components mounted on both sides is driven by demands for lighter and smaller portable products. It’s a high-tech solution to a packaging problem but not an approach that simplifies inspection.

X-ray imaging is the only practical way to determine the quality of the hundreds of hidden solder joints underneath a BGA device. But, because 2-D technology just measures the attenuation of X-rays passed through an object, it can’t distinguish between devices mounted on the two sides of a PCB. Their images interfere with each other.

Laminography is an early and limited form of 3-D X-ray technology that can form a distinct image of one side while ignoring the components on the other. In a laminography machine such as the recently discontinued Agilent 5DX, a single slice is imaged at the focal plane between the detector and source. All other points not lying on that plane are defocused and contribute to background noise.

More recently, 3-D X-ray processes have progressed largely along two lines. The first, computed tomography (CT), was developed in the early 1970s as a medical scanning technology. The name is derived from the Greek tomos that means slice and graphein to write. CT accumulates a large number of 2-D images taken from slightly different positions relative to the object.

In a CT-capable PCB inspection system, the X-ray source and detector maintain a fixed relationship to each other, but the object is incrementally rotated around a single axis as successive images are acquired. Medical CT scanners reverse the process and spin the source-detector assembly around the stationary patient.

CT computes the characteristics of elemental volumes (voxels) by combining information from all of the 2-D images. A key requirement of the filtered back projection reconstruction algorithm typically used by CT is for the set of 2-D images to represent a full 360° rotation of the object relative to the source and detector. If the 2-D Fourier transform is used for reconstruction, the angular increments must be equal.

Figure 1 shows 3-D images created from hundreds of 2-D projections. Pictures like these typically take several minutes to create. In contrast, because of the primitive computing resources available, the early CT scanners required a day or longer to create 3-D medical images.

The second direction of 3-D X-ray development is tomosynthesis: literally, a synthesized slice. The main distinguishing characteristic of this technique is the greatly reduced data set needed to generate 3-D images. In medical applications, this means lower X-ray exposure and faster treatment. For PCB inspection, tomosynthesis can operate with nine or fewer 2-D images of each area on the board. Given similar processing hardware, this approach is many times faster than CT. The trade-offs are reduced depth of focus and less capability to discriminate between slices.

Both CT and tomosynthesis are distinct from laminography: They acquire all the information to develop any view from the same set of images. This capability is shown in the three slices through the crimped contact in Figure 1. A traditional laminography machine requires a separate set of images for each slice.

X-Ray 2-D Imaging 101

If the X-ray source were a perfect point, imaging would be very easy. Instead, because the source has some finite size, the rays don’t form a simple cone. The rays passing the edge of an object have slightly different angles because they come from different locations on the source. This means that images of the object have a penumbra associated with the edges. The term penumbra originates with Kepler in 1604 and in his usage refers to the area between the darkest shadow (umbra) and the full light.

Use of a microfocus or nanofocus tube increases the cost of an X-ray inspection system but can minimize the size of the penumbra. This is important because magnification is achieved by changing the ratio of the object-to-source distance relative to the source-to-detector distance. For a fixed source-to-detector distance, moving the object closer to the source increases the magnification—and the penumbra.

Improved 2-D Resolution
This trade-off among magnification, image clarity, and cost has been reinterpreted in the Glenbrook Technologies magnification fluoroscopy X-ray 2-D real-time imaging system. With this technology, the object is placed close to the shadow plane—the plane where the image is formed—rather than close to the X-ray source. A large object-to-source distance minimizes the size of the penumbra and allows a low-cost conventional X-ray tube to be used.

The shadow plane is a high-resolution X-ray scintillator coating coupled to the input window of an image intensifier. The fine detail from the scintillator image is intensified 30,000 times at the output window. This bright image then is viewed by an analog CCD camera with optical zoom capability. Compared to the three to four line pairs/mm resolution of a conventional cesium iodide fluoroscopic imaging device, the Glenbrook innovation achieves 15 line pairs/mm and optical magnification up to 40X without requiring a microfocus source.

According to Gil Zweig, company president, this new technology permits the 2-D microscopic examination and video recording of electronic assemblies with significantly lower radiation exposure than can be achieved with any other approaches.

Reverse Geometry X-Ray^® Imaging
The Digiray 2-D X-ray system uses a large, raster-scanned source and a small detector. As in a conventional system, the object is placed close to the source to provide magnification, but that’s where the similarity ends. Because the detector is very small, only rays passing through the object on a straight line from the active point in the raster to the detector are recorded. This means that the background noise resulting from X-ray scattering in a typical system is virtually eliminated.

The detector directly digitizes the transmitted ray with 12 bits of resolution. There’s no need for large-area imaging devices, image intensifiers, or cameras. Zoom and pan are achieved by shrinking/expanding the raster and repositioning the pattern on the face of the source. A 10″ diameter field of view is possible depending on the required magnification.

2-D X-ray systems easily deal with things that optical inspection cannot, such as the solder joints under BGA packages and the connections to components within metal shields, according to Rohit Patnaik, president of Digix Scientific. In Mr. Patnaik’s experience, automated 2-D systems are best deployed in a high-volume, low-mix environment with single-sided boards.

Moving on to 3-D

In Digiray’s Motionless CT^™ X-ray process, several detectors simultaneously measure transmission values at different positions as the raster is scanned on the source. This information is equivalent to the multiple projections used in CT scanning and produces an image of a slice through the object.

The additional capabilities 3-D inspection provides beyond 2-D make it very desirable. Where a 2-D system can determine if a BGA joint is soldered, it takes a 3-D system to determine the volume of voids in the solder. And, 3-D techniques are needed to get unambiguous images of double-sided board assemblies. Machines intended for automated 3-D solder-joint inspection have a set of built-in algorithms that examine the reconstructed images.

Mr. Patnaik explained that 3-D X-ray imaging can keep up with online production speeds, but in the past, machine cost has been an issue. In addition, there are fewer large companies producing 3-D systems today so the choice of suitable machines is limited. Modern systems have the advantage of larger detector size and better X-ray tubes. They also improve throughput by using 2-D when possible and switching to 3-D only when necessary.

Ken Gribble, business development director at Metris/X-Tek, a long-time developer of CT systems, added his views: “Using our technology, reconstruction times for a 2,048-pixel cubed volume with 3,600 projections have been reduced from several hours to 18 minutes. A smaller 1,024-pixel cubed volume with 1,800 projections takes 38 seconds. Nevertheless,” he continued, “it is highly subjective whether CT will be practical for inline applications. Compared to tomosynthesis, CT will be slower because it requires the object to be rotated in view of the detector.”

Parallel Tomosynthesis
Goepel electronic’s OptiCon XLine 3-D System is an innovative X-ray machine based on a custom nine-segment detector. As shown in Figure 2, nine images are simultaneously recorded for each position of the board. Each image is of a different area, so the board position must be precisely incremented until sets of nine images have been accumulated for all the areas of interest.

As the figure shows, all off-axis features are illuminated at a different angle by the cone of rays. Illumination at different angles is a common theme in both CT and tomosynthesis. The resulting projections include features that have been translated in both X and Y depending on their height, location relative to the central axis, and the angle of the X-ray cone. Simultaneously imaging nine areas combined with high-speed image processing achieves a 6.2-in.²/s inspection speed for both 2-D and 3-D operation.

A 3-D reconstruction algorithm that can be used in tomosynthesis is described in an IEEE paper that discusses estimation of the charged particle distribution in an accelerator beam.¹ The algebraic reconstruction technique (ART) is explained in detail and some examples of its use given. For the relatively simple charge distributions being dealt with, a good reconstruction was obtained from three projections. Adding more projections only gradually improved the match between the actual beam distribution and the reconstructed one.

The paper gives no details of the computer system used to evaluate the algorithm, but the time to iterate a solution based on three projections was about 100X longer than for two. Going to four projections required only a small increase in time and actually took fewer iterations. For all iterative algorithms, some convergence accuracy criteria must be set such as variance, discrepancy, or Euclidean distance.

As stated in the paper, “The ART algorithms have a simple intuitive basis. Each projected density is thrown back across the reconstruction space in which the densities are iteratively modified to bring each reconstructed projection into agreement with the measured projection.”

The Speed-Detail Trade-Off
True CT typically takes a relatively long time to create detailed 3-D images. The Y.Fox µCT machine, previously the Feinfocus Fox, features Y.QuickScan, a fast form of µCT reconstruction that takes a couple of minutes. This still is too long to keep up with a high-volume production line.

YXLON positions the Y.Fox as a high-accuracy machine best suited to applications that need high-quality microfocus inspection. Some of the features that contribute to overall image quality are the large flat-panel detector, 16-bit real-time image processing, and geometric magnification up to 2,720X. Views of ±45° are possible with 2-D operation.

YXLON was acquired by COMET Holding AG in May 2007. Feinfocus had been part of COMET since being acquired in 2004. In October 2007, Feinfocus was renamed YXLON International Feinfocus GmbH, and the product line names gained the Y. prefix.

Separately in 2007, GE Inspection Technologies Division, now GE Sensing and Inspection Technologies, bought phoenix|x-ray. At the time, GE announced that the phoenix CT technology had a much finer resolution than the current GE technology and would improve the company’s offerings in both metal and composite nondestructive testing (NDT).

The most recently developed v|tome|x L 240 and L 300 Machines are intended for very high-power NDT on dense objects such as large castings and welded seams. For PCB inspection, the older microme|x and nanome|x machines feature oblique 2-D viewing that helps distinguish the geometry of individual solder joints as well as an optional CT mode. As with the YXLON Feinfocus products, the emphasis is on reconstruction accuracy.

Omron’s VT-X 3-D Inspection Machine features both a true CT mode as well as fast tomosynthesis. Gene Fujita, product marketing specialist—AOI at the company, said, “Increased component density on both sides of a circuit board continues to drive the need for X-ray inspection for solder-joint quality. BGAs and heel fillets concealed beneath the component cannot be inspected using techniques other than X-ray.

“Omron’s CT technology focuses on void inspection as one of the most critical operations that cannot be achieved using only 2-D techniques,” he explained. “Omron’s patented algorithms integrate board movement with X-ray detector rotation to provide maximum imaging data leading to highly accurate analysis at production line speed.” The Omron datasheet includes the qualifying comment that the fast tomosynthesis mode should be used when there are no parts on the other side of the board.

In another example, the YESTech X3 Automated X-Ray System offers a patented tomosynthetic method that requires a very small number of images to extract information pertaining to a solder joint or other relatively flat object. In the X3’s approach, the X-ray detector and/or PCB move to different positions to acquire images of a single region of interest taken from multiple angles. Speeds from 4 to 5 in.²/s are achieved in 2-D mode, dropping to 0.5 in.²/s in 3-D.

Of course, X-ray inspection can’t do everything. According to Don Miller, the company president, “YESTech has long promoted the use of AOI, X-ray, and electrical test to provide the widest range of fault coverage. X-ray arguably provides the best solder-joint inspection, especially on BGAs and other area array devices,” he said. “AOI generally is accepted as the most effective technology for inspection of component-related defects and electrical test for correct impedance signatures.”

Summary

The technology required to support inline X-ray PCB inspection at full-line rates continues to be developed. A few machines have reached that point, but even for those systems, no doubt there are lists of things to improve and simplify if only to reduce the cost. Better detectors, algorithms, and faster processing will improve the reconstructed image quality, which should lead to higher error-detection confidence.

Once an image is reconstructed, a wide range of algorithms is available to examine specific aspects such as solder fillets or solder-ball voids. These capabilities also are improving.

As a user, you care about results. Can a certain machine distinguish between a good joint and a bad one with better than 95% certainty? Can it do this regardless of a device’s X-Y orientation or what may be mounted opposite it on the other side of the board? Can it do all this at the required speed and within your budget?

It’s impressive to see the tremendous detail of a good CT image, but it’s also irrelevant if you need a 10-s inspection time and the CT reconstruction took 10 minutes. Because of the formal mathematical definitions underpinning the process, there’s generally good agreement among manufacturers on what CT means and that it’s the gold standard against which other forms of reconstruction are judged.

The fastest 3-D systems use tomosynthesis techniques with a minimum number of projections. Overall inline performance is governed by the combination of the number of projections, the mathematical and hardware implementations of the reconstruction algorithm, and the effectiveness of the automatic feature inspection routines.

Because of the complex dependencies among these factors, you will get the best indication of a machine’s suitability by actually inspecting samples of your company’s production PCBs. You know better than anyone else the problems these boards have posed in the past. How do the capabilities of the new machines stack up against your present needs and future expectations?

Reference

1. Raparia, D. et al, “The Algebraic Reconstruction Technique,” IEEE, 1998.

December 2009