[gta02-core] Board parameters update

Werner Almesberger werner at openmoko.org
Mon Aug 31 11:19:09 CEST 2009


Rene Harder wrote:
> I think we should try to stay as much within common standard specs as
> possible because the goal of gta02-core is to create a prototype and we
> are not going for production right away. Custom specs always means that
> our board has to be processed individually which adds up to higher cost.

Agreed. The easier we can make the board to produce, the more choices
and the lower the price. My estimates are based on the assumption that
the simpler circuit will help us to make a slightly less demanding
board than Openmoko's original, but that there won't be any
revolutionary simplifications.

If we can "dumb" down the layout drastically, that's great. It's just
not something I'd feel comfortable assuming when making cost estimates
or when evaluating an SMT site.

> Is 4 mil really necessary? Most fabs go down to 5-6mil as min.
> Trace/Space width.

The track width is constrained mainly by the BGAs. If the 2442 is our
most demanding component, then 5 mil may be enough: the ball to ball
pitch is 0.5 mm, with a ball diameter of 0.3 mm, footprint equal to
pad size.

As far as I can tell, it's unavoidable to route diagonally between
balls (e.g., B2, B25, etc.), so we have to fit a trace into
0.5 mm * sqrt(2) - 0.3 mm = 0.4 mm or 15.7 mil.
5 mil would be enough for this.

> I think for a prototype a regular chemical or HAL tin finish might be
> sufficient or are the any specific reasons why we need ENIG.

I have to admit that I'm not an expert on surface finishes. I did a
bit of Web research a while ago and found the following:

- HAL/HASL: cheap and easily available but with a relatively uneven
  surface. A potential problem for fine-pitch BGA. (There, we need
  better than 99.9% accuracy, or each of the boards will end up with
  some random flaw or another.)

- OSP (Organic): inexpensive and with great surfaces, but unstable
  and demanding on process parameters. With OSP, I would mainly be
  worried about test pads oxydizing and becoming unsolderable. It
  probably also constrains the choice of SMT fabs. OSP seems to be a
  good choice when optimizing the board cost over large quantities.

- ENIG (Nickel-Gold): the standard process for anything demanding.
  Old and well understood. Boards can take some abuse, such as heat,
  storage, or oxygen. Needs two process steps and is more expensive
  than the rest.

- Immersion Silver: a relatively new coating that competes with ENIG.
  Less expensive but susceptible to oxydation. Suitable for
  fine-pitch packages.

So ENIG seems to be the safest choice, particularly if we assume that
we may have to do significant rework. HASL doesn't seem to be suitable
for small-pitch, or it's at least risky. OSP looks more like the thing
you consider for mass production. Silver may be a possibility. I'm not
sure how delicate the handling is, though, and how quickly oxydation
would become a problem with test pads and unpopulated footprints.

> I guess 1080 woven-glass with 65% FR4 epoxy laminate is pretty much
> standard everywhere so shouldn't be too difficult to get. About 1078
> prepreg im not sure, haven't seen it anywhere in the specs so this might
> be only possible within custom specs.

The board is quite thin, only 1 mm. That may complicate it somewhat.
Going to 6 layers should bring us closer to standard thicknesses.
Besides that, I think we can adapt the layout to a wide range of
parameters.

> Are we planning on population all components in a fab? 140 unique
> components, that's quite much and I think this means an increase of
> setup costs for the placements. So maybe it's worth to solder some
> unique components our own, if that would reduce the costs.

I would leave out the shields (they can easily cause trouble and get
in the way of rework). If any of the electromechanical components
cause trouble, manually soldering them may be an option.

One thing to consider is that we need to have everything we want to
test at the SMT site on the board. Otherwise, the boards cannot be
tested on-site.

> Another thing, for the assembly fab we would need a square/rectangle
> board so we should keep in mind that we need to panelize the board with
> an additional outer frame.

Yup. That has to be specified by the fab.

> A serious problem are blind and buried vias though. They are our killer
> for any standard specs, if I'm not mistaken they require a huge amount
> of extra steps between every lamination process which is impossible to
> do within standard spec. So best would be to stick to PTH vias if that's
> not possible our cost will be probably more than doubled.

Hmm, I'm not sure we'll be able to pull this off with through vias.
Also, don't they increase the EMI risk by spreading signals all over
the place ?

> At the end, i think we should consider that this will be *only* a
> prototype and not a usable phone, so we don't need a lot of gimmicks
> which are not absolutely required.

Yes, but on the other hand, we wouldn't want to use a very fragile
processa that introduces problems we could have avoided. There's this
moment when you try to power up a new board and nothing works. It
really helps then if you don't have to suspect every single solder
joint to be a traitor to the cause ;-)

- Werner



More information about the gta02-core mailing list