When I started my career, I was working with the 4000 series CMOS devices. Unless they were running at very high speeds, they drew micro-amps, so any voltage regulation was handled by a simple linear regulator with no thought for heat sinking. Back in 1976, I was unaware of switch-mode power supplies, if they had been developed, and the linear regulator was all the rage. I seem to remember the LM317 being selected of the product of the year by Electronics magazine. (Anyone remember that?) I was lulled into a false sense of security.
A couple of years later, I moved on to NMOS micros like the 8085 and 8748. These drew tens of milliamps (or more), but 100 mA isn't much, is it? And yet the voltage regulators I was using were overheating. How could that be? Well, of course, I was using a high input voltage, and 10 V across the regulator at 100 mA is 1 W — well outside the absolute maximum of 750 mW for a TO92, and in reality pretty stressful for a TO-220 package without heat sinking.
Over the years, I have developed a rule of thumb: If I place my thumb on a component, and I can't keep my finger there, it is too hot. My thumb seems to be calibrated to about 55°C. Now, I know that cost is the imperative for some engineers, and they need to design right up to the limit, taking into account the maximum ambient temperature combined with the temperature gradient across the heat sink and package arriving at the maximum junction temperature. I work in a different world. We produce very small volumes, and the prototype is the final unit. It has to work the first time, so I give myself plenty of margin for error.
Most of my designs are still simple, with few ICs (one is often a microcontroller) in a limited space, so I still go for linear regulator. The newer complete switch-mode regulators (like the TI LMZ14201) can be very helpful, if a little pricy. But a mentor of mine once said, “When it comes to power dissipation, there is no magic. Bigger is better.” The micro may be speced at a particular current, but this current is not the amount drawn when an emulator is plugged in or there is some malfunction on the circuit. After several instances of overheating, I started to upsize. For instance, I use a DDPAK, rather than a SOT23, for a regulator package.
Often I have designed two linear regulators in series or a series zener before the regulator in order to distribute the heat and reduce the individual component power dissipation. Also, if I am driving LEDs or relays, I will power them directly from the incoming power (and switch them through open collector transistors or something similar) to save the additional current through the regulator. Even a relatively small temperature rise can affect your measurements in a case like a current shunt, and I have used two or more in parallel to limit this effect, though this introduces a conundrum if you are looking for a Thevenin connection.
Wait. There's more that I would like to point out. Make sure you read your components specification completely. Let's look at the PR02 resistor. This is supposed to be a 2W resistor, but look at page 10 of the data sheet. The hotspot temperature at the 1.6W power level leads to a temperature rise of around 120°C. The resistor itself may handle that temperature, but it will heat up everything around it. I have seen PCBs suffering from heat damage as a result of hot resistors, and it may even get hot enough to melt the solder connecting it to the board. The operating temperature will certainly fail my thumb test.
You may think that this resistor is unique. It isn't. Take a look at the Heat Rise Chart on this data sheet, which indicates that a 5W resistor (from this particular manufacturer) will go up more than 100°C at only 3 W.
Something else to note from the PR02 data sheet (on the same page) is that the length of the leads and the height above the PCB affect the power dissipation, as well. This is not restricted to resistors. Here is the data for an MR751 diode. Pay attention to Figure 3.
Another thing you should be aware of is derating. When the ambient temperature increases, the part's ability to dissipate its own heat drops off to zero. All the data sheets I've cited so far include a derating curve.
Now, I don't design switch-mode power supplies (SMPS), but I know enough that heat dissipation is a problem there, as well. In particular, I am told you need to pay attention to switching the MOSFETs with a sharp edge, or the transistor will be on for a period of time, and the conduction of current in the linear zone will generate heat. The coils, of course, must also be correctly rated — a topic that is way beyond my knowledge frontier.
You would think that capacitors wouldn't get hot, since they carry only reactive power. That's the theory, anyway. But capacitors, like everything (and everyone) else, are imperfect. There is a property of capacitors called equivalent series resistance (ESR), and the heating by any AC current is the same I2R as any resistor, so the capacitor can get warm, as well. This phenomenon is not restricted to a SMPS. Where safety is not an issue, there is a technique where a mains AC input is rectified and regulated to a low DC voltage with a device like the Supertex/Microchip SR10 regulator. A large portion of the voltage is dropped across a capacitor, and the ESR must factor into that.
Heat dissipation concerns are not restricted to discrete components. Integrated circuits will generate heat through the sheer number of transistors that are switching and through the current being sunk or sourced at their I/O pins. In today's world of miniaturization and surface mount technology, power dissipation has become even more of an issue. The use of copper thermal pads on the underside of the IC package is becoming very common, necessitating the development of heat spreading techniques on the PCB, including multiple vias for large surface areas on the top and the bottom of the board.
My advice to you: Whenever you introduce a component into a design, consider its power dissipation first. Unfortunately, it seems to be a case of “Do as I say, not as I do.” Even though 38 years of experience should have taught me that, it always comes back to bite me. Only last week, I made an error with 200 V across a high-voltage linear regulator, the LR8. The current through the voltage biasing resistors was a measly 4 mA with no other load, and 800 mW is just too much for a TO92.
Do you have burn callouses on your fingertips, as well?