20/20 foresight: detecting bridges before they occur.

The solder paste print operation is widely recognized as a primary source of defects in surface-mount assembly. One approach to increasing yields is to detect print defects immediately after the print operation and reject defective boards before the placement of electronic components. This practice enables surface-mount manufacturers to save time otherwise wasted in the assembly of defective boards and avoids costly rework.

Inspection Sequence

The general approach to inspecting printed circuit boards (PCBs) is illustrated in Figure 1. As an initial operation, the image of a section of the board is acquired either by an area or line scan camera. The acquired image is then processed so that regions of the board that are covered with paste are more easily identified. Once appropriate regions of interest are defined, various analysis techniques are used to either quantify paste coverage on or in the vicinity of pads, paste coverage in the gap between pads, or to characterize bridge features. The resulting measurements are compared to user-defined process limits. Historical data is used to monitor trends for effective statistical process control (SPC) of print functions.

[FIGURE 1 OMITTED]

Paste Detection: Enabling Image Processes

The term paste detection is used to describe an image processing method in which solder paste is separated from non-paste features, creating a new, paste-only image. The resulting image is not analyzed at this stage to determine the quantity, location or significance of paste deposits. Paste detection methods include direct application of single or dual thresholds to the captured image, image subtraction techniques and texture-based segmentation. Variations may include use of ultraviolet (UV)-dye enhanced paste, laser profiling to create a topographical image and x-ray techniques. All methods are intended to better separate paste from non-paste regions for subsequent analyses such as bridge detection.

At present, the most common techniques are based on single or dual threshold methods that use pixel brightness to isolate solder paste regions on a substrate. A single threshold, or brightness level, is chosen to create a binary image depicting paste and non-paste regions. Dual threshold methods work similarly but include a scalable transition region of brightness levels between thresholds to more accurately account for edge pixels. This transition region is preferred since small paste features can contain a relatively significant amount of edge data.

Subtraction methods create a difference image by subtracting one image from another. Subtraction-based paste detection methods typically compare differences between a reference image and a newly captured image to detect paste. These images must be registered precisely to each other before subtraction to minimize error. A relatively significant amount of storage, retrieval and buffer capabilities are needed to accommodate the many reference images.

Ideally, a complete set of reference images needs only to be acquired once, but, in doing so, performance can be adversely affected by typical variations found between otherwise identical substrates. To avoid such problems, new reference images must be acquired for each substrate to be inspected, which can slow the process considerably and may be inappropriate for use in a high-speed production environment.

Using threshold or subtraction methods alone to effectively separate paste from background information when brightness levels overlap may be difficult, if not impossible. Area and location of paste deposits cannot be determined with certainty under such conditions' thus limiting the value of SPC data and adaptive control responses.

A novel method for detecting solder paste that uses texture as a means of recognition has been developed and tested. In particular, the method performs a combination of mathematical and morphological operations on a digitized image to separate areas of texture, specifically that of solder paste, from other dissimilarly textured non-paste areas on any PCB, stencil, wafer or any substrate or similar surface. As long as the acquired image data in paste regions is valid with no loss due to clipping, the relative brightness of non-paste regions and features is not a factor. The paste detection routine processes live or captured images, or portions of images, that include at least some part of the total area to be inspected. Successful acquisition methods include line scan, area scan, analog and digital input devices with supporting hardware and software.

The texture-based routine relies on the unique image characteristics of solder paste that are most pronounced when the image is in shall focus. System magnification must be high enough to adequately sample paste texture, yet not so high that unique paste frequency characteristics are lost. The process creates a weighted, paste-only image where higher values indicate a higher probability that paste is present at any given location (Figures 2A and 2B).

[FIGURE 2 OMITTED]

Bridges and Other Print Defects

In solder paste printing operations, the term bridge generally describes a print defect where some amount of stray paste spans the gap between adjacent pads. At critical dimensions, a paste bridge may fail to pull back during subsequent reflow operations, causing a short or other related defect in the final assembly. Not all bridges or bridge-like defects have the necessary mass or geometry to adversely affect a given process. Conversely, gap defects that actually are significant to a process may not always connect adjacent pads to form a well-defined bridge. Figures 3A and 3B give a close-up view of a bridge defect and several typical non-paste (board) features as seen in a run-time image (Figure 3A) and in a processed paste-only image (Figure 3B).

[FIGURE 3 OMITTED]

Bridges are not the only type of print defect. Excess paste, poor print definition and poor alignment also increase the probability that similar defects, notably shorts, may appear at this location later in the assembly process. Although subsequent processes certainly play a part in determining the quality of the final PCB assembly, detection and accurate assessment of bridges and other defects, as printed, would provide the most direct feedback for process control of appropriate print functions.

Bridge Defect Characterization

Bridges or bridge-like features occur when a relatively well-defined deposit of paste spans the gap between pads, or nearly so, or beyond predefined limits. As the span of a paste feature increases across the gap, so does the significance of the feature. A classic bridge would span the entire gap, but span alone does not guarantee that a related defect will occur later in the process. Something less than full span could be equally troublesome or equally benign. Additional characteristics must be considered to gauge the true significance of a defect to the process.

Although sections along a bridge may be narrow or weak, or its bridge-like geometry poor, the probability that bridging will occur at some point during reflow may be great due only to the amount of paste involved. As the area covered by the paste feature increases, so does the probability that it will cause a bridge-related defect when sufficient span exists across the gap.

The thinnest point along the bridge feature--its weakest link--may indicate an ability or tendency to pull back during subsequent reflow. If a section along the bridge is sufficiently narrow, the probability that the paste will break and pull back from this point is greater than a deposit that remains relatively or critically wide at its narrowest point. As the width of the paste feature increases, so does the probability that it will cause a bridge-related defect, provided a sufficient span exists across the gap and the total amount of paste forming the bridge is sufficient to maintain it at reflow.

Assessment of bridges, bridge-like features and other print defects can be a subjective task with few hard rules or limits. Yet a machine can only analyze tangible characteristics. Reliable methods to classify and weigh the significance of print defects as they relate to the process are required to provide meaningful output and to define realistic and useful process limits. Bridge, bridge-like and too much paste are subjective descriptions of print detects. Without further technical assessment, none of these can predict with certainty that a bridge-related defect will appear later in the process. In fact, no form of measurement can predict this defect with certainty. They can, individually or together, indicate a process trend where the probability of such a defect is increased.

Detection of Paste-In-Gap, Bridge-Like Geometry

A unique gap defect analysis has been developed that provides reliable paste-in-gap area measurement and detects significant geometry and span of bridge-like paste features as they relate to the surface-mount assembly process. The total amount of paste in a gap and the effective span of bridge-like features across the gap are used together to determine the probability that a specific paste feature will cause a bridge-related defect.

A general paste-in-gap defect would occur when the total quantity of paste, regardless of shape or location in the gap, is beyond predefined limits. An actual paste bridge does not need to be confirmed, and no further characterization of the defect is required to qualify it as such. This condition indicates generally poor print quality, poor alignment, bridging, or all of these, with increased probability that bridge-related defects, primarily shorts, will appear at this location later in the assembly process.

A more specific bridge defect would occur when the span of a bridge or bridge-like feature is beyond predefined limits. The operator would select low, medium, or high bridge detection sensitivity to define what he or she considers to be a bridge of minimum sufficient bulk or width, in relative terms, for the purposes of measuring the equivalent span of bridge-like features that are at least as wide or wider. The low setting would require a bridge-like feature to have more critical mass, more bridging strength, more significant bridging geometry, or root mean square (RMS). The high setting would measure the span of much finer and less substantial wisps of paste--more like peak than RMS. So with bridge sensitivity set high and span limits set relatively low, a higher number of short, wispy, bridge-like features would be flagged for review. All of our span data were taken at high sensitivity.

Repeatability of the Bridge Analysis Technique

For the inspection method described here to be useful, the bridge measurements must have an acceptable degree of repeatability. For the purpose of this study, we have chosen to evaluate the repeatability of measuring both the span of bridges and the area of the gap covered by paste. We have also chosen to set the bridge detection sensitivity to what we feel is the highest practical level for a production environment. A lower sensitivity may prove to be more desirable in actual practice.

A test board was printed, and defects were intentionally introduced on two of the four sides of a quad flat pack (QFP) device. Figures 4A and 4B show part of one inspection group taken at two stages of the bridge detection procedure. Figure 4A is the actual run-time image of a portion of the test board. Figure 4B is the processed paste-only image of the same region of interest. Figure 4C is an overlay of the run-time and paste-only images for illustrative purposes. The test board was run through 120 dry print cycles, each followed by an inspection cycle, where area and span data were collected for each of 176 gap locations.

[FIGURE 4 OMITTED]

Figure 5 is a plot of repeatability vs. paste-in-gap area. Both axes are in percent of total gap area. The horizontal axis, mean reported gap area, ends at 25% since no paste-in-gap areas larger than that were present. In fact, less than 20% of the reported paste-in-gap areas were above 5% of the total area. These results highlight the need to also detect and accurately measure effective span of significant paste geometry across the gap, in parallel, since area measurement alone obviously provides incomplete bridge detection or prevention analysis.

[FIGURE 5 OMITTED]

Figure 6 is a plot of repeatability vs. reported bridge span. Both axes are in percent of total width of the gap. Results are least repeatable (at third standard deviation) when bridge detection sensitivity is set high and only wispy, bridge-like spans exist. In this case, fortunately, the probability of a bridge-related defect is remote, and span limits can be adjusted to avoid nuisance detection of shorter wisps. Statistical analysis would still reveal significant trends for successful process control. Of course, more substantial bridge features provide more repeatable span measurement at any sensitivity setting.

[FIGURE 6 OMITTED]

Conclusions

The precise inspection of bridges is a critical tool, not only for the detection of today's most common surface-mount print defects but also for correcting undesirable trends in the process. Since a relatively insignificant paste-in-gap area can provide significant bridge geometry across the gap, and vice versa, both paste-in-gap and span measurements are needed to reliably determine the true significance of bridge-like features to a process.

The measurement of paste-in-gap area is inherently more repeatable than that of span due to differences in what is being measured and the amount of data involved. Repeatability of partial span measurements is affected by the bridge detection sensitivity programmed into the system. We used what we feel to be a high level of sensitivity during our tests, although a lower setting may prove to be more desirable in actual practice. The detection of bridges that span the full gap width is most repeatable and could certainly be used for the purpose of identifying defective boards based on the classic definition of bridge.

Bridge detection can only be as successful as the paste detection method used to separate paste from background. This study shows that our novel, texture-based paste detection method and bridge analysis technique provides useful and reliable measurement of bridge characteristics that are relevant to the board assembly process.

David P. Prince is senior machine vision engineer, e-mail: [e-mail], and Dr. Gerald C. Pham-Van-Diep is director of advanced development, e-mail: gphamvandiep@cookson electronics.com--both with Cookson Electronics Equipment, Franklin, MA.