Troubleshooting experiments-STRIPPED FOR DOWNLOAD_Metal
Contents
1. Troubleshooting experiments S TRIPPED FOR DOWNLOAD Metal Tech S ppking Template USE THIS qxd 3 8 2012 10 21 AM Page 1 TECHNICALLY speaking BY PATRICK VALENTINE NORTH AMERICAN PCB AND EP amp F BUSINESS MANAGER OMG ELECTRONIC CHEMICALS MAPLE PLAIN MINN Using the Plackett Burman Family of Designed Experiments for Troubleshooting Process Controls 1946 R L Plackett and J P Burman published their now famous paper The Design of Optimal Multifactorial Experiments in Biometrika vol 33 The paper described the construc tion of economical designs with the run number a multiple of 4 rather than a power of 2 like full factorial and fractional factorial designed experiments Plackett Burman designs are very efficient screening designs when only main effects are of interest These designs are used for screening exper iments because main effects are in general confounded with two factor interactions Designed experiments that are confounded are described by their resolution A designed experiment that has a main factor confounded with a two factor interaction is a Resolution III An easy way to remember this is to hold three of your fingers up in front of you grab two of your fingers with your oppo site hand and you have one finger that is confounded with the two held in your other hand This simple example can be used with Resolution IV and V designs to visualize the con founding using four and f2. metalfinishing 21
3. in fact it is true And we do that by increasing the alpha value in the significance testing Figure 2 shows a power curve for a Plackett Burman 12 run experiment with 10 factors and an alpha value of 0 05 We can see on the horizontal axis that with 50 power we ll need to see an effect size of 5 standard devia tions And the effect size is the dif ference between the low and high fac tor level means Discussions on power calculations can get lengthy so we ll save that for another day Historically alpha values of 0 05 or 0 10 were used due to computations being done by hand With the advent of the modern computer we can easi ly alter the alpha value and deter mine its effects For screening experi ments I recommend using alpha val ues up to 0 15 be used to increase power in the experiment After all it s a screening experiment and we don t want to make the acceptance criteria too harsh There are six basic steps to follow when designing and running experi ments 1 State the problem define the objectives 2 Design the Experiment a Choose factors amp ee Figure 3 Stray plating defect seen on blue soldermask reasonable ranges for each b Determine appropriate responses amp how to measure c Select a design know your pros amp cons and review runs d Approximate power of the experiment 3 Randomize amp run the experiment 4 Analyze the experiment 5 Confirmation runs amp Pok esti
4. to aid the process engi neer in troubleshooting a process as it allows one to separate the vital www metalfinishing com Troubleshooting experiments STRIPPED FOR DOWNLOAD Metal Tech S pking Template USE THIS qxd 3 8 2012 10 22 AM Page 6 TECHNICALLY 3 SPEAKING zerenens wi Miiat aso Quality Press Milwaukee WI 5 Launsby R and Schmidt S few factors from the trivial many 2005 Understanding Industrial The process engineer must keep in Designed Experiments 4th Edition mind that statistics are only tools to Air Academy Press Colorado help us they don t replace the engi Springs CO neers skill and intelligence REFERENCES 1 NIST SEMATECH 2010 BIO e Handbook of Statistical Patrick Valentine is the North American Methods PCB and EP amp F Business Manager for 2 SARIS Laser Ablation ICP MS OMG Electronic Chemicals As part of his analytical technique developed by responsibilities he teaches six sigma green Balazs NanoAnalysis belt and black belt courses He holds a 3 Montgomery D 2005 Design Master s Degree in Business from Regent and Analysis of Experiments 6th University Six Sigma Master Black Belt Edition Wiley amp Sons Inc certification from Arizona State Hoboken NJ University and is an ASQ certified Six 4 Mathews P 2005 Design of Sigma Black Belt Patrick can be contact ed at patrick valentine omgi com www metalfinishing com January February 2012
5. 0 100000 AuNi Plating Defect Backside Signal Intensity c s Figure 6 Results of the SARIS analysis of the stray plate Blue Solder Mask Signal Intensity c s Figure 7 Results of the SARIS analysis of the blue soldermask 20 metalfinishing January February 2012 Figure 8 Metallic copper particles embedded in the slip sheets 10x Figure 9 Metallic copper particles embedded in the slip sheets 40x The more advanced the cross linking of the soldermask longer final cure time the less likely the soldermask was to pick up copper particles which makes sense These types of stray defects are usually amorphous in appearance and once the root cause of the problem was identified and eliminated the stray plating ceased Knowledge of the potential for sol dermask to pickup particulate should be disseminated throughout the manufacturing organization and lessons learned should be archived in a FMEA Failure Mode Effects Analysis These activities are known as process management SUMMARY When trouble arises in today s com plex manufacturing processes the process engineer needs to react quickly Many times the engineer is faced with dozens if not hundreds of process control options to change Using designed experiments as a tool for troubleshooting can be very effec tive saving both time and money The Plackett Burman family of designed experiments can be an effective tool
6. ett Burman design runs are a multiple of 4 beginning with 12 runs there is no 16 run experiment Each design can have up to k n 1 factors for example the simplest Plackett Burman the 12 run can have up to 11 factors 11 12 1J One can easily see the advantages to a 12 run experiment being able to model up to 2 2 048 different Power Curve for Plackett Burman Design 0 Effect Terms Omitted refers to main effects J Figure 2 Power curve for the Plackett Burman 12 run DoE with 10 factors and an alpha level of 0 05 e www metalfinishing com IL Troubleshooting experiments STRIPPED FOR DOWNLOAD Metal Tech S pking Template USE THIS qxd 3 8 2012 10 21 AM Page 2 TECHNICALLY speaking Factors Step Exposure Description A E Stouffer step wedge exposure level for the soldermask Developer Passes Number of passes through the soldermask developer Pre Cure Soldermask pre cure time 180 F Final Cure min Soldermask cure time 300 F UV Bump Pre Scrub No Yes Ultra violet bump of the soldermask prior to ENIG processing S Light brush scrub prior to processing the ENIG line No Cleaner Fee Use of acid cleaner in the ENIG line Micro Etch p in 5 4 Micro inch removal in the etch step in the ENIG line Palladium activator time in the ENIGdine Pre Dip min 1 Sulfuric pre dip time prior to the palladium activator in the ENIGtine Activator min Table 1 D
7. f the final model Table 5 that the Lack of Fit is sta tistically significant p value signifi cantly below 0 05 which indicates we are lacking higher order terms that of possible interactions and or quadratic terms With Lack of Fit www metalfinishing com detected the general protocol would be to abandon the tentative regres sion model in attempts to find a more appropriate equation Our primary objective is discovering sig nificant factors and our secondary ee objective is building a regression model that of predicting the value of stray plate dependent variable from the factors independent variables Y flx1 X2 X Due to the lack of fit we will abandon the secondary objec tive and ignore this condition We have detected two statistically significant variables that of Micro Etch and Final Cure which deserve further investigation Results of the SARIS analysis are shown in Figure 6 Of significant interest is the copper Cu that was detected There is high correlation between the copper detected on the backside of the stray plate defect and the statistically significant factor Micro Etch Why Because the Micro Etch is removing copper For a reference point the blue sol dermask was analyzed by SARIS and the results are shown in Figure 7 Note that copper is detected but at a much lower level than in the stray plate defect so we can conclude that the low level detected in the s
8. ive fingers respectively Another way to look at this is by correlation analysis With a Resolution III fractional factorial design the correlation between a main effect and a two factor interac tion is either 1 or 1 remember that correlations range between 1 and 1 and in correlation analysis we usual 16 metalfinishing January February 2012 Interactions AB AC BC Factors ABC Figure 1 Confounding effects of a Resolution Ill design with correlation of 1 or 1 ly examine correlations that are 2 0 70 An example of this is shown with a 2 half fractional design in Figure 1 The and signs repre sent the low and high factor levels and are coded as such Summing this up what we have with fractional fac torials 2P Resolution III designs is a main effect column X is either orthogonal to X X sums to 0 or is identical to minus one 1 or plus one 1 X X Even though the Plackett Burman designs are Resolution III they pos sess one unique attribute The two factor interaction column X X is cor related with every X for k not equal to or j at a correlation value of 0 33 or 0 33 with this weak correlation the effects of confounding are mini mized Furthermore the interactions are uniformly dispersed over all the experimental runs These two attrib utes make the Plackett Burman designs good choices for screening experiments when the number of fac tors being studied is 25 Plack
9. mations 6 Report amp recommendations It does make sense to use the Plackett Burman family of designed experiments to separate the vital few factors from the trivial many when troubleshooting a process Let s look at a real example using the Plackett Burman 12 run screen ing experiment Problem statement During elec troless nickel immersion gold ENIG plating of printed circuit boards a random defect is seen that is known as stray plating plating of nickel and subsequent gold on top of the soldermask See Figure 3 for an example of stray plating This defect leads to rework and potential shorts during or after assembly After a brainstorming session the factors and levels were chosen and are shown in Table 1 Concurrently the defect sample in Figure 3 was removed with Scotch tape and sent to an outside lab for analysis using SARIS Material Analysis which is a combination of high power laser beam technology used to ablate the sample and subsequent analysis with the high mass resolution capability of the ICP MS elemental survey and quantitative analysis of the material January February 2012 metalfinishing 17 Troubleshooting experiments STRIPPED FOR DOWNLOAD Metal Tech S ppking Template USE THIS qxd 3 8 2012 10 21 AM Page 3 TECHNICALLY speaking Standard Step Developer Pre Order Exposure Passes Cure Table 2 DoE design array Standard Step Developer Pre Order Exposure Passes C
10. ndividual pieces found on each test vehicle Analysis of variance ANOVA was used to analyze the full model Table 4 shows the ANOVA while Figure 4 shows a Pareto chart of the standard ized effects The final model reduced ANOVA and Pareto chart of the standardized effects are shown in Table 5 and Figure 5 We can see in final model ANOVA www metalfinishing com IL Troubleshooting experiments STRIPPED FOR DOWNLOAD Metal Tech S ppking Template USE THIS qxd 3 8 2012 10 21 AM Page 4 TECHNICALLY speaking Analysis of Variance for Stray Plate coded units sorce or sass f ass Pade rr Main Effects Step Exposure Developer Passes Preseure Final Cure UV Bump Pre Scrub Cleaner Micro Etch Pre Dip Activator SS SF SS SS SS i SS CS 1421 33 033 3530 8 33 408 33 75 00 033 27 00 736 33 5 33 75 00 Resauateros a saj saj oa n e e Table 4 ANOVA of the full model Pareto Chart of the Standardized Effects response is Stray Plate Alpha 0 15 Micro Etch Final Cure Developer Passes Activator UV Bump Cleaner Pre Cure Pre Dip Pre Scrub Step Exposure 2 3 Standardized Effect Figure 4 Pareto chart of the standardized effects for the full model Analysis of Variance for Stray Plate coded units Source f sass Ass agms OF P Main Effects Final Cure Micro Etch Residual Errors Lack of Fit Pure Error ue ee E ety i Table 5 ANOVA o
11. oE factors and levels process combinations These designs are also known as saturated main effect designs because all the degrees of freedom are utilized to estimate main effects If no p values probability are cal culated due to design saturation then the analysis can be done graph ically or by calculating percentages based upon the main effects Sum of Squares divided by the total Sum of Squares If we apply the principles of Pareto then we ll most likely be removing multiple main effects which will in turn give us the need ed degrees of freedom to calculate the p values If we use a maximum of k n 2 factors this allows one degree of free dom for the error term which allows the calculation of p values for each of the main effects in the model simultaneously Even with this we can still have up to 2 1 024 differ ent process combinations Saturated Plackett Burman designs typically have low power Power is defined as the conditional probability that you will avoid a Type II error and a Type II error is failing to reject accepting the null hypoth esis when in fact it s false Ideally we want power to be at least 50 with a preferred power value a 280 Certainly as we refine our model by removing non significant main effects we ll increase power Another www metalfinishing com way to increase power is to increase our risk of making a Type I error that of rejecting the null hypothesis when
12. older mask is a filler and the high level detected on the backside of the stray plate is an anomaly The screening DoE was able to eliminate a significant number of the study variables and study variables that are eliminated contain a wealth of information about the process itself and are key in developing enhanced process knowledge Therefore it is recommended that time be allotted to debrief team members and those responsible for the processes under study Armed with the identification of the two significant factors and the results of the SARIS analysis fur ther investigations were made into the entire upstream processing It was found that soldermask oper ators were slip sheeting and stacking final cured panels as they were com ing out of the oven The slip sheets were investigated as a potential source of contamination Figures 8 and 9 show what was found embed ded in the slip sheets themselves Contaminated slip sheets were transferring copper particles to the hot soldermask surface when the panels were stacked after final cure January February 2012 metalfinishing 19 IL Troubleshooting experiments STRIPPED FOR DOWNLOAD Metal Tech S ppking Template USE THIS qxd 3 8 2012 10 22 AM Page 5 TECHNICALLY speaking Pareto Chart of the Standardized Effects response is Stray Plate Alpha 0 15 1 574 Micro Etch Final Cure 2 3 Standardized Effect With Expanded Y Axis Full Scale Counts 100000
13. ure Cleaner Mitel xe Pre Final UV Pre Cleaner Cure Bump Scrub Activator Etch Dip Micro Pre Activator Stray Etch Dip Plate E E a ee ee Table 3 DoE responses of stray plate is then obtained Note that the backside of the stray plate was ana lyzed the side that was in direct con tact with the soldermask Minitab 16 software was used to create the Plackett Burman 12 run designed experiment and the design array is shown in Table 2 When running any designed exper iment it s important to do all you can to minimize any noise influence in the testing Noise or lurking vari ables are simply unidentified vari ables that change during an experi ment that cannot be controlled 18 metalfinishing January February 2012 Lurking variables can be tolerated when theyre managed correctly but disastrous when they re not To help in reducing noise it s important that the test vehicles mimic the real pro duction pieces so for this experi ment the test vehicles were actual multilayer printed circuit boards in panel form with a dimension of 16 x 18 inches Another critical step in negating noise is to randomize the experimental runs if noise is present we generally see an inflation of the error variance without adding bias to the real differences between the fac Q tor levels The responses to each of the 12 individual experimental runs are shown in Table 3 The counts of stray plate are actual i
Download Pdf Manuals
Related Search