Heatsink on FPGA?

Hi,

Anybody has experience with heatsinks on FPGAs?

In the V5 documentation, Xilinx says the heatsink can be glued to the
FPGA but that it
is safer to screw it to the board to avoid mechanical contraints to
the FPGA
ball when under vibrations.

But then the screws take some space on the board that we can't really
afford...
This would be at the expense of signal integrity (longer PCB
tracks...).

Any feedback welcome :o)

Many thanks,
Jean-Baptiste.

jean-baptiste.nouvel@jdsu.com wrote:

> Hi,
> 
> Anybody has experience with heatsinks on FPGAs?
> 
> In the V5 documentation, Xilinx says the heatsink can be glued to the
> FPGA but that it
> is safer to screw it to the board to avoid mechanical contraints to
> the FPGA
> ball when under vibrations.
> 
> But then the screws take some space on the board that we can't really
> afford...
> This would be at the expense of signal integrity (longer PCB
> tracks...).
> 
> Any feedback welcome :o)
> 
> Many thanks,
> Jean-Baptiste.
> 

I've been putting heatsinks on FPGAs for quite a while, including 
heatsinks with fans on them.  Gluing is fine for in the lab, but if it 
is going into the field, you'll want a better mechanical connection to 
the FPGA if the heat sink has any mass to it.  That can be done with 
screws to the board, or with some arrangement of springs on the chassis 
if your mechanical arrangement allows it. You could also fabricate a 
hold-down bracket that screws to the board, thereby only taking a little 
bit of board real-estate at two points (one we did had a bracket that 
went diagonally across the heatsink and held to the board with two screws).

Ray Andraka wrote:
> jean-baptiste.nouvel@jdsu.com wrote:
> 
>> Hi,
>>
>> Anybody has experience with heatsinks on FPGAs?
>>
>> In the V5 documentation, Xilinx says the heatsink can be glued to the
>> FPGA but that it
>> is safer to screw it to the board to avoid mechanical contraints to
>> the FPGA
>> ball when under vibrations.
>>
>> But then the screws take some space on the board that we can't really
>> afford...
>> This would be at the expense of signal integrity (longer PCB
>> tracks...).
>>
> I've been putting heatsinks on FPGAs for quite a while, including 
> heatsinks with fans on them.  Gluing is fine for in the lab, but if it 
> is going into the field, you'll want a better mechanical connection to 
> the FPGA if the heat sink has any mass to it.  That can be done with 
> screws to the board, or with some arrangement of springs on the chassis 
> if your mechanical arrangement allows it. You could also fabricate a 
> hold-down bracket that screws to the board, thereby only taking a little 
> bit of board real-estate at two points (one we did had a bracket that 
> went diagonally across the heatsink and held to the board with two screws).

Another small heatsing fastening method I have seen is a simple torsion bar hooked into a 
pair of through-hole eyelets placed on opposite sides. PCB real-estate for a pair of these 
is something like 3-4 square milimeters. No screws, easy to service, negligible PCB area 
and the eyelet tabs can be however long as necessary to clear nearby SMT components.

We have done before on Virtex-2. It's very good handle to remove the
top case of the FPGA if you have a desire to look inside so if you can
do have mechanical mounts to the pcb.

John Adair
Enterpoint Ltd.

On 12 Mar, 17:04, jean-baptiste.nou...@jdsu.com wrote:
> Hi,
>
> Anybody has experience with heatsinks on FPGAs?
>
> In the V5 documentation, Xilinx says the heatsink can be glued to the
> FPGA but that it
> is safer to screw it to the board to avoid mechanical contraints to
> the FPGA
> ball when under vibrations.
>
> But then the screws take some space on the board that we can't really
> afford...
> This would be at the expense of signal integrity (longer PCB
> tracks...).
>
> Any feedback welcome :o)
>
> Many thanks,
> Jean-Baptiste.

On 12 Mar 2007 10:04:30 -0700, jean-baptiste.nouvel@jdsu.com wrote:

>Hi,
>
>Anybody has experience with heatsinks on FPGAs?
>
>In the V5 documentation, Xilinx says the heatsink can be glued to the
>FPGA but that it
>is safer to screw it to the board to avoid mechanical contraints to
>the FPGA
>ball when under vibrations.
>
>But then the screws take some space on the board that we can't really
>afford...
>This would be at the expense of signal integrity (longer PCB
>tracks...).
>
>Any feedback welcome :o)
>
>Many thanks,
>Jean-Baptiste.

This is a non-trivial problem.  What is the environment?  Will it be
subject to shock and vibration?  What is the board orientation?

In our experience the only thing that works reliably is to screw it
down.  Our applications are ground-mobile military and passenger rail
vehicles.  

The forces have to be carefully controlled so that you don't deform
the BGA solder balls.  You will have to check with Xilinx to see what
compression forces are acceptable for each package.  To control this
we use spacers between the PCB and the heat sink, and a Bergquist
ultra-soft thermal gap pad between the heat sink and the BGA package.
You have to adjust the spacer height to get the right amount of
compression of the pad.

We have found that the PCB is not stiff enough and will bow under the
pressure.  So, on the bottom of the PCB we add a thin FR4 insulating
spacer and then a stainless-steel stiffener plate.  The plate has
countersunk holes, and we use flat-head screws from the bottom and
lock nuts at the top (heat sink).

We have been using Radian 'Icicle' series half-brick DC-DC converter
heat sinks.

We have not yet had to do this with Xilinx devices, but we have been
doing this with Freescale PowerPC processors (up to 480 pin BGAs).

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

Greg Neff
VP Engineering
*Microsym* Computers Inc.
greg@guesswhichwordgoeshere.com

-- 
Posted via a free Usenet account from http://www.teranews.com

Hello Greg,

thank you for your feedback.

We do have to consider vibrations. I think the standard says something
like [50Hz-500Hz]/3G.
But it is a telecom test equipment that will be horizontal and static.
So not so bad.

Reading you, screws is the right option. Makes sense.
PCB will have to be reinforced from underneath.

To control the force that is applied by the scew, Xilinx mentions some
spring based mechanism
that prevents the screw to deform the solder balls.

Thank you for taking the time to send me this.

Jean-Baptiste

Thanks Daniel!

Daniel S. a =E9crit :
> Ray Andraka wrote:
> > jean-baptiste.nouvel@jdsu.com wrote:
> >
> >> Hi,
> >>
> >> Anybody has experience with heatsinks on FPGAs?
> >>
> >> In the V5 documentation, Xilinx says the heatsink can be glued to the
> >> FPGA but that it
> >> is safer to screw it to the board to avoid mechanical contraints to
> >> the FPGA
> >> ball when under vibrations.
> >>
> >> But then the screws take some space on the board that we can't really
> >> afford...
> >> This would be at the expense of signal integrity (longer PCB
> >> tracks...).
> >>
> > I've been putting heatsinks on FPGAs for quite a while, including
> > heatsinks with fans on them.  Gluing is fine for in the lab, but if it
> > is going into the field, you'll want a better mechanical connection to
> > the FPGA if the heat sink has any mass to it.  That can be done with
> > screws to the board, or with some arrangement of springs on the chassis
> > if your mechanical arrangement allows it. You could also fabricate a
> > hold-down bracket that screws to the board, thereby only taking a little
> > bit of board real-estate at two points (one we did had a bracket that
> > went diagonally across the heatsink and held to the board with two scre=
ws).
>
> Another small heatsing fastening method I have seen is a simple torsion b=
ar hooked into a
> pair of through-hole eyelets placed on opposite sides. PCB real-estate fo=
r a pair of these
> is something like 3-4 square milimeters. No screws, easy to service, negl=
igible PCB area
> and the eyelet tabs can be however long as necessary to clear nearby SMT =
components.

jean-baptiste.nouvel@jdsu.com wrote:

> Reading you, screws is the right option. Makes sense.
> PCB will have to be reinforced from underneath.
> 
> To control the force that is applied by the scew, Xilinx mentions some
> spring based mechanism
> that prevents the screw to deform the solder balls.

Last time I looked at this, a few years ago, the best solution was 
something designed for the PC market. That market is big enough to 
support engineering design work on screws, springs, backplates, and so 
on. And the end product is often amazingly inexpensive, though that 
wasn't a consideration for my project.

You're welcome.

I personally like the torsion bar and I have seen it on a few PC motherboards. The bar 
regulates pressure and if you use a soft, sticky(-ish) thermal pad between the heatsink 
and FPGA, it will dampen vibrations while maintain contact and heatsink position. Of 
course, this method is suitable only for small thermal loads (<15W) due to low contact 
pressure and low-profile (10-20mm) lightweight (<100 grams) heatsinks unless you want to 
risk the heatsink ripping the eyelets off if the box is roughed up while either 
upside-down or standing up. The two points bar is only suitable for low vibration exposure 
applications.

For somewhat larger thermal loads and significantly improved vibration tolerance, instead 
of using a single bar, some designs I have seen used an H-shaped hold-down clip. With four 
anchor points, the heatsink is fully locked down on all axis and won't wobble anywhere 
near as much or easily as it would with the two-point torsion bar, significantly reducing 
the vibration stress on each anchor point and the likelyhood that the heatsink could fall 
off if someone (Fed-Ex employee?) kicked the box around repeatedly.

For applications where a device will be regularly exposed to vibrations, the four points 
H/X clips are noteworthy (the best I know) screw-less options.

jean-baptiste.nouvel@jdsu.com wrote:
> Thanks Daniel!
> 
> Daniel S. a &#4294967295;crit :
>> Ray Andraka wrote:
>>> jean-baptiste.nouvel@jdsu.com wrote:
>>>
>>>> Hi,
>>>>
>>>> Anybody has experience with heatsinks on FPGAs?
>>>>
>>>> In the V5 documentation, Xilinx says the heatsink can be glued to the
>>>> FPGA but that it
>>>> is safer to screw it to the board to avoid mechanical contraints to
>>>> the FPGA
>>>> ball when under vibrations.
>>>>
>>>> But then the screws take some space on the board that we can't really
>>>> afford...
>>>> This would be at the expense of signal integrity (longer PCB
>>>> tracks...).
>>>>
>> Another small heatsing fastening method I have seen is a simple torsion bar hooked into a
>> pair of through-hole eyelets placed on opposite sides. PCB real-estate for a pair of these
>> is something like 3-4 square milimeters. No screws, easy to service, negligible PCB area
>> and the eyelet tabs can be however long as necessary to clear nearby SMT components.
> 

On Mar 14, 6:54 am, Tim <t...@nooospam.roockyloogic.com> wrote:
> jean-baptiste.nou...@jdsu.com wrote:
> > Reading you, screws is the right option. Makes sense.
> > PCB will have to be reinforced from underneath.
>
> > To control the force that is applied by the scew, Xilinx mentions some
> > spring based mechanism
> > that prevents the screw to deform the solder balls.
>
> Last time I looked at this, a few years ago, the best solution was
> something designed for the PC market. That market is big enough to
> support engineering design work on screws, springs, backplates, and so
> on. And the end product is often amazingly inexpensive, though that
> wasn't a consideration for my project.


A note on glue.  We've been using glue to attach heatsinks for years
and
have thousands of boards in the field without detachment problems,
BUT...

1) We have found that the heatsinks that come self-sticking do not
hold.

2) We use only DeltaBond 156-K, which requires a long cure time.

3) We don't use fan-equipped heatsinks which may create their own
vibration
or supply boards for use in high-vibration environments.

That being said, if you're short on board space and your board is
going
into a typical P.C. or similar environment, glue can work for you.

Here's a picture of one of our high-volume boards:

http://www.alacron.com/FRAMEGRABBERS/FastImage1300new.jpg

4 of the 8 devices with heatsinks are Xilinx FPGA's.  The others are
DSP chips in a similar BGA package.

Regards,
Gabor