Too heat in MT7621D

Hello

I developed a custom card but warms too much.
MT7621D
Build in RAM and Build in Giga switch lan. (MT7530)

Any advice to reduce it ?
Not fan or heatsynk.

Tried under clocking?

1 Like

Any liquefied gas will do. Care should be taken to not chose a flammable or otherwise harmful gas.

1 Like

Not, Thomas.
I mean software or project tricks.

How many degrees? Why do you think it's "too much"?

The fingers cannot stand.
Now can be 60/70°C
But when I colse PCB on the enclosure, i think it rise to 80/100°C
I believe it can melt the plastic of enclosure.

According to the datasheet the mt7621 chip maximum junction temperature (plastic package) is 125 degrees C. So, it's ok. Regarding the melted plastic... May need to adjust the case design.

What about the obvious choice, back to the drawing board and fixing the faulty design?

3 Likes

Under which conditions (CPU load, ambient temperature, board orientation horizontal/vertical, ...)?

Have you actually measured this, or are you just guessing?
What temperature do you want to achieve?

Ok, but I need tips for reduce heat generated from the CPU

I can try to recuce CPU clock or ram clock or booth
image

The measured heat is:
10minuts from 29 to 71°C
30minuts from 29 to 78°C
No data traffic
Horizontal heatsynk 14x14mm h4mm
PCB and enclosure cannot be changed.
The CPU-heatsynk will not can put it when the board is inside the enclosure. Or it would be very difficult to put it.


Heatsynk on photo is for 60W PoE
image

I've seen 2 or 3 mt7621 boards and the all had a chunky heat sink on the SoC. This does not look like a well thought out design. The 60 watt PoE looks suspicious too.
Are you running from PoE?
Underclocking might help a little but probably not enough. Also powering at under voltage might help a little but will probably need a big electrolytic as a current reservoir if you can get away with not making the SoC unstable.
You might be able to use a big but flat metal plate as a heat sink as well.
But all this, if any combination works, is all a bit of a bodge and unlikely to end up with something that is reliable.

3 Likes

I create 60W bt PoE, through 4 lipo cells, to power on an external device.
I can join the CPU with heatsink, but I still have to reduce the generated heat.

There are 5 voltage regulator.
3.3V, 1.5V, 1,2V, 1.1V, 1.0V
Which to reduce ?
I studied how to reduce the number of regulators, but I didn't find solutions.
Although, for my mistake, 1.2V was 1.5v and everything worked regularly.

Cannot add electrolytic capacitor, but 220uF ceramics yes.

And ram clock do not affect heat generated ? Only CPU clock ?

Maybe lower CPU clock require less voltage, then more stability.

This CPUs have power saving ?
Sleep when no traffic.

I think you need to revisit the design - as it stands it is not at all well thought out or implemented.

The 4 lipo cells will give ~16V. So at 60W, that is about 3 to 4 amps drawn from the cells. If you are converting to 48 volts for the PoE, that is potentially a lot of heat that could be generated - You probably need a bigger heat sink there too.

I am sorry to say this but the only reliable fix is to redesign it properly.

I have 20A at 3.5V and my PoE circuit is prefect.
Heat is reasonable and acceptable.
I cannot modify or remake box mold.

I just need advice on reducing the heat generated by the cpu

There is no way to reduce this significantly. You need a big heat sink like that found in commercially available designs.

If you are pushing 20A at 3.5V into a cat 5 or cat 6 ethernet cable, it will probably melt.

1 Like

PoE is usually at 48 volts, as described in the link you provided.

Could it be that there is somehow excess solder and bridged legs? Have you checked the soldering work with a microscope?

ulpian

If the system cooling and dimensions are fixed then you have to change something else - like using a much lower power SOC to begin with.
Something like a low power arm core will run cool in a confined space.