Based on the number of requests I received requesting information on my SPC training program and the conclusion that I cannot supply that many copies free, I have generated a free outline of what I think most people are looking for. It includes recommendations on training methods, basic philosophy, and what elements I recommend to monitor. I hope you find it helpful. I would like to hear back off line from those who access this outline regarding any comments on its contents. I am always looking to improve my program and appreciate all feedback. The complete training manual includes simplified details of how to construct charts, calculate limits, Published articles I have written on SPC, and all of referenced training aids with instructions on how best to use them. If anyone is interested in obtaining the complete training manual please contact me off line. I can be contacted at [log in to unmask] . I would still like to hear opinions or recommendations on the outline.
Again, thank all of you for your interest.
Ed Cosper
Director, Quality Assurance and Engineering
Graphic Electronics Inc.
Tulsa, OK
RECOMMENDED SPC OUTLINE FOR PWB PROCESSES
By: Ed Cosper
The following information is provided under the assumption that the basic understanding of SPC principles is available. This is intended as a guideline to help implement an effective and useful SPC system in a printed circuit board manufacturing environment. It does not necessarily reflect textbook application methods but rather the methods I have personally found to be useful to me.
TRAINING PERSONNEL
I have found it a waste of time and money to attempt to train "everybody" in the statistics of SPC. Generally, you loose the average floor workers attention the first time you put up any formula. I have found the best approach is to expose and teach the manufacturing supervisor about normal distribution and chart interpretation using the following practice.
Typically I use what I have coined and an "Impact Chart" consisting of 30 plots. Basically this is a chart that shows consistent distribution around the mean for the first 20 plots. Then the chart shows a definite change in the average with the last 10 points significantly below the previous mean average. I am careful not to show any findings below .056. The mean average on the chart I use is .062 and represents panel thickness. The latter ( or changed) mean average is about .058. The specification cited is .055 - .069.
First I have the group complete a form. The form consist of a list of all 30 measurements and the specified parameter. However, they are not recorded in the same order as shown on the chart. Beside each of the measurements I have provided a space for the attendee to mark whether or not he/she feels the measurement is in or out of spec. Once they have completed the form, I ask if anyone found anything out of specified limits. Since none are, I expect the group consensus to be "no." I then ask if anyone sees anything wrong with the data. Generally a might hear "well, some are close but OK".
Once everyone has concluded that the data recorded in front of them shows no problem. I show them the impact chart. Explaining that the charted data is the same data they had just worked on, I emphasize the different conclusion one can reach from looking at basic recorded data verses charted data. This is where the impact comes in. I then explain that it is a normal tendency for an operator to evaluate one lot at a time and draw conclusions of how the process is doing based on individual lots. I explain that monitoring charted data gives us a larger picture of what is going on. Much like the class exercise, unless something is out of spec, we don't normally notice changes. I then explain about normal distribution and how it relates to charted data.
Regarding interpreting the charts, I provide samples of all the basic "rules". However, what I am most concerned with is watching for trends or changes in mean average. I instruct the class to notice single point occurrences outside limits but if the reason is not quickly evident, don't waste time looking for it. It is probably related to an operator set up error that is not going to be admitted.
The operators are then taught what to measure, how to measure it, and how to put in on a chart. The supervisor is chartered with the responsibility to show the operators how to monitor a chart. I have found that thru basic human interest, operators learn quickly how to notice concerns on charted data.
SOME BASIC CONCEPTS
1. SPC is not a quality program. It primarily is a manufacturing program. If implemented
properly, you could loose an arm trying to take away an operators chart they have
become dependent on the show everybody how good their process is.
2. The quality organization should enforce what management has decided to implement.
This is best done by audits.
3. Resistance. This is to be expected. One thing I have learned is that one of the first things
an effective process control / spc program will highlight incompetence. Be assured, if
there are personnel in positions that are unqualified, this will become evident to everyone.
As the system defines deficiencies, someone must be responsible to correct those
concerns. If the person currently responsible can't, won't or does not know how to correct
the concern, actions must be taken in this area. This is probably the most common
reason these types of systems fail and are pushed to the side. Bottom line, sometimes it is
hard for a companies management to address these situation for various reasons.
4. Chart only what you feel is important! Do not chart data and try to impose SPC just to
satisfy a customer. If it doesn't serve a purpose towards improvement, then it is a waste of
time and money.
5. Use and devise chart formats to suit your needs. We employ many "modified" charts that
provide me with the information in a format that makes most sense to us.
6. Use the data to perform capability studies. Keep the studies in the chart jacket it you want.
It is the capability of a process that is important. Therefore, I see no need in placing actual
statistical limits on a chart if you don't want to. As long as you are using and responding to
charted data, you are accomplishing the intent of SPC.
WHAT TO MEASURE ?
The following is a guideline which outlines the processes most conducive to the application of SPC.
It does not represent everything that can be measured.
1. Tooling and film storage.
This area is best monitored using an automated chart recorder for temperature and
humidity. Once data is collected for a few weeks, data points from the chart can be
selected to represent a sample. A process capability can be done based on those
selected data points. At this time the information can be evaluated by management to
determine if the controls are adequate for the purpose and to establish the limits the
process will be reacted to. Once the limits are established, they can be highlighted on
each new blank chart.
2. Receiving inspection.
Laminate thickness is very important in multilayer processing. Any laminate thickness that
is determined to be critical by management ( usually thin cores) can be sampled and
charted on an X bar R chart. Measurements should represent the middle and four corners
B-stage scale flow tests performed on a sample representative of a typical current
production panel size can be performed. 3 or 4 plies can be pressed which would most
typically represent the number of plies in a normal dielectric. I feel it important to replicate
as close as possible the actual production environment as this is how the product will
actually perform. An X-bar R chart can be used. Sample location will typically represent
the middle and 4 corners. Sample size and frequency is discretionary. The scale flow test
can also double to evaluate the flatness and pressure distribution of the press plattens.
The data obtained above is very helpful when trying to accurately calculate overall
expected post lam and dielectric thickness. I have come not to rely on suppliers test
results or test data as it usually does not accurately represent the products performance
in our specific process. I have found that charting and monitoring the consistency of
supplied material thickness can be very helpful in convincing suppliers to minimize their
variation. Although we must accept by the spec, no longer must we buy by the spec. We
can choose our preferred suppliers based on consistency rather than basic compliance.
3. Drilling
Periodically the drill accuracy of the machines need to be verified. The accuracy in
production mode is the most meaningful. A standard test pattern can be programmed
which exercises all axis movements at the extremes. The hole locations should be on a
.100" grid for easy reference. The bottom left hole in the bottom left corner of the panel
will become zero or datum. The test pattern should consist of several holes drilled in each
corner of the panel forming a 90 degree angle as is covers both axis of each corner. This
pattern can then be loaded or added to any program currently running and the test pattern
drilled on an actual production run. This will give an accurate insight as to the accuracy of
the process as it relates to production panels. The variations can then be plotted on a
separate X bar R chart for each axis or as a true position value on a single X bar R chart.
4. Electroless Copper
Although the baths should be analyzed by the lab and results recorded, I have found
charting these results to be ineffective. The critical element of this process is the
deposition rate. The deposition rate or weight gain should be obtained and plotted on a
moving range chart by the operator at a frequency determined by management. I feel
that the operator should be totally aware of what the process is producing and how
consistent the process results are. This instills a sense of security with regard to quality
and confidence with regard to his/her process.
5. Imaging Areas ( Photo Print )
Since many problems stem for improper lamination of the resist, it is critical that the
most probable contributing factor be eliminated. By monitoring and charting the exit
temperature ( or roll temp if preferred) you can monitor this element for consistency.
More that one chart may be needed for thickness variation. Erratic distributions will be
generated if thin innerlayers are charting on the same chart as .062 final boards.
Exposure light intensity can be monitored to determined the typical or expected
degradation rates of currently used light sources. Then the light source can be effectively
monitored for a change in predictable degradation. The measurements can be made in
either time ( seconds ) to a predetermined millijewel or millijewels at a predetermined
exposure time. ( I prefer the latter) The sample frequency should be based on number of
exposure. A standard single point chart will suffice.
6. Copper / tin lead Plating
This is perhaps the most difficult to monitor statistically. Again, I do not recommend
charting all the lab analysis. Especially if you are using titration. Chart only what you
feel is critical to the process. One thing to keep in mind is the operating windows. Some
elements have windows you can drive a semi though. So why try to monitor something
that has a variation of +/- 5 when the operating window is 50. Just as an example.
Plating is best monitored thru cross section but not on actual product. Current density
variation between designs prevent a normal distribution.
temperature to be a major concern needing that close of monitoring.
9. Hot Air Solder Level
Because copper contamination is a critical element in this process, a moving range chart
can be utilized to monitor the copper content. Although you will not generally obtain a
normal distribution, you can use the chart to monitor trends. It is a very effective tool in
determining how to maintain you solder bath with regards to how often to skim, decant,
and analyze.
Regarding the solder thickness, I have been successful in establishing a thickness model
which depicts the solder thickness deposition across the width of a panel in the plating
thru holes. Basically, we take an 18x24 panel with a test hole pattern drilled at .5"
increments in both axis. We then electroless and flash plate the panel so that the holes
will take solder. Then we hot air level the panel and cross section across the panel. By
plotting the thickness on a model, we can see how the solder thickness varies across the
surface of the panel. Because air knives have a tendency to warp and bow, we can tell
when the effects are to a degree that the air knives need to be changed.
As far as monitoring SMD pad thickness, I have found the results to varied to monitor.
This primarily due to puddling.
10. Fabrication
Generate a standard rout program. Make it simple yet insure it exercises all axis
movements. Rout the test program on scrap material and measure the dimension you
feel are important. You can use a moving range chart for each axis or dimension.
Frequency is at management discretion.
11. Yields
Although the text books say to use a P chart for overall yields, I have found the limits on
a P chart useless due to the varying sample sizes. Yields should be monitored. I recommend plotting yields on a chart against a predetermined goal. You can monitor the trend as it relates to the goal. I would recommend calculating and plotting average lot yields and not overall yield. As an example: If you ran two lots one for 5000 and one for 50. Suppose the 5000 lot yielded 4990 and the 50 piece lot yielded only 25. If you plot the overall, you would show a 99.3 % yield whereas you had one lot that actually had 50% yield. It is important that the 50% lot be represented on the chart.
End
C Copyright 1991 Ed Cosper All rights reserved
***************************************************************************
* TechNet mail list is provided as a service by IPC using SmartList v3.05 *
***************************************************************************
* To subscribe/unsubscribe send a message <to: [log in to unmask]> *
* with <subject: subscribe/unsubscribe> and no text in the body. *
***************************************************************************
* If you are having a problem with the IPC TechNet forum please contact *
* Dmitriy Sklyar at 847-509-9700 ext. 311 or email at [log in to unmask] *
***************************************************************************
|
