Two flat metal surfaces touch across about one per cent of their area. The rest is air, and that is what cooks your laptop

Press two precision-machined, mirror-polished metal blocks together as hard as you like. Under a microscope they are not touching. They meet at a scattering of high points, and the real contact area is on the order of one per cent of the area they appear to share. Everything in between is a thin, ragged layer of trapped air.

Air conducts heat about fifteen thousand times worse than copper. So the join you cannot see is, thermally, very nearly a wall — and in a laptop or a phone, that wall sits directly between the chip that makes the heat and the metal that is supposed to take it away.

This is the first surprise in thermal engineering: the bottleneck is almost never the material. It is the boundaries between materials. The second surprise is larger, and it runs against the instinct that most of nanotechnology trains into people — at the nanoscale, making something smaller usually makes it a worse conductor of heat, not a better one.

How a solid carries heat at all

There are two mechanisms, and which one dominates tells you a great deal about a material.

In metals, heat is carried mainly by the same free electrons that carry electric current. This is why the two properties track each other so closely — good electrical conductors are almost always good thermal conductors, a relationship formalised as the Wiedemann–Franz law. Copper is excellent at both for the same underlying reason.

In non-metals and semiconductors, there are no free electrons to speak of, and heat travels as vibrations of the atomic lattice itself. Those vibrations, quantised, are called phonons — sound waves in the crystal, carrying energy from hot regions to cold ones.

Phonon conduction can be spectacularly good. Diamond conducts heat better than any metal despite being an electrical insulator, and single-layer graphene is better still. The recipe is stiff bonds, light atoms and a highly ordered lattice — stiffness makes the vibrations fast, light atoms make them easy to excite, and order lets them travel a long way before something scatters them.

That last word is where the nanoscale enters.

Why small structures conduct heat badly

A phonon does not travel forever. It scatters — off other phonons, off impurities, off defects, off boundaries — and the average distance it covers between collisions is its mean free path.

In silicon at room temperature, a substantial share of the heat is carried by phonons whose mean free paths run from hundreds of nanometres up into micrometres. That number is the crux. It means that when a feature in a device is smaller than a few hundred nanometres, its own boundaries start scattering the phonons that would otherwise have carried heat across it.

The consequence is direct and measurable: a silicon nanowire conducts heat far worse than bulk silicon. Same element, same crystal structure, same purity — different size, and thermal conductivity falls by a large factor. A thin film conducts worse than a thick one. A nanocrystalline material conducts worse than a single crystal of the same substance, because every grain boundary is another scattering surface.

This deserves emphasis because it inverts the usual story. Elsewhere in nanotechnology, dividing a material finely buys you something: more surface for catalysis or adsorption, shorter diffusion paths in a battery electrode, new optical behaviour in a quantum dot. For heat conduction, dividing a material finely costs you, and there is no way around it — the boundaries you created are the scattering sites.

The joins, not the metals

Now back to the wall of air, because in a real device this is the dominant problem long before phonon physics gets a say.

There are two distinct interface effects, and they operate at different scales.

At the atomic level, even a perfect, gap-free interface between two dissimilar materials resists heat flow. The two solids have different atomic masses and bond stiffnesses, so their phonon spectra do not match, and a vibration arriving from one side is partly reflected rather than transmitted. This is thermal boundary resistance, and it is a real and unavoidable term in any layered structure — which matters enormously in modern chips, where a heat path may cross a dozen thin layers.

At the practical level, the problem is the one we began with: two nominally flat surfaces contact at a small fraction of their apparent area, and the gaps are filled with air.

This is what thermal interface material — the paste, pad or phase-change film between a chip and its cooler — is actually for. It is a widely held misconception that the paste is there because it conducts heat well. It does not. A typical thermal paste has a thermal conductivity of a few watts per metre-kelvin, against roughly four hundred for copper. Its entire job is to be much better than air, which is a low bar it clears by a factor of a hundred or more, and to fill the microscopic valleys so that the two solids are thermally continuous.

That reframing settles a long-running argument. More paste is not better; it is worse. Every micrometre of extra thickness is more of a mediocre conductor in the path. The ideal is the thinnest continuous layer that still fills every gap — which is why application technique matters more than the brand, and why a pump-out or dried-out paste after some years shows up as a machine that throttles under load when it never used to.

The material in the middle of a heat path is rarely what limits it. The boundaries are — and a boundary you did not design is a boundary filled with air.

What is actually inside a phone

Follow the heat out of a modern phone and the chain is instructive.

The processor die generates heat in an area of a few square millimetres, at a power density that would be alarming if you scaled it up. That heat must first spread sideways, because the surface it eventually escapes from is the whole back of the phone — hundreds of times larger. Spreading is therefore its own engineering problem, and it is solved with graphite sheets: layers of highly ordered graphite whose in-plane thermal conductivity exceeds copper's, at a fraction of the weight. Almost every phone contains them.

In higher-power devices this is upgraded to a vapour chamber: a sealed, flat, partly evacuated cavity with a wick and a small amount of working fluid. Fluid evaporates at the hot spot, the vapour moves to the cooler region essentially instantly, condenses there, and wicks back. Because phase change carries a great deal of energy per unit mass, a vapour chamber moves heat across a device with a far smaller temperature difference than solid conduction could — an effective conductivity many times that of any metal.

And at the end of the chain, always, is the slow step: getting heat from a surface into air. Convection to still air is poor, which is why fans exist, why laptops have fins with a great deal of surface area, and why a phone — which has no fan and cannot have one — is fundamentally limited by its outside surface and the air around it.

Throttling is what happens when this chain cannot keep up. The controller reduces clock speed and voltage to keep the silicon junction below its limit. It is a protective response, not a fault, and it means the cooling path — not the processor — is the component that ran out of capacity.

Graphene and carbon nanotubes appear constantly in thermal proposals for good reason, and the honest position on them is the same as the one for nanotube strength: the individual object is extraordinary and the assembly is not. A film of graphene flakes is not a sheet of graphene, and heat crossing from flake to flake meets exactly the interface resistance this article is about. Aligned, well-bonded assemblies do genuinely well; randomly dispersed flakes in a polymer usually disappoint relative to the headline number of the material they are made from.

The same physics, deliberately inverted

Here is the part that makes phonon scattering more than a nuisance.

A thermoelectric device converts a temperature difference directly into voltage, with no moving parts. Its quality is captured by a figure of merit, ZT, which rewards high electrical conductivity and a large Seebeck coefficient while punishing high thermal conductivity. The difficulty is that in an ordinary solid those properties are coupled: things that conduct electricity well usually conduct heat well too, so improving one term degrades another.

Nanostructuring is the way out, and it works precisely because of everything above. Grain boundaries, nanoscale inclusions and interfaces scatter phonons strongly — while scattering electrons much less, because electrons have shorter wavelengths and different scattering rules. So a nanostructured material can be made to block heat while still passing current: what the field calls a phonon glass, electron crystal. Nanostructured bismuth telluride and related materials improved ZT substantially on exactly this principle.

The honest caveat matters. Even the best ZT values around one to two translate to modest conversion efficiency, so thermoelectrics remain a technology for places where reliability and silence beat efficiency — waste heat recovery, spacecraft power supplies, small Peltier coolers — rather than a general answer to power generation. But the design principle is a clean example of a nanoscale effect being turned from a defect into the entire product.

The same inversion appears in insulation. Aerogels and nanoporous insulators work partly because their pores are smaller than the mean free path of air molecules, which at atmospheric pressure is around seventy nanometres. Confine air in pores smaller than that and gas molecules collide with the pore walls more often than with each other, which suppresses their ability to carry heat. Deliberately blocking transport by making the structure small — again.

Why it matters for students and researchers

Thermal management has quietly become the limiting factor in computing. Transistor density kept rising long after the power per unit area became the hard constraint, and a great deal of modern chip design — clock throttling, dark silicon, chiplet layouts, the packaging decisions that now matter as much as the transistors — is a response to heat rather than to any limit on making transistors smaller. This morning's piece on why AI systems consume so much power is the same problem seen from the electricity bill; this is what happens to that energy once it has been spent.

It is also an area where measurement is genuinely hard and worth learning. You cannot put a thermometer inside a 100 nm film. Techniques like the 3-omega method and time-domain thermoreflectance exist because nanoscale thermal properties have to be inferred from carefully designed transient experiments, and the same caution applies as in any indirect measurement — the number you get depends on the model you assumed.

The open problems are unglamorous and consequential. Thermal boundary resistance is poorly predicted from first principles and is often the dominant term in a layered device. High-conductivity materials that are also cheap, light and manufacturable in bulk remain scarce. Interface materials that do not pump out, dry or degrade over years are a real and underrated need. And phonon engineering — deliberately designing structures to steer or block heat the way photonic crystals steer light — is an active field where the fundamental science is still being written.

Frequently asked questions

Does applying more thermal paste improve cooling?

No — it makes things worse. Paste is a compromise material, far poorer than metal, and its only purpose is to displace air from the microscopic gaps between two surfaces. The best result comes from the thinnest layer that still covers completely. Excess paste adds thickness to the heat path and can also spill where it is not wanted.

Why does my phone get hot while charging as well as while gaming?

They are two separate heat sources. Gaming heats the processor and graphics core. Charging heats the battery and the power conversion electronics, through internal resistance and conversion losses. Doing both at once puts two independent sources into a device with one modest cooling path, which is the worst case — and, as covered in why batteries age, also the condition that damages the cell fastest.

Do laptop cooling pads actually work?

Modestly, and mostly for a specific reason: many laptops draw air from underneath, and a soft surface such as a bed or a lap blocks those intakes entirely. A pad that simply restores clear airflow can help substantially. A pad with fans, on a laptop already sitting on a hard desk, typically produces a few degrees at best, because the bottleneck is inside the machine rather than in the air around it.

Would a graphene heat sink fix phone cooling?

Graphite and graphene-based spreaders are already used in phones and they do help — but with spreading heat sideways, not with the final step of getting it into the air. That last step is limited by the phone's outside surface area and by convection, and no internal material can improve it. This is why a passively cooled device has a hard ceiling on sustained power regardless of what it is built from.

Why do large data centres use liquid cooling?

Because air runs out of capacity. Water carries far more heat per unit volume than air and transports it with a much smaller temperature difference, so as power density per rack rises, air cooling needs impractical volumes and speeds. Direct-to-chip liquid cooling and immersion cooling both address the same bottleneck this article is about — moving heat away from a small, very hot area quickly enough that the silicon never has to slow itself down.