Designing the proper thermal solution around your product is critical to success. I'm sure you've heard that more times than you can count, which in turn creates that internal eye roll we all have when we are fed reminders of what we consider common sense. Just like power, thermal design is often left to the end. Ironically most product reliability failures end up being the result of power or thermal issues, which most of us know yet there is still a tendency to de-emphasize it during development. While it might not be as flashy as the processor or radio selection, designing the power and thermal solution alongside the rest of the system is critical to getting the best performance out of it. I've become much more proactive in the thermal designs around Avnet boards, which is especially important as we have created more designs that are intended to be used as products rather than pure evaluation platforms. Boards that are created to be used as part of customer end systems need to be reliable for much longer periods of time and also function in a variety of environments beyond simple bench top evaluation.
I decided to post this blog because we have a great case example currently in production, the Ultra96-V2. The Ultra96 platform was originally thought of as more of an internal board that wasn't planned for market release. As the team worked with it though, it was realized that it could benefit a number of customers working with the Zynq UltraScale+. We then began the process of validating the board for market release. After releasing the version 1 solution (V1) we tweaked the design to add features such as a PMBus power architecture and changed circuits to align better with Avent's support capability. The original board had a custom bracket and fan designed to cool the Xilinx device. This configuration worked fine for most applications, but was limited in operational temperature range. We like to offer our platforms with industrial temp ranges whenever possible, so as I looked into how we could accomplish that I realized that the fan / bracket solution wasn't going to cut it. My new goal was to design a thermal solution that would work for both commercial and industrial temp ranges and, if at all possible, get rid of the fan.
Ah the fan. In most high power applications a necessary evil. From the first time I saw the platform though something about it didn't sit right with me. The Ultra96-V2 is by design a lower power platform. I felt that a fan running on it gave it the appearance of a higher power consuming design. My mission became clear, design a passive thermal solution that could enable industrial temp operation and eliminate the fan for low temperature operation.
To get there I needed the help of a partner experienced with thermal relief solutions. I reached out to Aavid Thermalloy, now a division of Boyd Corp. I had worked with Boyd in a previous life as a telecom hardware designer. I started remembering the days of running tests with a thermal chamber, a dozen thermal probes, a stack of gap pads and about half a dozen extrusions trying to find the perfect mix to get where I needed to be. I was confident we could get there, but I needed their help.
I was able to send them a mechanical sample of the existing board. From there we had a discussion about the power consumption of the UltraScale device as well as discussing other potentially power hungry devices that could be limiting our performance. From my testing I knew the power management ICs on the board were also getting warm. Not too warm to operate, but remember your power limits go down as the temperature goes up so if you can reduce the temperature its a good idea to jump at the opportunity. You'll see in the image below the red arrows point out the Zynq, two PMIC via stitched ground floods on the PCB and the single output regulator.
Boyd was able to prototype an extrusion that, along with gap pads, would mount to the board and contact the Xilinx device as well as the PCB below the PMICs. The PMICs have a via-stitched ground flood underneath the device. On our board the parts are on the opposite side from the XIlinx device, however we were still able to contact the PCB flood underneath the devices. We went through a couple of gap pad options before finding the right fit, but the end result was a completely passive (audible noise free) thermal solution that actually provides better relief than the bracket / fan assembly we started with. Note the different gap pad heights designed to contact the varying heights of the ICs and PCB. Better performance, no audible noise and lower cost, yes we hit the trifecta. Unfortunately we haven't figured out how to defy physics yet, so at temperatures approaching 85C you will still need some airflow. Whether that be from a fan mounted to the mounting holes provided on the new heatsink or from somewhere else in your system, the newly designed heatsink gives you the lowered thermal impedance from the die to the air that you need to make it work.
Keep an open mind when designing your thermal solutions. You may not be able to get there with off the shelf solutions, however custom solutions are available from companies like Boyd that can dramatically improve the operation of your product. Boyd also has thermal modeling capability so its definitely worth your time to see what they have to offer. Have the conversation and get a quote, the costs are very reasonable and the payoff can be substantial.