BGA's are killing my project

"Arlet" <usenet+5@ladybug.xs4all.nl> writes:
> BGA solder problems can also be related to the PCB design. Make sure
> your pads have the recommended size and shape. Vias in the pad, or
> very near the pad can also cause problems by wicking the solder away
> from the pad and into the hole.

What about finish?  I mean, HASL vs flash vs immersion vs whatever?

Also, what about adding solder paste to the pads, or flux?

Inquiring minds want to know...

Hello John,

> 
> My project uses several BGA parts.  We're in the second board rev stage
> (rev 2).  Our board manufacturer doesn't seem to be able to 100%
> reliably solder those parts down.  It makes SW debugging troublesome
> because one never knows if the SW or the part is misbehaving.  We've
> had 15 boards built and I estimate 20% have BGA problems.
> 
> I'd like to hear other's experiences with and solutions to BGA
> soldering reliability problems.  I have a few questions to kick off the
> discussion:
> 
> Is it a manufacturer process problem that can be totally fixed with
> some soldering process improvement? (Corollar: Is my board builder not
> up to the task?)
> 
> Is this a natural and expected problem that I just have to live with
> (result = reduced board yield)?
> 
> How much does the lead-free directive contribute to the problem? (My
> rev 2 boards we're built using no-lead solder but rev 1 boards were
> built using leaded solder and had the same BGA problems.)
> 
> Does X-ray inspection *really* identify badly soldered parts 100% of
> the time?
> 
> What is a typical delivered board BGA solder failure rate that you guys
> will accept?
> 
> Thanks for sharing your opinions and experiences.
> 

I don't use BGA if at all possible so I can't recommend an assembler. As 
others have said the first thing is to look whether your board layout 
and fab instructions are top notch and ideal for each BGA location.

If they are then my suggestion would be to find an assembly house that 
has a good track record with BGA. Maybe post in sci.electronic.design, 
"Recommendations for BGA assemblers" or similar. There are a lot of 
folks who have to deal with BGA. You could also call the sales guys at 
the manufacturer of your BGA chip. They know good places and if not they 
can certainly find out via a few calls to their other clients. After all 
they have a vested interest in you sticking with their chip and not 
looking for alternatives.

IME when there is a reliability problem with an assembler and they can't 
immediately point to QC test results and say "aha!" then chances are it 
won't get better fast enough.

Oh, and make sure the engineers at the assembly house didn't take the 
liberty to "improve" your layout. I had that happen a month ago where 
all my prototypes failed. Every one of them. Turned out they had 
increased some clearances and inadvertantly split a power plane island 
into four floating chunks.

-- 
Regards, Joerg

http://www.analogconsultants.com

Joerg wrote:

> 
> Oh, and make sure the engineers at the assembly house didn't take the 
> liberty to "improve" your layout. I had that happen a month ago where 
> all my prototypes failed. Every one of them. Turned out they had 
> increased some clearances and inadvertantly split a power plane island 
> into four floating chunks.
> 

I meant at board fab. That's where it happened, not at the assemblers.

-- 
Regards, Joerg

http://www.analogconsultants.com

Jim Stewart wrote:
> PeteS wrote:
> 
>> Arlet wrote:
>>
>>> johnspeth@yahoo.com wrote:
>>>
>>>
>>>> My project uses several BGA parts.  We're in the second board rev stage
>>>> (rev 2).  Our board manufacturer doesn't seem to be able to 100%
>>>> reliably solder those parts down.  It makes SW debugging troublesome
>>>> because one never knows if the SW or the part is misbehaving.  We've
>>>> had 15 boards built and I estimate 20% have BGA problems.
>>>
>>>
>>>
>>>
>>> BGA solder problems can also be related to the PCB design. Make sure
>>> your pads have the recommended size and shape. Vias in the pad, or very
>>> near the pad can also cause problems by wicking the solder away from
>>> the pad and into the hole.
>>>
>> A specific piece of board design that will _kill_ you in BGAs is the 
>> pad. You really _must_ use non Solder Mask Defined pads if you want 
>> high yield (and I define that as 100% - I don't accept failures due to 
>> failed soldering).
>>
>> I once worked at a company that had soldermask defined pads under a 
>> pretty large (768 balls) BGA, and the yield was horrendous. Went to 
>> non-SMD and the yields went way up.
>>
>> I would be very surprised, assuming the layout is properly done and 
>> the CM has edxperience with BGAs, to _ever_ see a BGA failure rate 
>> higher than that for TQFP for soldering issues. BGAs are _easier_ on 
>> the process.
> 
> 
> Could you explain the difference?

Not sure which difference, so I'll do both :)

A solder mask defined pad is simply that; the pad opening is defined by 
solder mask, which covers some part of the pad. A non-SMD pad is defined 
by the pad itself, and no solder mask is in contact with it.

On a solder mask defined pad, there is, by definition, a raised 'lip' 
through which the solder ball has to flow for soldering, and this can 
cause no connection at all to high impedance connections (which are 
really frustrating - the JTAG tests will all pass, but nothing works at 
speed).

There are some really good app notes on this; I'll see if I have any. I 
seem to remember app notes from Xilinx, Altera, TI, Intel and Micron at 
the least.

As to BGAs being easier on the process - on SnPb the solder ball wicking 
effect actually 'pulls' the BGA onto the pads. The placement accuracy 
for a BGA is 0.5 pad spacing because of this, which is less accurate 
than final soldered placement accuracy and very easy to achieve with 
automated pick'n'place.
Whether this holds true in a non SnPb world has yet to be seen.

For a TQFP, because there are pads on each side, the wicking effect 
doesn't work this way (the forces apparently cancel) so the placement 
accuracy has to be the same as the final soldered placement accuracy. 
For a highspeed part (to minimise impedance offsets) this may be as low 
as 0.1 pad spacing, which may itself be 0.5mm. That's a very tight 
tolerance, and very difficult to achieve.

Empirically, I have also had many more problems with cold joints, 
soldersplash and solder shorts with TQFPs than with BGAs. I've 
personally designed many dozens of boards with both and I have had very 
few problems (except the non-SMD issue) with BGAs. Most of my solder 
problems are with TQFP and QFN (and that's a process issue although the 
process is harder to get right the first time for those).

Cheers

PeteS

PeteS wrote:
> Jim Stewart wrote:
> 
>> PeteS wrote:
>>
>>> Arlet wrote:
>>>
>>>> johnspeth@yahoo.com wrote:
>>>>
>>>>
>>>>> My project uses several BGA parts.  We're in the second board rev 
>>>>> stage
>>>>> (rev 2).  Our board manufacturer doesn't seem to be able to 100%
>>>>> reliably solder those parts down.  It makes SW debugging troublesome
>>>>> because one never knows if the SW or the part is misbehaving.  We've
>>>>> had 15 boards built and I estimate 20% have BGA problems.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> BGA solder problems can also be related to the PCB design. Make sure
>>>> your pads have the recommended size and shape. Vias in the pad, or very
>>>> near the pad can also cause problems by wicking the solder away from
>>>> the pad and into the hole.
>>>>
>>> A specific piece of board design that will _kill_ you in BGAs is the 
>>> pad. You really _must_ use non Solder Mask Defined pads if you want 
>>> high yield (and I define that as 100% - I don't accept failures due 
>>> to failed soldering).
>>>
>>> I once worked at a company that had soldermask defined pads under a 
>>> pretty large (768 balls) BGA, and the yield was horrendous. Went to 
>>> non-SMD and the yields went way up.
>>>
>>> I would be very surprised, assuming the layout is properly done and 
>>> the CM has edxperience with BGAs, to _ever_ see a BGA failure rate 
>>> higher than that for TQFP for soldering issues. BGAs are _easier_ on 
>>> the process.
>>
>>
>>
>> Could you explain the difference?
> 
> 
> Not sure which difference, so I'll do both :)
> 
> A solder mask defined pad is simply that; the pad opening is defined by 
> solder mask, which covers some part of the pad. A non-SMD pad is defined 
> by the pad itself, and no solder mask is in contact with it.
> 
> On a solder mask defined pad, there is, by definition, a raised 'lip' 
> through which the solder ball has to flow for soldering, and this can 
> cause no connection at all to high impedance connections (which are 
> really frustrating - the JTAG tests will all pass, but nothing works at 
> speed).
> 
> There are some really good app notes on this; I'll see if I have any. I 
> seem to remember app notes from Xilinx, Altera, TI, Intel and Micron at 
> the least.
> 
> As to BGAs being easier on the process - on SnPb the solder ball wicking 
> effect actually 'pulls' the BGA onto the pads. The placement accuracy 
> for a BGA is 0.5 pad spacing because of this, which is less accurate 
> than final soldered placement accuracy and very easy to achieve with 
> automated pick'n'place.
> Whether this holds true in a non SnPb world has yet to be seen.
> 
> For a TQFP, because there are pads on each side, the wicking effect 
> doesn't work this way (the forces apparently cancel) so the placement 
> accuracy has to be the same as the final soldered placement accuracy. 
> For a highspeed part (to minimise impedance offsets) this may be as low 
> as 0.1 pad spacing, which may itself be 0.5mm. That's a very tight 
> tolerance, and very difficult to achieve.
> 
> Empirically, I have also had many more problems with cold joints, 
> soldersplash and solder shorts with TQFPs than with BGAs. I've 
> personally designed many dozens of boards with both and I have had very 
> few problems (except the non-SMD issue) with BGAs. Most of my solder 
> problems are with TQFP and QFN (and that's a process issue although the 
> process is harder to get right the first time for those).
> 
> Cheers
> 
> PeteS

FWIW, the board finish makes a lot of difference. My preferred finish 
for BGA boards is gold over nickel, and since RoHS that's the finish I 
use for all boards.

HASL is _not_ suitable for large BGAs (or any BGAs, imo), especially 
full grid types. Many BGAs have outer rows, and a central area where 
power and ground connect, and a nice clear zone between them. Full grid 
devices have no such clear zone (See the FT256 package from Xilinx for 
an example).

As an interesting note on solder mask defined pads, I worked with one 
device from IR (IPwr200x, ISTR) where a _solder mask defined pad_ was 
specified as a requirement. Although that was unusual, I did it that way 
and never had a failure. The ball spacing on that part was quite wide, 
however.

Cheers

PeteS

PeteS wrote:

[...]

>As to BGAs being easier on the process - on SnPb the solder ball wicking 
>effect actually 'pulls' the BGA onto the pads. The placement accuracy 

[...]

>For a TQFP, because there are pads on each side, the wicking effect 
>doesn't work this way (the forces apparently cancel) so the placement 
>accuracy has to be the same as the final soldered placement accuracy. 

which shouldn't be a problem, IMO.

And I can hand solder TQFP even at 0,5mm pitch without any problem.
And desolder it with little effort and without damaging the board.

Try this with BGAs.

Oliver
-- 
Oliver Betz, Muenchen (oliverbetz.de)

Oliver Betz wrote:
> PeteS wrote:
> 
> [...]
> 
> 
>>As to BGAs being easier on the process - on SnPb the solder ball wicking 
>>effect actually 'pulls' the BGA onto the pads. The placement accuracy 
> 
> 
> [...]
> 
> 
>>For a TQFP, because there are pads on each side, the wicking effect 
>>doesn't work this way (the forces apparently cancel) so the placement 
>>accuracy has to be the same as the final soldered placement accuracy. 
> 
> 
> which shouldn't be a problem, IMO.
> 
> And I can hand solder TQFP even at 0,5mm pitch without any problem.
> And desolder it with little effort and without damaging the board.
> 
> Try this with BGAs.
> 
> Oliver

In an automated process, things are completely different

Have you ever been in those situations?

Cheers

PeteS

PeteS <peter.smith8380@ntlworld.com> wrote:

[...]

>>>As to BGAs being easier on the process - on SnPb the solder ball wicking 
>>>effect actually 'pulls' the BGA onto the pads. The placement accuracy 
>> 
>> [...]
>> 
>>>For a TQFP, because there are pads on each side, the wicking effect 
>>>doesn't work this way (the forces apparently cancel) so the placement 
>>>accuracy has to be the same as the final soldered placement accuracy. 
>> 
>> which shouldn't be a problem, IMO.

>> And I can hand solder TQFP even at 0,5mm pitch without any problem.
>> And desolder it with little effort and without damaging the board.
>> 
>> Try this with BGAs.
>
>In an automated process, things are completely different
>
>Have you ever been in those situations?

certainly I referred to automated placement when I wrote "which
shouldn't be a problem, IMO" - we use an several years "old" Siemens
so I can't imagine that "current" equipment is worse.

But I love the possibility to hand solder prototypes or to replace a
microcontroller when I zapped it during ESD tests. All I need is a
matching funnel (wording?) and hot air to remove the IC and a
soldering iron with a hoof shaped tip and a magnifier to solder the
new one. Well, some solder wick could be necessary.

Oliver
-- 
Oliver Betz, Muenchen (oliverbetz.de)

Hello Oliver,


>>>>As to BGAs being easier on the process - on SnPb the solder ball wicking 
>>>>effect actually 'pulls' the BGA onto the pads. The placement accuracy 
>>>
>>>[...]
>>>
>>>
>>>>For a TQFP, because there are pads on each side, the wicking effect 
>>>>doesn't work this way (the forces apparently cancel) so the placement 
>>>>accuracy has to be the same as the final soldered placement accuracy. 
>>>
>>>which shouldn't be a problem, IMO.
> 
> 
>>>And I can hand solder TQFP even at 0,5mm pitch without any problem.
>>>And desolder it with little effort and without damaging the board.
>>>
>>>Try this with BGAs.
>>
>>In an automated process, things are completely different
>>
>>Have you ever been in those situations?
> 
> 
> certainly I referred to automated placement when I wrote "which
> shouldn't be a problem, IMO" - we use an several years "old" Siemens
> so I can't imagine that "current" equipment is worse.
> 
> But I love the possibility to hand solder prototypes or to replace a
> microcontroller when I zapped it during ESD tests. All I need is a
> matching funnel (wording?) and hot air to remove the IC and a
> soldering iron with a hoof shaped tip and a magnifier to solder the
> new one. Well, some solder wick could be necessary.
> 

It's usually called nozzle. As you said, you can buy square nozzles for 
stations such as Metcal that blow hot air only onto the pin rows. In our 
systems production it was never a problem to swap a flat pack, even 
large ones. The techs could do that in minutes. BGAs are a whole 'nother 
matter, a real pain to get off the board without damage.

-- 
Regards, Joerg

http://www.analogconsultants.com

Joerg wrote:

[desolder TQGP]

>> microcontroller when I zapped it during ESD tests. All I need is a
>> matching funnel (wording?) and hot air to remove the IC and a
>> soldering iron with a hoof shaped tip and a magnifier to solder the
>> new one. Well, some solder wick could be necessary.
>
>It's usually called nozzle. As you said, you can buy square nozzles for 

thanks, nozzle is the word I didn't remember. We have one with a
spring loaded vacuum cup, so the TQFP is lifted automatically as soon
as the solder is molten. Really easy to use and not very expensive.

Oliver
-- 
Oliver Betz, Muenchen (oliverbetz.de)