PCBs correct first time?

[Greg Neff]

No one in my experience (especially me) ever seems to get PCBs of any
complexity designed correctly straight off, there's always at least one
error on the prototype. They are usually OK the second time round, of
course. How general is this observation?

I've got a feeling that thorough checking of the prototype PCB design
probably costs more ultimately than just getting it made and fixing any
problems subsequently.

Leon

These days our PCBs tend to be on the complex side. We do
double-sided SMD, BGAs, fast signals, 16-layer construction, >800
components, etc. We usually have to spin the board a second time.
Occasionally a third time, but not too often (if we get bit at this
point it's usually due to a signal integrity or ESD/EMI issue). We
get close because we carefully control the design process:

1) We have a library of verified footprints. If we need a new
footprint then it is made and checked against a mechanical sample
prior to it being used in a design.

2) Our schematic libraries include every part number in our part
database. Each part in a library includes the part value,
manufacturer, manufacturer's part number, our part number, and
footprint. Yes, this means that we have a library part for each value
and size of resistor we use. The generation of the parts in the
schematic libraries is where most of our mistakes are made.

3) A schematic is entered that includes only mechanically critical
components. A PCB is designed with these parts placed as per the
specification. A copper-clad piece of FR4 is drilled and routed by
our PCB house. The mechanically critical parts are soldered and/or
glued to this board. This mechanical sample PCB allows verification
of form and fit in the target equipment. It also allows the
mechanical guys to work on the enclosure and wiring harness design.

4) A complete schematic is entered and run through DRC. The schematic
is peer-reviewed. A human-readable netlist (wirelist) is generated.
You would be surprised how many errors can be detected by reviewing
this wirelist.

5) A PCB netlist is generated. The netlist is imported into the PCB
tool.

6) The PCB layout is done, with several reviews along the way by the
design engineer. Any changes required at this stage are carefully
forward or back annotated as required.

7) A BOM is generated from the schematic, and a kit of parts is pulled
to allow building of two or three prototypes.

8) We build the prototype, and then the fun begins...

Breadboarding is a thing of the past for digital designs. It simply
won't work. The last wire-wrap prototype that we did was about 15
years ago.


================================

Greg Neff
VP Engineering
*Microsym* Computers Inc.
[email protected]

 

[John Devereux]

These days our PCBs tend to be on the complex side. We do
double-sided SMD, BGAs, fast signals, 16-layer construction, >800
components, etc. We usually have to spin the board a second time.
Occasionally a third time, but not too often (if we get bit at this
point it's usually due to a signal integrity or ESD/EMI issue). We
get close because we carefully control the design process:

3) A schematic is entered that includes only mechanically critical
components. A PCB is designed with these parts placed as per the
specification. A copper-clad piece of FR4 is drilled and routed by
our PCB house. The mechanically critical parts are soldered and/or
glued to this board. This mechanical sample PCB allows verification
of form and fit in the target equipment. It also allows the
mechanical guys to work on the enclosure and wiring harness design.

Well this bit is cheating!

4) A complete schematic is entered and run through DRC. The schematic
is peer-reviewed. A human-readable netlist (wirelist) is generated.
You would be surprised how many errors can be detected by reviewing
this wirelist.

Interesting; I think I will try this idea.

 

[John Larkin]

No one in my experience (especially me) ever seems to get PCBs of any
complexity designed correctly straight off, there's always at least one
error on the prototype. They are usually OK the second time round, of
course. How general is this observation?

I've got a feeling that thorough checking of the prototype PCB design
probably costs more ultimately than just getting it made and fixing any
problems subsequently.

Leon

I'd guess that we've had two or three throw-away rev A boards in our
20-year company history. We expect to be able to manufacture and sell
the first board, and we formally release it to manufacturing as rev A.
They build it, they inspect it, we test it, and we are embarassed if
anything serious is wrong.

We don't breadboard entire designs. Once in a while we breadboard one
specific circuit, generally because the electromagnetics is too
complex to simulate, or because a part isn't characterized well
enough. If an engineer can't design the product architecture right
without a breadboard, we just fire him.

Correct-the-first-time is a matter of policy and culture, and I try to
encourage it here. Most boards do get revved past 'A', to add
features, replace obsolete parts, or improve manufacturability, but
it's seldom under duress.

John

 

[John Larkin]

I just did a board with only a couple of minor errors:

o A 176 pin QFP with 0.5mm pin pitch. Except it should have been
0.4mm.

o A flash chip with all the address lines off-by-one.

All in all, not my best work...

See my post to a.b.s.e. Luckily, I can fix this with a diode or
something.

John

 

[Frank Bemelman]

See my post to a.b.s.e. Luckily, I can fix this with a diode or
something.

Nah, you have to do better, to win this 'most horrible PCB blooper
contest'.



Wish I had a digital camera some 20 years ago...

 

[Brian]

I breadboard everything.... on a prototype pcb

I may test parts of a design, but the first "whole show" test is done on a
prototype pcb. I hardly bother checking this first board for layout mistakes
in PADS. I use this board as a tool to shake them out. Then I'lll run a
second proto, hopefully w/o errors to check the circuit and layout. 3rd time
I send artwork, its for real.

With SUPER cheap protos available fast and the fact I am always working
multiple projects, this is most time efficient for me. Too time consuming to
check that first board for my own stupid mistakes, and many times a databook
error sneaks in anyway.

No I just put lots of time into checking and fixing what gets
forgotten. It works better if you can set it aside and look at it
later, or get someone else to check it. It's almost always something
really stupid and obvious, like using the wrong width for a SOIC. But
it's probably only a matter of time (knock on wood).


I'd probably go back and check that about 4 times. Those thru-hole
PLCC sockets have a really irritating pin pattern (I remember making
an 84 or 68 footprint). Then there are relays and oddball connectors
with metric dimensions, asymmetrical layout, pinout numbering
including various phantom pins, and the drawing given from the
*bottom*. Yuk.


That's when you get the boards right out in the trash so you don't
have to be reminded of the monumental screw up.


Hahah. I haven't heard that expression in years. Then there was the
guy with only five- but his underwear fit like a glove.

Best regards,
Spehro Pefhany

http://www.speff.com

 

[Brian]

Don't feel bad... the guys here claiming to go from concept to production
with one layout are:

A) Designing simple toasters
B) Have ESP and can predict data book errors
C) Spend the time doing everything we find on a proto on paper (holding
parts up to monitor to verify footprints, heehehe), just spending 25 times
the time we need
D) Do only one design every 3 years
E) Lying

Brian

 

[Frank Bemelman]

Don't feel bad... the guys here claiming to go from concept to production
with one layout are:

A) Designing simple toasters
B) Have ESP and can predict data book errors
C) Spend the time doing everything we find on a proto on paper (holding
parts up to monitor to verify footprints, heehehe), just spending 25 times
the time we need
D) Do only one design every 3 years
E) Lying

You forgot F) And when there was a stupid mistake, it was made
by a co-worker.

;-)

 

[John Larkin]

Don't feel bad... the guys here claiming to go from concept to production
with one layout are:

A) Designing simple toasters
B) Have ESP and can predict data book errors
C) Spend the time doing everything we find on a proto on paper (holding
parts up to monitor to verify footprints, heehehe), just spending 25 times
the time we need
D) Do only one design every 3 years
E) Lying

F) Disciplined.

G) Smart.

H) Profitable. Very profitable.

John

 

[Mike Page]

No one in my experience (especially me) ever seems to get PCBs of any
complexity designed correctly straight off, there's always at least one
error on the prototype. They are usually OK the second time round, of
course. How general is this observation?

I've got a feeling that thorough checking of the prototype PCB design
probably costs more ultimately than just getting it made and fixing any
problems subsequently.

Leon

"Thorough checking" is a culture thing. You have to have a certain
expectation of correctness, because there are hardware engineers whose
mantra is "this isn't the final board, so it doesn't matter".

Missing a deadline or an expected performance level has systemic costs
that aren't always obvious, and usually hurt the whole team.

 

[Mike]

You forgot F) And when there was a stupid mistake, it was made
by a co-worker.

or G) A client who has an attack of featureitis and decides to move
the goalposts in the middle of the prototype phase.

 

[The Real Andy]

No I just put lots of time into checking and fixing what gets
forgotten. It works better if you can set it aside and look at it
later, or get someone else to check it. It's almost always something
really stupid and obvious, like using the wrong width for a SOIC. But
it's probably only a matter of time (knock on wood).

I screwed up my first ever production pcb by designing a VGA connector
upside down (if that makes sense). Didn't read the data sheet properly.
Since then I have never had a wrong footprint on the PCB. What I do now is
print out the overlay with the holes on a decent laser printer and stick the
paper to a piece of polystyrene foam. I then insert every through hole
component into the overlay and lay every single SM part ontop.

 

[Brian]

Thats covered by E)

 

[Brian]

John,

Is toaster control that profitable?

I'm betting you fall under E)

 

[Brian]

I find protos to be cheaper than thorough checking many times. Unless I have
breadboarded the entire design, I often don't find certain errors until the
1st proto is made. These include data book errors, misinterpretation of
data, and wrong assumptions. Often, this first proto also leads to new ideas
for the design (better performance, lower cost) and change is needed on the
pcb. If you aren't finding that you often want to change the way you did
something on your initial design, chances are you have put out a product
that isn't as good as it could be. Occassionally I even rip 2 protos with
differences to similar circuits just to see whats better (Simulation still
sucks). With protos being so EXTREMELY cheap, I can't see why not to use
them this way. Spend 100 bucks on a proto to save $1 on every unit, heck ya!

 

[Frank Bemelman]

John,

Is toaster control that profitable?

I'm betting you fall under E)

Well, we are all just kidding a bit I guess, but when
I see the pictures of John's work, I rank myself as
'toaster-business' category.

I never did anything > 4 layers and with perhaps $75
worth of components on it. And 95% has been 2 layers
anyway. At those levels, a proto does not cost much.

But the point was made, and there is some truth in
it, for some of us.

[snip]

 

[budgie]

I screwed up my first ever production pcb by designing a VGA connector
upside down (if that makes sense). Didn't read the data sheet properly.
Since then I have never had a wrong footprint on the PCB. What I do now is
print out the overlay with the holes on a decent laser printer and stick the
paper to a piece of polystyrene foam. I then insert every through hole
component into the overlay and lay every single SM part ontop.

I glue the laser printout to cardboard (cereal box) and "drill" all
the required holes with a special pin. To avoid static risk, I use
already-dead IC's of the various pinouts. It's a straightforward way
to ensure mechanical fit, and often shakes out a problem or two before
boards are fabricated.

 

[budgie]

Don't feel bad... the guys here claiming to go from concept to production
with one layout are:

A) Designing simple toasters
B) Have ESP and can predict data book errors
C) Spend the time doing everything we find on a proto on paper (holding
parts up to monitor to verify footprints, heehehe), just spending 25 times
the time we need
D) Do only one design every 3 years
E) Lying

you left out the most important part ....

OR

F) brought up in a culture of checking things thoroughly. Goes with
the idea of accepting responsibility for one's own ****-ups, and the
old-fashioned notion that it pays to do it right once rather than
half-right a number of times.

 

[Frank Bemelman]

you left out the most important part ....

OR

F) brought up in a culture of checking things thoroughly. Goes with
the idea of accepting responsibility for one's own ****-ups, and the
old-fashioned notion that it pays to do it right once rather than
half-right a number of times.

I guess you never have software problems either.

 

[Spehro Pefhany]

I glue the laser printout to cardboard (cereal box) and "drill" all
the required holes with a special pin. To avoid static risk, I use
already-dead IC's of the various pinouts. It's a straightforward way
to ensure mechanical fit, and often shakes out a problem or two before
boards are fabricated.

Seems most of us use somewhat similar independently developed methods
to catch as many errors as possible as early as possible.

Does anyone else edit the NC drill G-code files manually to fit the
proto house's standard rack?

Best regards,
Spehro Pefhany