0.031" Thick FR4 PCB

[Jamie]

Some day I might build a particularly high performance planar transformer,
using a PCB stack thicker than it is wide.

A number of manufacturers are offering "heavy copper" these days (> 10
oz.), pretty cool.

Tim

I kind of wished we did that way back with one of our older projects. It
was only a double sided board however, we didn't use heavy copper and
many of the traces are very thin to say the least, just something we
over looked or didn't think it was important at the time.

These items are used in some areas that has modest moisture and PCB
failures are starting to show due to trace deterioration.

Jamie

 

[Robert Baer]

Yep. That's the reason we're starting to go to less than 1.5mm (1mm,
IIRC) boards; can't do the vias on .5mm BGAs. Don't know that we're
going to actually do any products with .5mm BGAs but we need them for
prototypes.

Isn't it rather stupid to do a "prototype" using parts that is NOT
gong to be used in production?

 

[Tim Williams]

I kind of wished we did that way back with one of our older projects. It
was only a double sided board however, we didn't use heavy copper and
many of the traces are very thin to say the least, just something we
over looked or didn't think it was important at the time.

Was testing a new power board today: 80 mil wide, 2oz traces carrying up
to 8A. Looks really cool on the FLIR, turn on the power and the traces
light up like neon signs before the glow from resistors and transistors
overwhelms the scene.

Tim

 

[[email protected]]

[email protected] a écrit :

I have a small board (8mm dia) that's top&bottom stuffed with 0402,
0201, uBGA,... and we also needed some accuracy on the max board diameter.

8mm?! That's not big enough to have setback to put one part in the
middle! ;-)

Usual separation was... scary, at least, so we ended to laser cut them:
fast, accurate, no strain. What's more to be desired? Ah, yes, low cost...

Yeah, that cost thing will get ya'. (Mechanical) routers are used
too, but also expensive.

 

[[email protected]]

Isn't it rather stupid to do a "prototype" using parts that is NOT
gong to be used in production?

Not at all. The parts are first available in .5mm pitch BGA. Several
months later the versions intended for our market hit the streets. We
get at least six months head start by prototyping with the commercial
parts. The software guys like the extra time to play. When the real
parts are available, it's a quick, low risk, spin.

 

[[email protected]]

the problem is in the other direction, when you get very
thin dielectric you can't make a trace that is thin enough
to get up to 50R

Wasn't a problem, though there weren't many cost issues, either. ;-)

 

[Fred Bartoli]

Tim Williams a écrit :

Was testing a new power board today: 80 mil wide, 2oz traces carrying up
to 8A. Looks really cool on the FLIR, turn on the power and the traces
light up like neon signs before the glow from resistors and transistors
overwhelms the scene.

I just repaired my daughter's SONY mini hifi (broken belt).
As I had the FLIR cam ready on hand I just pointed it to that, well, POS

At rest, some PCB portions around small heatsinks (regulators I think)
are more than 90 celsius!
I changed some 'litycs that are beginning to leak, but same temp.

Crap.

 

[Allan Herriman]

Right. If you can etch 4 mil wide traces, a stripline embedded in FR4
needs a plane separation of about 12 mils (6 above the trace, 6 below)
to get 50 ohms. That suggests big trouble for, say, a 16 layer board.

I just don't understand things like 20 or 30 layer boards, unless they
are really thick, or made of low Er stuff, or have insanely low
impedance traces. Anybody done one? What was the stackup?

Yep.

Thick board for passive backplane application in something that went into
a phone exchange. This was from the days when buses were parallel, so
there were lots of traces between the press fit connectors. It needed a
lot of planes to keep the crosstalk down. I don't recall the exact
stackup but it was pretty much a regular board with normal thickness Cu
and prepreg, just more layers.

The thick board also helped with mechanical stability.

I don't see things like that these days. PC motherboard costs have
forced buses to be serial so it's now possible to move many Gb/s around
with only a modest number of PCB layers.

Regards,
Allan

 

[MrTallyman]

Yep.

Thick board for passive backplane application in something that went into
a phone exchange. This was from the days when buses were parallel, so
there were lots of traces between the press fit connectors. It needed a
lot of planes to keep the crosstalk down. I don't recall the exact
stackup but it was pretty much a regular board with normal thickness Cu
and prepreg, just more layers.

The thick board also helped with mechanical stability.

I don't see things like that these days. PC motherboard costs have
forced buses to be serial so it's now possible to move many Gb/s around
with only a modest number of PCB layers.

Regards,
Allan

You've obviously not seen a modern VME or Compact PCI chassis with
modern connector technology.

PCB backplanes are at 0.090" and even 0.120" or possibly even more
Mainly for rigidity since huge cards are inserted into dense connectors
(far more dense than in telco days) against them on both sides (front of
chassis and back of chassis)

Telco was 40 or 80 pins per connector, max... on a good day.

These are like 250+ pins per connector with two per card plus power
sections.

Motherboards were not "forced" to be serial, and it never had anything
to do with "costs" either.

Hard drives and their interface technology is from THEIR industry, and
THEN the PC Motherboard makers follow suit.

SAS get placed on a motherboard because SAS drives exist, not because
MOBO makers wanted to have lower costs.

Stop making shit up, boy.

 

[[email protected]]

Clark boards for the IBM 3081 through 390 series machines were about 100
layers, to allow escaping all the traces from a big thermal conduction
module (TCM).

With twisted pair overflow wires. Sometimes thousands per board. That
was a neat machine to watch.

 

[MrTallyman]

PCI Express didn't evolve in the drive industry.

I never said it did, you dippy twit. Learn to read.

It was a replacement
for PCI.

Ummm, no. It is STILL tertiary to the main PCI bus, which is STILL the
master. PCIe was made as the replacement bus for the AGP video bus, It
was an evolution of the PCI bus. The AGP bus simply wasn't quite quick
enough.

 

[MrTallyman]

PCI ran out of steam on throughput because of skew issues. And it was
too expensive, in terms of pins and copper and PCB layers, both on the
mobo and the connectors/add-in boards. The lanes of PCIe are tiny
pairs, as few as one pair to a slot, and skew between lanes doesn't
matter.

Motherboards still have PCI slots, and PCI cards are still sold
EVERYWHERE computer cards are sold. PCI is STILL the master bus on the
board, and all else is tertiary to it, with PCIe being at the same level
with direct hooks to the memory bus. It is what replaced the AGP
graphics bus, which is what was not robust enough and it gets direct
hooks, but all other busses are tertiary to and controlled by the PCI
bus.

PCI and PCI 32 still exist, but are dying out because their capacity
is not up to modern data rates.

Super Micro (among others) still puts PCI 32 slots on their
motherboards and there are still PCI 32 SCSI controllers and the like in
the channel, and PCI itself is still found on nearly all motherboards.

 

[Allan Herriman]

Telco was 40 or 80 pins per connector, max... on a good day.

The particular system I'm thinking of used high density 2mm pitch
connectors. They looked a lot like these:
http://www.te.com/catalog/cinf/en/c/10453/721
I think there were about 200 pins per connector with up to five
connectors on a card. Most cards just had one connector.

Mustn't have been a good day, I suppose.

Regards,
Allan