EC-5.6 Thermal Behaviour, and Why Everything Drifts

What this is and why it exists

Nothing in electronics holds still.

Resistance moves with temperature. So do gain, threshold voltage, oscillator frequency and amplifier offset. A design that ignores this works perfectly on a bench at twenty-five degrees and fails in a car, in a loft, or in February.

This topic collects the thermal behaviour later subjects each assume separately. Which way each thing drifts and why, where the noise floor comes from, and how heat actually leaves a device. It is short, because it is a habit rather than a body of theory. The habit is asking, of every design, what happens when this gets hot.

The vocabulary

• Temperature coefficient — how much a parameter changes per degree.
• Drift — slow change in a parameter, usually with temperature or age.
• Thermal noise — the voltage produced across any resistance by the random motion of warm carriers.
• Shot noise — the fluctuation from current being carried by discrete charges arriving at random moments.
• Noise floor — the level below which a signal cannot be distinguished.
• Thermal resistance — how many degrees of temperature rise each watt of heat produces.
• Junction temperature — the temperature inside the device, which is what its rating is really about.
• Thermal runaway — a loop where heating increases current, which increases heating.

The mental model

Start with the reading habit. A temperature coefficient tells you how much something moves per degree. On its own it means nothing. Multiply it by the actual temperature range the product will see, and then decide whether it matters. A part specified at one temperature is a part specified nowhere useful.

Which way things drift is worth knowing rather than looking up each time. A metal's resistance rises with temperature. A semiconductor's falls. So a resistor and a diode drift against each other, and some circuits are deliberately built to let one cancel the other. A base-emitter voltage falls by roughly two millivolts per degree, which is why a bias network that ignores temperature does not stay biased.

Now noise, and there are two kinds worth telling apart.

Thermal noise exists because carriers move randomly whenever the material is above absolute zero. That random motion appears as a fluctuating voltage across any resistance. It depends on temperature, resistance and bandwidth, and on nothing else — not on the current flowing. So the only ways to reduce it are to cool the circuit, lower the resistance, or narrow the bandwidth. That is the floor, and no amplifier design gets below it.

Shot noise is different. Current is carried by discrete charges crossing a barrier at random moments, so it fluctuates in proportion to the current itself. Reducing the current reduces it.

That difference has a practical consequence. If thermal noise dominates, reducing current does not help. If shot noise dominates, it does. Knowing which you are fighting decides whether an idea will work.

Heat flow is the easiest part, because it is a circuit you already know how to solve. Treat heat flow as current, temperature difference as voltage, and each interface as a resistance in series. Junction to case, case to sink, sink to air. Add them up, multiply by the power, add ambient, and you have the junction temperature. That temperature is what the rating is about, not the one you can measure with a finger.

Finally, the failure mode to recognise before building anything. Suppose warming a device increases the current through it, and that increased current warms it further. That is a loop with positive feedback, and it ends with the device destroyed. Thermal runaway is why a bipolar power stage needs emitter degeneration and why paralleling devices without ballast is a bad idea. Spotting the loop on paper costs nothing.

What you should now be able to explain or do

• Turn a temperature coefficient into a real worst-case figure.
• Say which way a metal and a semiconductor each drift, and name a circuit that exploits it.
• Distinguish thermal from shot noise, and say what reduces each.
• Compute a junction temperature from a chain of thermal resistances.
• Recognise a positive thermal loop before it destroys something.

Check yourself