Authored By:
Andrew Kelley
XACT PCB Ltd
North Shields, UK
Transcript
As PCB designs become more complex with more layers, tighter annular rings and a broader range of advanced materials, understanding the effect of material distortion is critical to maintaining yields.
Determining the inner layer scale factors to compensate for various material movements is also becoming increasingly difficult. With the additional constraints of quick turn, running scale factor test batches are no longer feasible. The first batch made must be delivered to the customer.
Drilling and imaging equipment provide the ability to automatically scale the drill program or image to match each panel manufactured. However, if scale errors have already been made on the inner layers, these errors will be followed throughout the manufacturing process.
This papers shows that variations can not be accounted for by simple look up tables and a more complex model is required to accurately predict the correct scale factors for new product designs.
Summary
As PCB designs become ever more complex with more sequential build up layers, tighter annular ring designs and broader range of advanced materials, understanding the effect of material distortion is critical to maintaining yields. Determining the inner layer scale factors to compensate for various material movements is also becoming increasingly difficult and with the additional constraints of the "quick turn" market there is no longer the luxury of running scale factor test batches - the first batch made must be delivered to the customer.
Latest generation drilling and imaging equipment provides the ability to automatically scale the drill program or image to match each panel manufactured. These capabilities ensure accurate registration of the relevant processes, but if scale errors have already been made on the inner layers, these errors are being followed throughout the manufacturing process and will ultimately be delivered to the end customer. To provide products of nominal dimensions to the customer, the original inner layer scale factors must be accurate and an intelligent compensation system should be used at these processes to reduce or eliminate the effect of variations upon the product received by the customer.
This paper analyses the data from a variety of product designs and constructions made at numerous facilities worldwide and demonstrates the influence of this variety upon the required scale factors. By comparing production results over two years and additional experimentation by the author, this papers shows that the variations exhibited can not be accounted for by simple look up tables and a more complex model is required to accurately predict the correct scale factors for new product designs. The paper will also outline the concept of an intelligent compensation system and how this could be applied to the manufacturing processes.
Conclusions
In this paper we attempt to quantify the influence of discrete features known to affect scale factors. We have shown that the variation defined by the standard deviations indicates that a model utilizing these features alone would not be sufficient to accurately predict inner layer scale factors. As the number of materials and design combinations increases, a simplistic model would become even less accurate.
The typical size of a laser drill capture pad is 75 to 100 microns over the drill size creating an annular ring of 37.5 to 50 microns, assuming scale errors in one axis only, the largest dimensional error across the length of a panel without break out of the drill holes from the pads would be 75 to 100 microns i.e. less than 0.017% scale error. With the lowest observed standard deviation of 0.013% for any of the groups identified, there is at best around 81% confidence of predicting within the required tolerance i.e. the applied scale factors would be incorrect for 19% of cores in the best case.
A more complex model considering many more parameters and influences is required. Such a model also needs to be capable of "tuning" the influence of each factor according to the overall combination of materials and to learn rapidly from only a small set of data for each possible factor.All the data required to develop such a system is widely available to the modern PCB fabricator, but the quantity of data prevents engineering resource from interpreting and reacting in the required time frame. An integrated artificial intelligence is required to automatically enhance prediction models and provide optimum production data in real time.
Initially Published in the IPC Proceedings
|
