OK, I just got done talking to Spark Fun directly. They verified to me that they can only do 8 mil spacing “sometimes”. There are some discrepancies between what the DRC bot chaecks against and the actual gerbers produced. Therefore for anything approaching 8 mil spacing their service will be hit or miss.
I thought this important enough to place in its own thread instead of buried in another one.
If you plan to do a design with 0.5mm pitch parts, you should look elsewhere, or at least plan on “just trying out BatchPCB”. The chances are high your design will not be manufacturable here.
If I can manage 8 mil tracks on my home-made PCBs (made in the kitchen sink) with 0.5 mm lead spacing connectors I can’t see why SFE’s supplier is having problems!
Hmmm… something must have changed. I seem to remember them setting 8 mil as their limit, but Gold Phoenix being able to stretch that down to 6 or 7 mil for wiggle room.
Here’s something to think about. How are they doing the 0.4mm pitch Hirose connectors on the camera breakout boards (to be released someday)?
The only thing I have problems with is the ground pour, that I set at 16 mil. I have had a few different boards with 8 mil spacing (16 for the pour) and they all passed n(and worked, unless I messed up).
8mil is a limit to keep costs down. While Goldphoenix can make boards down to 6mil, I believe they charge extra if you drop under 8. That means if only one design on a panel needs less than 8mil sizes, the entire panel gets charged the higher rate.
I was hoping that as the PCB pooling service got bigger, there would be enough demand to start accepting finer pitch boards. I have a couple boards that make use of 0.3mm pitch connectors.
As for the 0.5mm parts, you shouldn’t have any problem making them through BatchPCB. 0.5mm is 19.68mils which means you can use 10mil pads and traces and still have 9.68mils of clearance. I haven’t had any problems through BatchPCB when using 0.5mm pitch parts.
As mentioned, some designs make it, others do not. It all depens on what the temperment of the DRC checker is.
You pad dimensions are nearly identical to mine. In fact, it is a 10 pin part, and only 4 of the pins on the latest DRC are shown as design violations. I have tried submitting the design with just this part placed.
Different positions lead to different rule violations.
Perhaps the online DRC has a grid it is using for measurements and if you are lucky enough to land on the grid, you can pass with 8 mil spacing.
GP quotes 6 mil spacing on 1 oz outer layers, 7 mil for 2 oz. I’ve mentioned this before - the 8 mil limit seems overly restrictive but it may be due to practical experience. I recall reading somewhere on their site that it was .2 mm which is 7.8 mil but could not find it when I looked just a minute ago.
OK, I just got done talking to Spark Fun directly.
I knew I should have told you we don't do phone support for BatchPCB :) Our website says don't call about BatchPCB, but ya do anyway...
What you talked about on the phone and what you post are two different things. You called and inquired why your designs with copper ground pours were failing the DRC bot when your Protel software DRC was passing them without problems. You also stated that your 0.5mm footprint was failing DRC. I should not have taken your call about this, but I obliged. It is up to you to figure out how to make a 0.5mm footprint not violate an 8mil DRC. Please refer to the forum - you are not alone.
The DRC bot checks against 7.6mil miminimums. We actually give you guys better tolerance than we advertise. Why? Because GP actually does all the way down to 5mil. They can go down to 5mil, but like any fab house, their yields go down as the tolerances go down. So 8mil is the cheapest price point. We regularly use GP for 6mil designs without problems. Their QC and testing has been pretty much dead on. The proto panels that come back at 8mil have very nice quality - there is no ‘sometimes’ about it.
Gerbers are interpreted different ways by different software. This has been our pain in the ass since the beginning of BatchPCB. GP uses CAM850 to DRC our panels. We use Protel. They simply do not agree. One would think that gerbers are just numbers, but how different software interpret them is the killer. I’m sure your board passes your 8mil DRC just fine, but it doesn’t even pass our 7.6mil check. Time to dig a little deeper.
I recommend you layout a very simple board with 8mil traces and spacing. Just a few traces, no more than a part or two and send it through the DRC bot. Then make it more complicated. I’m sure you’ll find that your 0.5mm part or ground pours are part of your problem.
OK, I just got done talking to Spark Fun directly.
I knew I should have told you we don't do phone support for BatchPCB :) Our website says don't call about BatchPCB, but ya do anyway...
You did, and I did anyway My bad.
sparky:
What you talked about on the phone and what you post are two different things. You called and inquired why your designs with copper ground pours were failing the DRC bot when your Protel software DRC was passing them without problems. You also stated that your 0.5mm footprint was failing DRC. I should not have taken your call about this, but I obliged. It is up to you to figure out how to make a 0.5mm footprint not violate an 8mil DRC. Please refer to the forum - you are not alone.
Not sure where you are getting different issues. Same issue here. I have a board which has greater than 8 mil widths and spacing and it is not passing the DRC, same issue as the call to you. When I mentioned a 0.5mm you said clearly and I quote, “Oh no we will not do that.”
sparky:
The DRC bot checks against 7.6mil miminimums. We actually give you guys better tolerance than we advertise. Why? Because GP actually does all the way down to 5mil. They can go down to 5mil, but like any fab house, their yields go down as the tolerances go down. So 8mil is the cheapest price point. We regularly use GP for 6mil designs without problems. Their QC and testing has been pretty much dead on. The proto panels that come back at 8mil have very nice quality - there is no ‘sometimes’ about it.
I wasn’t referring to the ability of your fab house to ‘sometimes’ produce designs and ‘sometimes’ not. I was referring to your DRC bots propensity to ‘sometimes’ pass >8mil designs and ‘sometimes’ not. As mentioned, the design passed Protel’s DRC as well as two other online DRC checking services.
sparky:
Gerbers are interpreted different ways by different software. This has been our pain in the ass since the beginning of BatchPCB. GP uses CAM850 to DRC our panels. We use Protel. They simply do not agree. One would think that gerbers are just numbers, but how different software interpret them is the killer. I’m sure your board passes your 8mil DRC just fine, but it doesn’t even pass our 7.6mil check. Time to dig a little deeper.
Agreed. For the time being I had to use another design house (again with 8mil tolerences) that accepted the design. Other than opening it up to 10mils (for which the pad size of the 0.5mm part gets pretty small) I am not sure what else to dig for.
sparky:
I recommend you layout a very simple board with 8mil traces and spacing. Just a few traces, no more than a part or two and send it through the DRC bot. Then make it more complicated. I’m sure you’ll find that your 0.5mm part or ground pours are part of your problem.
-Nathan
It is clear it is that part that the DRC bot is incorrectly flagging, thus the call in the first place.
DRC is hard to do, really hard. I tried to write them a better one once - it parsed gerbers fine, but the actual DRC part is extremely difficult to do properly. [Someday I’ll finish that for ya ;)]
Protel is a really big name.
Sparky [and co] don’t have any control over how protel works / does not work.
For the price you’re getting for these boards, [at least I] can live with a bit of fiddling required to make them pass DRC.
I submitted some files to Olimex once and they failed their DRC, although they were obviously OK. I downloaded an evaluation copy of the same checking software that they used and established that it had a bug, which was confirmed by the software company.
My bad.