Heat Dissipation
Heat dissipation is the removal of waste heat from an electronic device so that its temperature stays within safe limits. Essentially all the electrical power a chip consumes ends up as heat, because the electrons do work only in the sense of moving charge through resistance and charging capacitances, and none of that energy is stored or exported in useful form. The engineering problem is therefore not whether heat appears but how fast it can be carried from a patch of silicon a few hundred square millimeters in area out to the surrounding air or water.
Joule heating is the root cause, and it scales with current squared. When current I flows through a resistance R, the power turned into heat is P = I^2 × R, equivalently V × I. In a transistor the dissipation shows up in the channel while it conducts, in the wires that connect millions of devices, and as charge is dumped from tiny capacitors on every switching event. A processor drawing 100 watts is, thermally, a small electric stove element, with the difference that the heat is released in a pocket of silicon smaller than a postage stamp. Vacuum tubes shed their heat as glowing glass; transistors shed far less per device, but they are packed in vastly greater numbers.
Junction temperature is the number that matters, and thermal resistance links it to power. The hottest region of a semiconductor is the junction, and reliability rules cap it, typically around 125 to 150 degrees Celsius for silicon parts, with consumer chips throttling lower. The relationship is a heat-flow version of Ohm's law: junction temperature = ambient temperature + power × thermal resistance, where thermal resistance is measured in degrees Celsius per watt and is the sum of the resistances along the path from die to package, to heat spreader, to heat sink, to air. Lowering any link in that chain lowers the temperature rise. Silicon itself is a fairly good thermal conductor, but the interfaces between layers, including thermal paste, are often the bottleneck. Heat also speeds up failure mechanisms such as electromigration, so a part run hotter lives shorter, a rule of thumb that is the thermal side of reliability.
Cooling hardware climbs a ladder from conduction to convection to phase change. A heat sink is a block of aluminum or copper with fins that increases the area in contact with air; a fan forces convection and lowers the thermal resistance of that last step by a large factor; a heat pipe uses an evaporating and condensing fluid to carry heat sideways with very little temperature drop; and liquid cooling, using cold plates and pumped water or coolant, moves much more heat per unit volume because liquids carry far more heat than air. Laptops and phones balance this with a thin chassis and thermal throttling, while data-center GPUs and AI accelerators, with power ratings in the hundreds of watts per chip (some current parts are rated around 700 watts or more), increasingly use direct liquid cooling, and some servers immerse boards in dielectric fluid.
Power density, not total power, is what limits clock speed. A chip's watts per square centimeter measure how concentrated the heat is. Typical high-end processors are generally cited in the tens of watts per square centimeter on average, and local hot spots, such as an integer unit running at full tilt, are far higher, in comparisons that put a modern processor's heat flux well above a kitchen hot plate. Because dynamic power rises with frequency (and with voltage squared), pushing clock speed up raises power density until the cooling solution cannot keep up. This is a core reason mainstream clocks stalled near a few gigahertz after the mid-2000s, and why chips now use many cores at moderate speeds, short boost bursts, and dark silicon, regions deliberately left off at any moment.
Thermal runaway is the classic failure of bipolar transistors, and it is a positive-feedback loop. In a BJT, the base-emitter voltage needed for a given current falls by about two millivolts per degree Celsius, so if the junction heats up and the bias is fixed, the collector current rises, which adds heat, which raises current again. Without a stabilizing resistor, feedback, or a current-limited drive, the device can destroy itself; early germanium transistors, with their larger leakage currents that double roughly every ten degrees, were notoriously vulnerable. Designers answer with emitter resistors, balanced biasing, and thermal protection that shuts a circuit down. Modern power electronics add careful current sharing among parallel devices for the same reason.
A frequent misconception is that heat is a defect to be eliminated; it is a conserved consequence of computation that must be managed. Better cooling does not reduce the energy a computer uses, it only moves the energy away, and a hot data hall can spend a large fraction of its power on cooling. The more durable fix is to make each operation cost less energy, which is where low power design enters.
Read heat dissipation as the physical bill that arrives for every bit flipped: the part of an electronic device's work that cannot be avoided, and the ultimate ceiling on how densely and quickly a chip can be made to think.