Thermal pad pressure on DM6467T

There are two important things you'll need to keep in mind when you do this.  First, we do have compressive loading requirements, and they relate to the lifetime you expect the device to last, and the temperature you expect the device to see.  We specify 28g per ball at 90C or 47g per ball at 70C for a ten year lifetime.  This is for a constant compression over the lifetime of the part, which is I think what you want based on the fact that you're looking at thermal pads.

While I'm not an expert in compressive package loading, from a system perspective It would seem to me that a big part of the problem for any compressive loading in real systems (on a PCB) is the board flexing under the pressure and balls getting torn off, especially after thermal cycling caused by regular use.  You mentioned standoffs, and this is good because it shows you understand that board flexing is an issue, and the best way to avoid it is to support the board as closely as possible to the pressure.  Increasing board thickness is another option that works well.

Second, while the DM6467 is designed for a heat sink, sounds like you're also trying to cool other chips that may not be.  One thing to know about chips that are not designed for heat sinks is that most of the heat flows through the balls of the chip rather than the top.  Chips not designed for a heat sink will dissipate ~75% of the heat through the balls, and 15% of the heat through the lid.  Therefore a heat sink on the top of the chip is not as effective as it would be for chips designed to run heat sinks. This is true for all manufacturers. 

Chips that are not designed for heat sinks lack an efficient path for the heat to flow from the die to the top of the package, so the heat flows through the substrate (that is like a tiny PCB that the silicon die is mounted on) and then through the balls of the part on to the PCB, primarily using the power and ground planes as heat spreaders to get the heat out of the part.

One method I typically recommend for these chips, that works great with thermal pads, is to put the thermal pad on the bottom of the board directly under the chip rather than on top.  The heat will flow out of the area better this way.  This is just something to keep in mind for cooling the other chips if they are not designed for a heat sink.

I hope this helps!

Keven

Few more comments from another engineer:

Max loading per BGA ball: 28g compression-load/ball @90C and 47g/ball @70C. The customer needs to translate this to total loading. They may assume 100% uniform loading (# of balls times strength) which is probably unrealistic. If sensitive, we recommend the customer do some type of stress concentration analysis.

Regards, Srirami.

Keven & Srirami,

Fantastic advice.  Thanks very much.

Do either of you have a spec for maximum board flex under a BGA?  And perhaps even a nominal multi-layer board flex coefficient?

Below is my response, and this may provide insight to others reading this thread in the future...

Note for DM6467T at 529 balls, 85C max op temp (use 28g 90C Srirami spec), that's 14.8Kg or 32.6 pounds.  At 0.75" square chip, that's 58 psi.  This is dramatically more capacity than I expected, much more than the 10 psi limit I was thinking of, and definitely high enough that board flexing is the limiting issue.  And in fact, I'm looking at Bergquist Gap-Pad 5000S35 which can be used effectively with under 2 psi, I believe, as well as their Gap-Filler liquid that should exert zero pressure (but adheres and so is not [easily] removable).

I have four holes right next to the DM6467T, and am looking to fasten (ref "chipscrew") through those, to keep the "bridging" distance and thus flexing to a minimum.  It's not as simple as just a spring loaded screw, because I'm integrating the heat sink with the enclosure wall, to which the standoffs also attach.  Thus I have standoff-vs-chipscrew length interaction, where assembly itself could flex the board.  Still working on this.

I've looked at pulling heat out of the board.  DARN!  Earlier today I realized I didn't defeat the thermal isolation from my standoff mounting holes to the internal ground planes.  That is, there IS thermal isolation there.  I need to short that across to get solid ground plane copper into my standoff mounting holes.  This will help draw heat into the standoffs and out the enclosure.

I've thought about back-side heat sinking.  Of course, it's a double-sided SMD board.  Thanks Keven for the advice about BGAs not intended for heat sinks sending some odd 75% of their heat into the board.  Meanwhile, a number of the leaded chips have copper heat pad connections under them, so they'll be putting a great deal of their heat into the board.  All the more reason to fix that standoff isolation as well as thermal pad to the back if I discover I need it.

Thanks,

Helmut

Sounds like you're headed in the right direction Helmut.  I agree with your conclusions, although be careful about ground loops when grounding to the case.  As long as no current flows through the case, you should be O.K., but you might want to have a back up option to disconnect the ground paths to the case in the event that you find ground loops interfering with an interface.  This could be plastic standoffs, etc., just keep them as a back up option.  I'm not an expert on case grounding, I've just seen a lot of ground loops in systems causing problems.

I've seen too many situations where a PCB works great until it's put into the system and then for some reason it doesn't.

Have a great day.

Keven

Thanks.

Note I've done a lot of data acquisition for race cars.  Talk about an EM noisy environment!  I've learned to always ground the board to the case in exactly one spot, consistent with preventing ground loops.  At the same time, I've also learned that sometimes you create ground loops on purpose, in order to have a smaller loop than you would be forced to have [externally] if you didn't do so (not applicable to today's situation).

Since I have thermal isolation between my ground planes and mounting holes, I won't get much heat out the standoffs right now, so plastic [nylon] on all but one standoff is workable.  If I need more heat conduction, I might be able to replace the nylon with aluminum, but put thin thermal conducting electrically isolating pads under them.

-Helmut