A DoE for characterizing solder paste stencil printing and measuring paste bricks by 3-D AOI is described.

It has been generally recognized that a significant proportion of SMT defects can be attributed to the solder paste stenciling process. Improved printing equipment makes it possible to exercise more control over various stencil print parameters. Understanding and quantifying stencil print cause-and-effect relationships requires experiments. New developments in machine vision have led to fielding 3-D automatic optical inspection equipment that can measure critical parameters of stenciled solder paste bricks as a response output in designed experiments, and later, as part of inline process monitoring and control.

The intent of the designed experiment was to characterize the stencil print process and develop a mathematical model. The company had recently standardized at this location on the Speedline MPM Accela, and was adopting inline 3-D AOI of solder paste. Having an accurate model of the stencil process and realtime SPC would be of great benefit to all manufacturing lines. The goal was to finetune printing to obtain the best and most consistent solder paste deposits from a given stencil design.

An ideal function diagram of the paste printing process was generated to sort out the various inputs, outputs, controls, and noise effects (Figure 1). The Taguchi DoE approach was chosen and controls selected. Because this effort was to understand and optimize an existing process, a three-level experiment was set up on an L9 array with a sample size of three boards per setup. The existing process was Level 2, with each control reaching up a step to Level 3 and down a step to Level 1. The controls manipulated in this experiment were:

• Squeegee speed.
• Squeegee pressure.
• Separation speed.

Constants in the experiment were identical to production processing:

• Stencil printer.
• Stainless squeegee blades.
• Single stroke printing.
• Kester R560 water-washable solder paste. (Rheology differences between pastes would require establishing a new model.)

The same three bare PWBs were used in each setup, and all nine experimental runs were conducted in a single afternoon. This minimized noise effects from PWB variability and noise that might come from paste aging.

A test vehicle was selected from the current mix of active assemblies, one that employed a good cross-section of component parts and stencil apertures common to many other assemblies currently in production. The standard production stencil fabricated by an AVL source was employed for the DoE, and print processing the DoE PWBs was done the same way as production boards.

3-D AOI was conducted on a Koh Young Technology model KY-3030 VAXL, installed inline and immediately downstream of the stencil printer. Programs for this machine are generated from stencil CAD files. In operation, this machine locates fiducials, then scans all solder paste bricks to measure their sizes and locations. These values are compared against a database of design values with a user-defined tolerance window. The system flags any solder paste bricks that violate specification, and those that are marginal.

AOI measures and records:

• Volume (% of designed value and actual cubic µm).
• Area (% of designed value).
• X and Y offset (mm).
• Brick height (µm).

All readings are stored in database tables in a multi-table structure. With the above parameters recorded for each solder paste brick, and some boards having upwards of 20,000 lands, this database can grow to large proportions. Part of this experiment was to determine which of these parameters were most useful for SPC tracking and which ones need not be recorded.

The experiment was conducted in a single session. Fresh solder paste was drawn from MRO stores for this experiment, the printer controls set up for each run from the L9 array, and three PWBs printed as if they were production boards. Boards were pulled off the line after passing through AOI, solder paste wiped from the surface, and the boards processed through water wash. While boards were being washed, the printer controls were set to the levels indicated for the next run in the L9 array. We progressed through all nine setups in the array, and the AOI machine automatically logged data under serial numbers: “DOE_1A,” “DOE_1B,” “DOE_1C” through “DOE_9C.”

Findings and Analysis

Area measurements from AOI trended the same as volume (Figure 2). Because volume measurements were more useful, area was not studied any further.


Next, x-y offset was examined. This parameter is largely governed by fiducials and the printer's ability to line up the PWB and stencil. No patterns were noted in this data, and x-y offset did not appear to be affected by the control settings (Figure 3). Variability never exceeded 0.2 mm; signal-to-noise measurements showed that changing control levels never exceeded 0.4 dB. A value that low is more background noise than anything else. As a  general rule of thumb, a 2 dB effect is sufficient to be significant and easily measured, and a 6 dB effect is sufficient to reduce standard deviation by one-half.


Studying printed volume for all package types revealed that control levels had little effect on the printed volume for most apertures (Figure 4). Effects of squeegee speed, squeegee pressure and separation speed were generally very small: 3% or less of the brick volumes (Figures 5 and 6). This indicated that with the present process and equipment, stencil design was the major driver of paste volume.


The large population sizes for each aperture design – 48 to 937 bricks deposited per package style with up to five of each package to a board – made it easy to measure variation in the form of standard deviation s. Means and signal-to-noise analysis of measured standard deviation revealed that the controls did have a statistically significant influence on the variability of paste deposits for several, although not all, aperture designs. Response data for standard deviation also showed that not all apertures shared the same response. Levels that would minimize s for one aperture would be far from optimal for another.

Several key and often-used package types were selected, and tables (means and signal-to-noise) for s were generated (Figure 7). These component packages in descending order of importance were:

• BGA-304-S2.
• CBGA-1681-S1.
• FBGA-100-S0.
• FBGA-165-15RD-S3.
• FBGA-165-S2.
• FBGA-256-S1.
• FBGA-400-S0.
• FBGA-725-S0.
• FBGA-783-S1.
• HBGA-937-S0.
• RN0603 8-S0.
• SOT23-S0.
• C0402-S0.
• C0603-S0.
• R0603-S0.

Because optimum for one package is not optimum for all, an algorithm to find the “best fit” level settings was developed. The packages most critical and difficult to rework came first when determining settings. These were arranged as a population-weighted average for determining what levels to use for obtaining minimal s values.

A programmed speadsheet was developed as a software tool to deploy these findings. It generates a population-weighted average prediction for standard deviation based on the number of apertures for each aperture design studied in the experiment. One can look at a single aperture design or create an amalgam of several apertures in different amounts.


Confirmation Run

Several reference designators were initially sampled to represent the various package styles on the board. Response tables for variability were examined for optimal level settings for either lowest standard deviation, or highest signal-to-noise ratio if no substantial improvement in standard deviation was found. A “best fit” level setting was determined for the test vehicle, and the model was used to predict volume and standard deviation of volume (Table 2). Settings to minimize variability as measured by standard deviation were selected as:

• Squeegee pressure Level 3: 14.25 kg [1.65 lb./in.].
• Squeegee speed Level 2: 25 mm/sec.
• Separation speed Level 2: 0.3 mm/sec.

Measurements from a confirmation run of three PWBs verified to the model with standard deviation measuring within 2 to 30% of the prediction (Table 3). A second confirmation run with different level settings was performed several weeks later, and again, measured values of s matched predicted values within 2 to 30%. For purposes of process control, predicting s within 25% was considered good enough for the production environment. The primary goal at this time was to reduce variability in the stencil process, and the DoE succeeded in identifying control level settings that minimized brick volume s. Solder defects linked to stencil printing would be confined to those driven by aperture design.


Because of the large variety of package styles and PWB footprints, this experiment effectively addressed only a sampling of the available apertures. The “library” of response tables for s can be expanded by repeating the L9 experiment on other boards, using the same standard water-washable paste, existing production stencils and solder sample PWBs. It is intended to expand the address response table library with subsequent experiments to aperture designs for all critical component types (BGA, µBGA, TSOP, etc.). Likewise, a model can be developed using the same method for other solder pastes, other printers or any other change in the designated constants. It would be possible to generate response tables for the combination of every variety of solder paste and every printer in use on the campus.

The experiment’s findings were integrated into setup sheets and the printer software. As part of the implementation and standardization, setup sheets permitted only the standard three levels from the DoE for squeegee speed, squeege pressure and separation speed. In the event manufacturing believed a nonstandard control level was required, engineering would run a test to validate or refute this. Response tables were used to update printer profiles for other assemblies, optimizing for variability.

Findings from the DoE were also used to set operating limits on brick size as measured by AOI. Inline AOI monitored printer performance, functioning as realtime SPC. Boards whose solder paste prints fell outside of the limits were flagged (AOI went into alarm status, requiring operator intervention) and wiped clean.

Conclusions

Printed paste volume is almost exclusively a function of stencil design for the stencil print process in use at this campus. On the Accela printer, settings for squeegee pressure, squeegee speed and separation speed generally have no statistically significant effect on the size of the printed solder paste brick. The data indicated that control settings have an effect on printed volume variability (as measured by standard deviation) on most BGA and fine-pitch package types.

Because the data indicated that stencil design has the greatest effect on the volume of paste deposits, this is an indicator that stencil design needs to be standardized. With design being the control that has the most influence, then reducing variability from design will result in reduced variability in stenciled paste deposits on boards. This would be another experiment.

A working model of variation in stencil printed solder paste was developed and adapted to a mathematical tool for setting optimum print parameters. It is believed that reducing the variability of the printed volume (as measured by standard deviation) provides the greatest immediate benefit.

Ed.: This article was first published at SMTAI in September 2006 and is reprinted here with permission.

Tim Wright is a senior process engineer at Jabil Circuit Inc. (jabil.com); tim_wright@jabil.com.

• Prev
• Next