We have had a pretty hot week in New York City, so when I was sitting on a friend's couch I couldn't help but notice a consistent source of heat radiating from the entertainment center that was about a foot away. Upon closer inspection, I noticed that his Motorola cable box was putting out a TON of heat – while it was OFF.
After I thought about the amount of waste heat this device was dumping into a non-air conditioned room 24/7, I started to wonder about the parts and how long they will survive if they are always warm enough to make a breeze. Is it just a few chips that are blazing hot? If so, those suckers have to be operating outside of suggested conditions.
Since the designer of the Motorola box made me so keenly aware of the poor thermal designs that plague some consumer products, I thought I'd write about how an engineer can determine when kinda hot is too hot. Who knows – maybe it'll save some part someday from getting slowly fried!
For this article's example, if you'd like to follow along I'll be using a datasheet for an LM317, a standard 3-terminal linear voltage regulator.
Here's what you'll need to determine thermal design quality:
Thermal resistance of the part given the heat sinking design.
There are two parameters that need to be looked up: (1) the thermal resistance from the datasheet and (2) the conditions under which that number is determined. With our LM317, we find theta JA, the thermal resistance from the die junction to the ambient temperature, to be 70-245°C/W depending on the package selected. Through-hole components make the conditions for this spec straight forward since they are expected to stand up off the board, but there is a third consideration for SMT parts: heat sinking to the PCB. This is where the datasheets can be a little... optimistic. With roughly ½ of the part's surface area soldered to the board, this can either help or hurt the thermal conductivity. In the case of the soldering the large thermal pin to a tiny pad, the board will act as an insulator. However if the PCB designer uses a large hunk of copper for the thermal pad, heat is quickly wicked away from the part giving better performance. Sometimes this information is hidden away in an app note, but in our case the plots can be found on page 9 of the datasheet. Thankfully ON semi specs their parts without the use of a huge copper pad, however one must be wary of 'front-page' thermal specs without considering the heatsinking conditions under which they are valid. For our LM317 example, we will use a figure of 70°C/W.Ambient temperature in the area surrounding the chip.
This is not the ambient temperature that the human operator will be in, but the temperature of the ambient air around the part itself. Will the unit be placed in an enclosed cabinet that doesn't allow air flow? How about the enclosure of the product itself? Will the inside of the case get much warmer than the air of the room? The part's ambient temperature acts as the lowest temperature a part can be, so any increase in temperature from the part's power dissipation will add to this minimum. For our LM317 example, we will assume it will be in an ambient temperature of 40°C.Power dissipated in the part
This is my favorite point to consider because it is pure circuit analysis and one can actually calculate a quantitative number. The key here is to be sure to find the worst-case power consumption of the part in question. In the simple case of our LM317 linear regulator, let's say it's dropping a 5V rail to create a 3.3V rail and will need to supply 500mA of current at most. Therefore we can calculate full-on power dissipation as:
P = (Voltage In - Voltage Out) * (I) or (5V – 3.3V) * 0.5A = 0.85 Watts
Of course not all parts are that straightforward, and some switchers may require the use of SPICE to determine the dissipated power (LTspice is my favorite for this kind of simulation). But in the end, a figure in Watts is what you want to have.
Max die temperature of the part
This can also be found in the datasheet, but often times engineers use this figure to insert a margin of safety. For instance, some parts have a max die temperature of 150°C, but many experienced engineers will tell you that a part that has a die temperature of over 100°C deserves at least a second look. Keeping the part at a lower temperature than is needed not only gives protection against design errors, but it also ensures that the users who will inevitably operate the unit outside suggested operating conditions will not be disappointed. In our LM317 case, we won't want it to rise above 100°C. Somewhere out a person is enjoying the extra caution I took as she sits in a sauna with my design running like a champ. I just know it.
Now that we have all that, follow this equation:
Die Temp Max >= (Ambient Temperature) + (Theta JA) * (Power Disipation)
So in our 317 example, we find:
(40°c) + (70°C/W) * (0.85W) = 99.5°C
Since the assumed worst-case conditions will result in a temperature rise of 99.5°C inside the part, we can safely say the design constraints have been met with 50°C remaining for safety margin. Success!
I know it can be troublesome to complete a calculation like this for every part that could be hot, but by the time proto boards come back from manufacturing, you'll be happy that you haven't created a glorified hot plate that needs redesign!