Duane over at Screaming Circuits had a great little post today that reminded of a problem I recently had (again): the need for thermal relief on small surface-mount parts, particularly discretes. I’ve been working away on a basic <1GHz wireless board as a weekend project, and the first version of the board worked well, but the Balun between the RF transceiver and the antenna failed to properly reflow every time. It would shift to the bottom in the image above, and because it’s a fairly fine-pitch part (it’s the six-pin chip in an 0805 sized package), the other side was never properly soldered and wouldn’t work. A bit of magnification and some time with the soldering iron fixed it, but it was an avoidable problem and you can’t do rework in a production environment.
The reason for the failure was because the other side was connected to a large chunk of copper, which was itself connected to a solid GND plane on layer two below (it’s a four layer board with solid GND and power planes, with signals on layer one and four), and one side would heat up much faster than the other pulling the part down. I’ve been bitten by this enough that I know to avoid it and add thermal relief, but because this is for RF I wanted the absolute lowest possible impedance to GND, and I thought it was worth trying with no thermal relief. Lesson learned. What did work for me? This little adjustment which kept a more balanced reflow pattern on both side since there was a more even distribution of heat during reflow, and the part stayed nicely in place: