Thermal Design & Hot Spots
Heat Is a Hard Wall
A chip turns electricity into work and, as a side effect, into heat. That heat must escape, or the chip cooks. Once a small area gets too hot, the chip slows, errors creep in, or it shuts down to save itself. So thermal design is not optional. It sets a real ceiling on how fast and how dense a chip can run. This chapter explains heat in plain words: power density, junction temperature, thermal resistance, hot spots, and the dangerous loop between heat and leakage. It also covers how chips sense and manage their own temperature.

Power Density, Not Just Power
Total power matters, but power density matters more. Power density is power divided by area: how much heat is packed into each square millimetre. A chip can burn a lot of power and stay cool if it is spread out. The same power in a tiny area makes a hot spot. As transistors shrank, more of them packed into each area while each one's power fell only slowly. So power density climbed. This is why modern chips fight heat far harder than older, larger ones did.
| Quantity | Meaning | Why it matters |
|---|---|---|
| Total power | all heat made | sets cooling system size |
| Power density | heat per area | sets local temperature |
| Hot spot | small high-density area | the actual failure point |
Worked example — power density
A block burns 4 W in an area of 2 mm².
power density = 4 W / 2 mm² = 2 W/mm²
Two watts in a square millimetre is intense. Spread that same 4 W over 20 mm² and the density drops to 0.2 W/mm² — ten times easier to cool. Placement that spreads heat is a real design lever.
Junction Temperature and Thermal Resistance
The hottest point inside a transistor region is the junction. Its temperature is the junction temperature. Chips have a maximum allowed junction temperature; cross it and reliability and speed suffer. Staying under it is the goal of all thermal design. Heat flows out through a chain: from the junction, through the package, to a heat sink, to the air. Each step resists heat flow, like a resistor resists current. This is thermal resistance: how many degrees the temperature rises per watt of heat pushed through. Add up the chain and you can predict the junction temperature.

Worked example — junction temperature
A chip burns 30 W. The total thermal resistance from junction to air is 1.5 °C per watt. The air is 40 °C.
temperature rise = power × thermal resistance = 30 × 1.5 = 45 °C
junction temperature = 40 + 45 = 85 °C
If the limit is 100 °C, this design has 15 °C of margin. Burn more power or cool worse, and that margin vanishes.
Hot Spots
Heat is never uniform. A busy block — a processor core, a fast multiplier — runs much hotter than idle memory beside it. These local peaks are hot spots. The chip can fail at a hot spot even while its average temperature looks fine. Hot spots drive design choices. Busy blocks are spread out or placed near good heat paths. Activity is balanced so no single area runs flat-out for long. Temperature sensors are scattered across the die to catch peaks where they happen.
| Cause of hot spot | Mitigation |
|---|---|
| Dense, busy block | spread it out, place near sink |
| Sustained full activity | balance load, throttle |
| Poor local heat path | thermal vias, better placement |
The Heat–Leakage Loop
Here is the dangerous part. Leakage current rises sharply with temperature. More leakage means more power. More power means more heat. More heat means more leakage. This feedback loop is called thermal runaway risk. In most chips the loop is stable: cooling carries away the extra heat before it spirals. But in a poorly cooled chip, or one with very high leakage, the loop can run away — temperature and leakage chase each other up until the chip fails. Managing leakage is therefore also a thermal job.

Worked example — leakage climbs with heat
Suppose leakage roughly doubles for every 20 °C rise. A block leaks 2 mA at 40 °C. At 100 °C:
rise = 100 - 40 = 60 °C = 3 steps of 20 °C
leakage = 2 mA × 2³ = 2 × 8 = 16 mA
An eight-fold leakage jump just from heat. This extra power feeds back as more heat, which is why hot chips get hotter than power-at-cold numbers predict.
Sensing and Managing Heat
Modern chips watch their own temperature. Small sensors across the die report local temperatures. A controller reads them and acts before damage occurs. The main lever is throttling: slow the clock or drop the voltage when a sensor reads too hot. Power scales with voltage squared and with frequency, so a small throttle sheds a lot of heat fast. Other moves include moving work to a cooler block, gating idle parts, and, in extreme cases, an emergency shutdown.
| Response | Speed | Effect |
|---|---|---|
| Lower frequency | fast | less dynamic power |
| Lower voltage | fast | big power cut (V²) |
| Migrate work | medium | spreads heat |
| Emergency shutdown | instant | last resort |
Worked example — throttle saves heat
A core at 1.0 V / 3.0 GHz is too hot. Throttle to 0.85 V / 2.4 GHz.
voltage factor = (0.85/1.0)² = 0.72
frequency factor = 2.4/3.0 = 0.80
dynamic power ratio ≈ 0.72 × 0.80 ≈ 0.58
Dynamic power drops to about 58% — a big, fast cut in heat for a modest speed loss. That is why throttling is the front-line thermal defence.
Interview Q&A
system you need, but power density — power per unit area — sets the local temperature. The same power spread out stays cool, while packed into a tiny area it makes a hot spot, and hot spots are where chips actually fail.
point inside the transistor region, and it must stay under a set limit. You estimate it by multiplying the dissipated power by the total junction-to-air thermal resistance to get the temperature rise, then adding the ambient air temperature — e.g. 30 W × 1.5 °C/W + 40 °C = 85 °C.
pushed through a path, much like electrical resistance for current. Heat flows from junction through package to heat sink to air, each stage adding resistance; summing the chain predicts the junction temperature.
area that runs much hotter than the chip's average. It is dangerous because the chip can fail there even while the overall average temperature looks safe, so design must spread busy blocks and place sensors to catch local peaks.
more heat causes more leakage, which adds more power, which makes more heat. Usually cooling
keeps this loop stable, but with poor cooling or very high leakage it can run away — temperature and leakage chase each other up until the chip fails.
controller that throttles the clock or voltage when too hot. Because power scales with voltage squared and frequency, a small throttle sheds a lot of heat quickly; the chip may also migrate work, gate idle blocks, or, as a last resort, shut down.
Key Takeaways
- Power density (heat per area), not just total power, sets local temperature and creates hot spots.
- Junction temperature = ambient + power × thermal resistance; it must stay under the chip's limit.
- Hot spots can fail a chip even when the average temperature looks safe.
- The heat–leakage loop can cause thermal runaway if cooling cannot keep up.
- Chips sense and throttle — cutting voltage and frequency sheds heat fast because power scales with V² and frequency.
ChipBuddy
← Home
Comments
Leave a Reply