How closely do you pack your thermal vias?
The above I would consider borderline excessive. That is, if the pads with holes, were directly under the thermal pad.
- Too many holes
- Holes too close to each other (or too big)
- Holes too far away
Holes should be 10 mil in diameter and spaced 25 mil apart. That leaves 15 mils of copper between vias . The diagonal holes have less copper than the diameter of the vias. It appears the vias may be too big.
The vias in contact with the thermal vias are the only really effective vias. Vias further than 3 mm from the thermal pad are ineffective. The copper not under the thermal pad is best used for convective and radiation heat transfer.
This is actually a horrible thermal design. The thermal vias should be on the thermal pad. The thermal area (the A in Fourier's Law) is the cross sectional area of the copper's thickness. PCB copper is too thin for sufficient conductive heat transfer. In this design, the heat transfer is first lateral then through the vias. The lateral heat transfer is unnecessary, inefficient, and ineffective.
Thermal Vias Points of Diminishing Returns Bottom Line
- 10-15 mil diameter holes (10 better than 15)
- hole centers spaced 25 mils apart
- Location of holes directly under the thermal pad, no further than 3 mm from the thermal pad.
it seems like more vias are better for thermal dissipation
The diminishing returns are significant so drilling holes to the point of causing wakened mechanical integrity is pointless.
So more holes is not better, more like useless.
There are two variables in the thermal dynamics conduction formula (Fourier's Law) where you can still improve.
- Area: thickness of the via copper
- dx Length/Distance: Length of via/thickness of PCB
If you have not done so, reduce the thickness of the PCB. By going from 0.062" to 0.031" you reduce via thermal resistance by 50%. I used 0.020" boards.
The Cree Optimizing PCB Thermal Performance documents referenced in a previous answer (@Wesley Lee) is very good. A few years ago I spent a lot of time researching thermal vias and found a lot of studies on the topic. The Cree document is a great summary of what I found elsewhere.
Also from the Cree document.
I think the number of holes is very relevant here. It shows the diminishing returns.
Much of the current research on thermal vias have been done for LED.
Even though they cannot spell platform, there is a good section on vias.
LUXEON Rebel Plaform Assembly and handling information
From the above document on number of holes and diminishing returns.
OSRAM has a nice little theraml management primer on Conduction, Convection, and Radiation.
OSRAM App Note: Thermal Management of Light Sources Based on SMD LEDs
Example from OSRAM
I do high power LED strips and the thermal vias did not help much. I concluded thermal vias were inadequate to conduct thermal flux.
What I had to do is mount the heatsink or a copper bar to the component side of the board as close to the heat source as possible. I never tried aluminum because I would leave bare copper where the bar connected to the PCB and was concerned about electrolysis. I now (got first PCB this week) use ENIG rather than bare copper so I can now try aluminum.
There is are alternating Cree XP and Lumiled Rebel footprints on this strip. The pads with the holes are the thermal pad. The screw holes for the heatsink are about 0.128" for the 4-40 machine screw.
This is how well the thermal vias worked:
Below is the heatsink mounted to the PCB. The heatsink here is a copper water pipe with cold water being pumped through it.
This approach was to have 100% thermal conduction where the thermal path was 100% copper.
I left the thermal vias in place to measure the temperature of the thermal pad.
The point is mounting the heatsink to the component side (if posible) is much better than thermal vias.
In testing the copper bar and the water pipe were almost the same temperature. The measured test points are below. Ice water water was pumped through the copper pipe.
The thermal pad, left most point, 10°C higher than the pipe right most. The point between the screw and thermal pad was a few degrees higher than the copper pipe.