AD9951/AD9859 temperature-depended stability issue

Hello everyone,

I'm working on a project containing the AD9951 as frequency generator for 145 MHz. Previously to that, the AD9859 was used, but was swapped for the pin-compatible AD9951 because of availability reasons. The issue described in the following post applied to both devices.

The DDS chip is used to generate an approx. 145 MHz carrier (HAM-radio band). When turned on, it works nicely and generates a clean signal on the output transformer when viewed on the spectrum analyzer.

This trace, however, suddenly becomes much broader and unusable when the chip and the board heats up. And by heating I do not mean hundreds of degrees, it is not at all uncomfortable to touch by hand.

It is important to state that the whole setup was not touched at all between good and bad signal. The behaviour can be reverted to the first picture by placing a small heatsink on the AD9951. If however, a hot soldering iron is placed approx. 1cm above the heatsink (without touching, just to raise the temperature by a couple of degrees), the issue re-emerges.

First I was suspecting some oscillation issues with the 145MHz power amplifier on the same board, but switching off the power and disconnecting the amplifier from the DDS does not change the behaviour at all. The issue seems to be coming from the DDS itself.

The used schematic and board layout are shown here:

The spectrum analyzer is connected directly to the output of the transformer with a 4n7 coupling capacitor and a 20dB attenuator (for safety). Attempts were made to amplitude-modulate the DDS via the DAC_RSET pin, but this feature is currently disabled, so the 3k9 resistor is just connected to ground.

Furthermore, the issue occured on two different PCBs, one with the AD9951 and one with the AD9859. The layout, however, is almost the same.

Now it is clear to me that it is probably hard to tell exactly what is happening on my board from a distance via a couple of pictures, but I just wanted to ask if this has been observed bevore in some situations or maybe I've left something obvious out. In case more information helps, I could also upload more design data, Settings, microcontroller code etc.

Thank you very much for your time in advance!

Best regards,

Josef

  • The EPAD on the underside of the device is not only a GND connection, but a heat sink, as well. That said, make sure the EPAD is adequately soldered to the GND plane, and that the GND plane is sufficiently large to draw heat away from the device.

    The two most common causes of heat-related signal quality issues:

    1. poor (or inadequate) solder connection to the EPAD
    2. a GND plane not substantial enough to transfer heat away from the device
  • Thank you very much for your quick reply!

    1. In terms of soldering: The part is hand-soldered on this prototype PCB, but the EPAD was tinned manually before assembly and the 9 vias in the board were flooded with tin. Then a broad tip was used to melt the tin in the vias from the underside and the chip could be gently pressed down, with everything flowing together.

    2. The board is standard 35um copper, only two layers for cost reasons. On one side, there are a couple of traces, but other than that the GND plane is not cut. Do you think it is too small as it is?

    I mean putting a small heatsink also increases this "threshold" a little bit, but I felt this was not really a reliable solution because the chip does not even get hot when this happens. Also, in the final design, it could easily get 50-60 degrees if located in the sun.

  • Another question: Is it likely that hand-soldering the part (no exact control over temperature gradient etc.) can damage the part and cause the described behaviour?

  • The package is intended for flow soldering (not hand soldering). So, hand soldering is potential issue (though not absolutely). The data sheet specifies the leads as having a 300C for 10s max. with regard to soldering.

    If the GND plane is "red" in PCB picture, then there is probably not enough copper area to adequately heat sink the part.

    Be aware it is very difficult (though not impossible) to get good solder flow through the vias and onto the EPAD using a soldering iron. Furthermore, your description of a nearby soldering iron creating the problem and a heatsink alleviating the problem points strongly in the direction of a thermal issue.

  • Hmm. I am almost sure that the solder connection is good because the EPAD as well as the mating pad on the PCB have been tinned before. It was therefore not possible to push the chip flat on the PCB. Then, the underside was heated until the tin would melt and only then it was possible to even push the chip flat on the PCB. Much flux was added, so I can hardly imagine it did not flow together.  I was more concerned to cause a short by excess tin. But anyhow, I will strip it off, examine and solder back on for comparison.

    The second thing that is very unclear is how it is even possible to exceed the devices temperature range. The maximum power dissipation is stated around 170mW, and the worst thermal resistance mentioned is 38K/W. This would only result in about 6.5°C difference from ambient. I measured the top of the case at 42 degrees and the PCB under the EPAD to 46°C. How can I possibly exceed the 150°C junction temperature?

    And in some sense, red and blue are both GND-Planes, but the main one (connected through the 9 vias) is the blue one.