The Quest for Perfect Design Data Packages
There’s an ongoing problem in the PCB industry: fabrication shops are receiving incomplete or inadequate design data packages, leaving manufacturers scrambling to fill in the blanks. For a quick-turn prototype shop like Washington-based Prototron, with over 5,000 customers and up to 60% of orders coming from new customers each month, that can add up to a lot of wasted time and effort just in the quoting stage. Dave Ryder, Prototron president, and Mark Thompson, engineering support, delve into this continuing issue and more.
Barry Matties: Dave, can you give us a little information about Prototron?
Dave Ryder: Prototron was founded in 1987. We saw a need in the Northwest for a quick-turn prototype shop. The area had needs that just weren't being met by most of the traditional production houses. We were welcomed into the community right out of the gate, which was a pleasant surprise. From there, we've continued to grow over the years. We acquired a local competitor here in 1995, which greatly added to our list of tricks in our bag. We picked up the shop in Tucson, Arizona, in 1999, which has certainly been a successful addition to the Prototron family. We're in year 32 here, and we're motoring down the highway.
Matties: A lot has changed in 32 years, hasn't it? Yet we still use hot air leveling [laughs].
Ryder: So true. Technically, a big challenge for us was finding a fax machine back in the day.
Matties: Mark, what's your role here at Prototron?
Mark Thompson: I work in a pre-engineering and signal integrity capacity. As a customer liaison, if there are any CAM concerns or issues after the data packages have been released, then I'm the guy that gets a hold of the customers and bugs them about what needs to be done to get the data packages fixed and make the boards more manufacturable.
Matties: And Mark, you are the author of the eBook The Printed Circuit Designer's Guide to… Producing the Perfect Data Package. Why is this topic so important that you wrote a book about it?
Thompson: I’ve spent the last 25 years reviewing incoming data sets, even from the largest customers who you would think have excellent programs to review these data sets before output, but the data is still seriously flawed. It’s very simple things: missing drill files, drill counts that don't match the drawings, referencing impedance tray sizes that don't exist, layer counts that don't exist and providing stackups with dielectrics that aren't possible—any number of things like that. We spend a fair amount of time in the front end with the design data making the part producible.
Matties: Is there a general attitude, “It doesn't cost me anything more to send incorrect data. I send you the data file, and you'll sort it out, right?”
Thompson: Yes, but by continuing to educate our customers, at some point, they will realize that this is a benefit to them; if they could send us a cleaner data package, they would get things faster. They're going to get their quotes back faster, and a more accurate board in the end. They're going to have better functionality with their board if the board manufacturer understands what their end result is supposed to be.
Matties: One of the things that I hear from many people is that 99% of every order that comes into a PCB shop needs some sort of design data adjustment. Even though you tick off all of those benefits, it's not enough for people to change.
Thompson: There needs to be an impetus for folks to change. And as you said, if you give them everything right out of the gate, then what's their impetus to change? I hearken back to the fact that we're attempting to educate our local customers. We have over 5,000 customers, and a lot of them are local, especially in the space industry. Many new space companies startups have opened around this area, and they have brand new folks. They have fantastic scientists and engineers. But as far as practical knowledge of the ins and outs of building a PCB and then having a tangible functioning piece of hardware when all is said and done, that's hugely lacking in this industry right now.
Matties: Do they need to know that to provide you the data you need to build the board?
Thompson: I think so. We spend a fair amount of time doing presentations for our customers to do exactly that. You don't just throw me out a dielectric drawing on a material that smacks of clock traces and differential pairs and is just rife with that stuff. They're throwing it to us as a PCB manufacturer with the assumption that, "These people are dumb. They don't even know what they're looking at." However, most of the folks in our CAM department have well over 25 years of experience. If they see differential pairs and clock traces, they know that the board is controlled impedance. If they see a dielectric control or stackup, even if there are no notes regarding any controlled impedance whatsoever, we know what it's supposed to be doing.
And if we see a huge mismatch, what do I do, Barry? I might think, "What if they don't call it out? Does that imply that they don't care?" We don't make that assumption. I go through it and do a calculation. Even if I don't know what the threshold is, if it's an odd number like 107 ohms for a 90-ohm threshold, I usually can communicate with the customer and say, "Is that really what you were shooting for?" And 99% of the time, they come back and say, "No, I wasn't. And why is there this mismatch between what I'm calculating and what you're calculating?"
The big answer is this: We use a field solver. Most of our customers just use an impedance package that's tied to their CAD package. If it's embedded in their CAD package, it's not intuitive enough. It's not like a field solver, such as the Polar Instruments tools that we use. I spend a lot of time daily defending that software package. Our customers will say, "That doesn't match my calculations. Can you tell me why?" Fortunately, what they do most of the time is send me a screenshot of what system they're using, and I'll look at it. Right away, I’ll say, "Here are a couple of problems right out of the gate."
First, they're modeling it as a rectangular shape. Etchant doesn't work at perfect 90° angles. It's a trapezoid. Ultimately, there's a crest and a foot to every trace, and that's a big part of it. Second, those systems usually can’t model things like solder mask over the top of the traces. When everything was 8/8 spaces and traces 10 years ago, an additional one or two mils of polymer plastic over the top of the trace didn't mean anything. Today, we're talking about 0.1-millimeter lines and spaces. I guarantee an extra two mils of plastic over the top of the trace is going to affect the impedances.
Matties: So, the designers just may not have the right tools available.
Thompson: I think the tools are adequate; it's the knowledge of the tools that is lacking in many cases for folks who are brand new in the industry. Even if you have a mentor who brings you in and says, "This is how this software works," they’re only going to tell you what they know about that software. They aren’t giving you the definitive use of everything inside that software.
A great example that we get all the time is a netlist mismatch, which is a huge deal. They'll send us a data set, and we come back almost immediately because it's one of the very first things we do in the process—organize your files, set up the drill strings, and run your IPC netlist. If we come back with a mismatch, the very first thing I do is I communicate it back with you. We try really hard not to try to interpret them.
Some things are obvious. If it's a net zero or one or an A-ground to D-ground short, it's most likely intentional, and we approach it that way. If we get 16 brokens and they're 16 castellated pads around an outside periphery, you've heard me say it before; we make that a foregone conclusion. When can I not do something like that? What if I come up with over 200 broken nets or open nets? At that point, I'm looking at their data for whole sections of missing copper pour or plane because that's a lot of connections not being made.
Matties: So, we know the problem is data is not getting here correctly. There's maybe some education needed. But is this a problem that can be solved, or is it something that we just have to rely on the way we do it?
Ryder: I think that you'll never completely resolve that particular issue. Certainly, with repeat customers, we steer them in the right direction once they have a better understanding of the challenges of the manufacturing process. They get it, and it doesn't go away.
Thompson: A great example of some successes in that respect is that we have been able to nurture many of our customers and get that sort of consistency from them. When their data comes in now, it takes a fraction of the time for us to review it and get it out into manufacturing.
Matties: When we talked about the data package, there's the design element. Some of it may be a poor design, but then there's the matter of providing the correct data set.
Thompson: That's correct. An output package should be a template. It should be all done by rote. Common sense would dictate we have to drill the boards to be able to get through-hole continuity, so we have to have an NC drill file. We're going to need image layers for all layers that are involved. If it's a 10-layer board, I need 10 image layers. If it's going to take solder mask, I'm going to need solder mask image layers. If it's going to take silkscreen, I'll need silkscreen image layers. Whether it's going to be sent to a local assembler or not or if we're panelizing it, I'm going to need any paste files associated with the job. Beyond that, any PDF drawings, or anything really unique about the job—such as unique reference planes for impedance scenarios or unique alternative surface finishes and those kinds of things—need to be called out, which is always helpful. That’s the nuts and bolts of what we need. It's the accuracy of this stuff that we see coming in.
Matties: Any designer, whether they have firsthand manufacturing knowledge or brief knowledge in manufacturing, should be able to provide you a good data set.
Matties: What's the obstacle for them to do that?
Thompson: The only thing I can think of is a lack of design reviews. They could be crunched on time. Maybe they've gone through a service bureau to do their circuit boards, and they're really cutting it close on time-to-market issues, and they didn't do an adequate review of the data set once it has been outputted. Nine times out of 10, when I said, "You didn't provide us with an NC drill file," they said, "Oops, I'm so sorry."
Matties: It's really just a simple checklist. How much does it take in terms of time for someone to review the data set?
Thompson: It depends on the complexity of the board. All I could think is that perhaps they're not operating off a checklist in some cases. A Class-3 IPC-6012 job dictates that I have to do an IPC netlist compare. They didn't provide me a netlist or an NC drill file, and they're missing all of their inner layers and vias. Sometimes a customer will come to us with preliminary data and ask for a preliminary quote on a board when it hasn’t been filled out. It will have 115 holes with all of the tooling holes but have no vias on it. We'll give them this arbitrary charge, it will come in, and there will be 9,000 additional vias associated with that job. We see that kind of stuff too; it's not at all uncommon.
Matties: How much time would designers save you if the data set was accurate to start?
Thompson: It would save our inside salespeople and me a lot of additional time and work.
Matties: Are we talking 30 minutes an order?
Thompson: It could be up to as much as three hours per order, depending on the severity or the amount of the issues.
Matties: Let's say the average is an hour for easy math because you have to go through all of this just to make a quote, and it doesn't even necessarily translate into revenue.
Thompson: That's right. It may not even come to fruition.
Matties: How many quotes would an average board shop do in a month?
Thompson: For every inside salesperson you have, you should have at least one per hour their entire day. If they work eight hours a day, they should have a minimum of eight per day. If you have three inside salespeople, you should have 24 jobs a day.
Matties: You're spending an hour per job fixing the data with 24 quotes a day. That would be 24 hours of waste. Why do we put up with it?
Thompson: It's hard to say because we are still considered a third party. For many years, the fabricators were seen as those who built us the boards. “They don't know anything about dielectrics, stackups, or the functionality of my board. They don't know anything about electronics.” All of that is untrue. However, that was the thought process many years ago as well.
Matties: They must have faith that you do know this stuff or they'd be sending the right data. I think they're relying on you to provide them with the right data. Is that cost factored into the board or is that just a part of the overhead?
Thompson: I'm assuming it's part of the overhead because I don't think we factor that in. We've always said our services are above and beyond; it's a value-added service. Whether or not your board ever comes to fruition at Prototron, you still have my time to be able to run calculations or give you engineering assistance.
Matties: It's a problem that's not going to go away unless there's a collective solution. When I hear multiple times that 99% of all the work that comes into any given PCB factory is not accurate, it’s a big problem.
Thompson: It's alarming, isn't it?
Matties: And maybe it's just the cost of making a sale, but if the data package was right to begin with, on average, how much time would you save out of that hour if it was correct?
Thompson: Most of it. It would take us a fraction of that amount of time.
Matties: If I'm running my company, this would be a top priority.
Thompson: And it has been at Prototron. Hence, the reason for writing the eBook and my drive to educate the customer through these presentations.
Ryder: We try to limit how much effort we put into it before actually getting a purchase order, but you can't necessarily scale that back because ultimately, we have to look at it like it's actually going to be an order for us. We have to look at it like we're going to build it.
Matties: Do people look at this on their P&L statement as a cost of sales?
Matties: Until you have the order, you're still in the selling process.
Ryder: That’s correct.
Matties: And it's a cost to sales. We can lower our cost of sales by training people, incentivize it, and do a lot of things, but it's not an order. Once it's an order, then it becomes engineering value-add. The value-add is them saying, "They will solve my problems for me, and I may or may not even buy the boards from them."
Ryder: We certainly see what happen.
Matties: Gentlemen, this has been very interesting. Thank you both.
Ryder: Thank you.
Thompson: Thanks, Barry.
To download your copy of Prototron’s eBook, The Printed Circuit Designer's Guide to… Producing the Perfect Data Package, written by Mark Thompson, click here.