kicad-developers team mailing list archive

[Rick Walker <walker@xxxxxxxxxxxxxx>]

Hi Veronica,

> I wonder if anyone has any insight to this.
> I have vias set with a 5mil annular ring as per PCB makers
> minimum but the PCB maker reports them as 4.859mil and thus below
> the 5mil they want.
> Is this an artefact of some setting (gerber file perhaps) or is
> it a math rounding error internally, or something else?

Vendors often uses a design rule checking program optimized for speed. 
To achieve this they often approximate pads as octagons.  Of course an
octagon is smaller than a perfect circle.  In fact, any non-infinite
sided approximation will be smaller across the facets than a true
circle. 

A second approximation is that they often render the geometry onto a
pixel grid and then look for clearance on the grid by counting
pixels.  There is of course round-off error. 

Basically, it is a hard job to precisely model distance in an efficient
way for ideal circles against polygons. 

Often, if you ask to talk to an actual line engineer, they will look at
the test results and waive the errors.  The sales people often won't
understand the situation well enough to help you. 

To avoid this, I usually don't run my pads all the way down to the
minimum.  If they can do 5 mil, I make it 5.5 mils.  Then you'll never
have a problem.  However, if you really need the clearance for something
tight like a BGA, then you just have to stand your ground and insist to
have the board reviewed by the fab manager. 

kind regards,
--
Rick Walker