Termperature testing

June 7, 2013 in 8 Channel PDU by Joshua Oster-Morris

With the new access to temperature sensor data from the PDU, I embarked on a study of the PDU’s temperature characteristics. This is important to understand because the temperature of the silicon affects how the circuits behave. It is especially critical for the ADCs (analog to digital converters) which are used to measure current flow. Essentially all the bits and bobs placed on the PCB (printed circuit board) will behave differently based on their temperature though. Characterizing the temperature movements therefore will help inform the maximum currents we can work with and how the final enclosure will have to be designed to handle the expected temperatures.

IMG_20130318_151431_870[1]
The PDU was designed to keep the heat creating parts away from the logic circuitry. You may recall that the design is two PCBs stacked together but separated by about 17mm (.65″). The power FETs (field effect transistors) are on the bottom board and all the logic is on the top board. The FETs are what directly control the current flow and are where the majority of the heat is expected to be “created”. The FETs are sensitive to temperature as well but I selected rather expensive FETs that can each handle up to 100A continuous current and 1400A peak current and I am limiting them to 15A each (some 85A below their maximum continuous figure). In this configuration they are dropping a maximum of ~225mW of power @ 15A. If all the FETs were dropping this amount of power (not possible because I’ve set the total current to 100A and 8*15A = 120A) it still represents only 1.8W. You should all be relatively well familiar with that it feels like to hold something burning ~2W in your hand as that is about the same as the amount of power used to transmit voice over with a cell phone. It gets a little warm but that’s about it. There is another source of heat on the power PCB, the resistance of the copper on the board to the currents flowing. The current goes in on one side and flows across the board to the get to the farthest FETs. This is not necessarily a small amount of heat and must be managed. I solve this problem by using thick copper on the PCB as well as busbar (a big chuck of copper) to further reduce the resistance across the board. The thickness of the PCB is equal to 9 AWG wire and the busbar is equal to another 4 AWG wire. Combined they are just over 26mm² of copper conductor moving current across the board which is roughly equal to 3 AWG wire. The power conversion there is (0.06465mΩ/meter * 2″) * 100A² and equals ~320mW. That adds to the 1.8W above and still places us comfortably with cell phones in the 2W realm. By the numbers, things are looking fine but nothing replaces a good lab test so lets see how that went.
(0.06465mΩ/meter * 2″) * 100A²
I performed a “controlled” test so I could get a measure of the heat transferred from the power PCB to the MCU (micro controller unit) where the temp sensor resides and the heat generated on the logic board itself where the MCU is. The first test I performed was to run 30A (about the most I can do with my current lab power supply) through the board until we achieved homeostasis of temperature. It took about 30 minutes for the temp stop climbing but I let it sit for an hour to confirm it was done warming. It raised 10°C above room temp (the starting point). That figure is affected by the test conditions somewhat. A relatively significant issue is that the loads are within 10″ the PDU and they are radiating a lot of heat, ~375W. The power supply is also blowing all of its heat in the general direction of the PDU via it’s cooling fan. I used two sheets of copier paper folded over the PDU to protect it from directly radiating heat of the bulbs (I made shade :P) but that does little to ameliorate the heat moving around the PDU from the two cooling fans blowing heat around from the power supply and the giant .6Ω load and it’s cooling fan. Taking all that into account, the numbers are still quite encouraging. Of course, a test cannot be “controlled” without a control test so lets discuss that one. The control test was running the PDU with no load until we achieved homeostasis of temperature. I again let is sit running for an hour but turned all the outputs off. The power supply was still blowing it’s warm air but as it was doing less it would have been running cooler than the previous test, there was no radiant heat coming from the bulbs though as they were not on. In this test the it raised 5°C above room temp. To me, this means that the 30A from the previous test accounted for only 5°C of the 10°C shift. So, what is responsible for the other 5°C? Well, I suspect most of that is the switch mode power supply which is converting the 12V DC to 3.3V DC to run the electronics. All the logic electronics do consumer power too so they all contribute a little to that total as well. I logic board is burning about 330mW when on but not driving any loads and 375mW when driving the 30A so that by itself contributes about .66°C of the 5°C. The heat contribution of the power PCB while sourcing 30A therefore is responsible for 4.33°C. Assuming this is a linear function (a huge assumption that SHOULD NOT BE MADE) at 100A the power PCB will contribute 14.5°C to the system at 100A while the logic board will contribute about 6.5°C to the mix for a total of 21°C at 100A (with the poor assumption that the power board’s contribution is linear). These numbers are all relative to the board sitting on my bench in relatively still air and not encapsulated. The epoxy used for encapsulation can actually improve the thermal characteristics. The actual numbers at 100A will require further testing which is waiting on the new boards and a new power supply (a Jeep in this case).