Applying Knovel Math to Real World Problems – Identifying the Root Cause for a Drop in Yield
December 8th, 2009In this solution story, an Operations Manager at an electronics manufacturing company troubleshoots the root cause of a drop in yield using basic search and Knovel Math.
Knovel Math provides fully documented Mathcad® worksheets for engineering calculations from trusted reference works, reducing the time it takes to find, solve and document equations. Let’s get started.
Identifying the Root Cause For a Drop in Yield
An electronics manufacturing company specializes in Printed Circuit Board (PCB) assembly in multiple locations globally. Typically, the PCB assembly operations run at a first pass yield of 98% but in the last two months the yield has dropped drastically, by 76%, which adds significant re-work and increases in cost. The Operations Manager is tasked with identifying the root cause and restoring the processes back to 98% yield.
The company’s current process is as follows:
* Incoming material inspection (printed circuit board, solder paste, components)
* Solder paste printing, solder paste print inspection, component placement, reflow, visual inspection, electrical test, storage
The Operations Manager reviews each of the processes and determines that there may be failures at the solder paste printing step. He finds that the defects are mostly related to solder joint shorting or solder bridge.
Determination of Mean and Standard Deviation of Inspection Data
The solder paste inspection system measures solder paste height at selected locations on the printed circuit board. The Operations Manager looks at historical data from the inspection step after solder paste printing and compare it to data from the last two months.
The Operations Manager types ‘mean and standard deviation’ into Knovel’s basic search to find the information on calculating this value.
He chooses the Mathcad enabled Handbook of Civil Engineering Calculations which contain formulas needed for calculating mean and standard deviation and starts to calculate the mean and standard deviation of inspection data (solder paste height) for the past two years. (Click on the image below to zoom in on details)
Then he calculates the mean and standard deviation of inspection data (solder paste height) for past two months. (Click on the image below to zoom in on details)
The historical data of past two years has a mean of 4.952 mils and a standard deviation of 0.488 mils whereas the data from the last two months shows an increase in the mean from 4.952 mils to 6.488 mils with a standard deviation at 0.513 mils.
After calculating these data, he can confirm that there is a problem in the solder paste printing process. He looks at all the factors (variables) that can have an impact on the solder paste print height, i.e. solder paste rheology, stencil thickness and aperture size, solder paste printing speed, printing pressure, and release height.
The solder paste properties were measured and compared to the previous batches. The Operations Manager found no statistically significant differences between the solder paste properties. The stencil apertures were measured for wear and it was determined that the stencil did not contribute to the variation is solder paste height.
The three process factors (print speed, print pressure and release height) were further evaluated. A full factorial (33) design of the experiment approach was used.
He types ‘design of experiment’ into Knovel’s basic search to find the background data on this approach.
He sees a useful methodology of full factorial design (33) of experiment approach in Design for Six Sigma – A Roadmap for Product Development (2nd Edition) under section ‘12.3.1 Notation for Two-Level Design’.
On the basis of full factorial (33) design of the experiment approach, the Operations Manager creates 27 designs of experiment points and inspects four samples for each design point. Results of the DOE are shown below. (Click on the image below to zoom in on details)
After comparing the average solder height with all three parameters, he notices that the solder height increases with the increase in release height and printing pressure and decreases with the increase in print speed. Out of these three parameters, release height shows the most significant change while printing pressure is the least significant parameter.
On the basis of DOE results, he suggests the following parameter ranges:
Release Height: 0.5 – 1.0 mils
Printing Speed: 2.0 – 3.0 in/ sec
Printing Pressure: 1.75 – 2 lbs/in2
He inspects the data for four weeks after implementing the above process parameter settings. He then calculates the mean and standard deviation of the solder height using the Mathcad enabled Handbook of Civil Engineering Calculations from Knovel as shown below. (Click to zoom in to details)
The data from the last four weeks shows the solder height mean 4.987 mils with the standard deviation at 0.505 mils and yield has come close to 98%, restoring the processes to optimal yield.
December 10th, 2009 at 6:25 am
I was involved in a similar situation where a Ball Grid Array and a few other surface mounted components caused major head aches for manufacturing.
In similar fashion, some very impressive spread sheets revealing thoughtful calculation were made.
When the dust settled, it was discovered that a line worker loaded a block of solder he found in the back of a storage room.