Thermal Paste and Interface Materials: What Is Actually in the Tube

June 10, 2026 18 min read Chemistry & Materials

hardwarecoolingthermal-managementchemistryhomelab

Contents

Every cooler review benchmarks thermal paste as if the difference between brands were the deciding variable, but almost none of them explain what the material is actually doing or why it works. The thermal interface material (TIM) sitting between a CPU die or integrated heat spreader and a cooler’s contact plate is doing something surprisingly subtle: it is not conducting heat better than copper, it is eliminating the air that would otherwise fill the microscopic voids between two surfaces that look flat but are not. Understanding that distinction — and understanding the chemistry that makes different materials better or worse at filling those voids — changes how you think about every thermal compound decision from a desktop repaste to a server maintenance schedule.

This post covers the physics of surface contact, the chemistry of the major TIM categories, the real failure modes that matter in production hardware, and the honest answer to the application method debate. It is aimed at people who want to understand what they are doing rather than repeat forum wisdom.

The Problem: Surfaces That Look Flat Are Not

When a machined aluminum or copper heatsink base contacts a CPU integrated heat spreader (IHS), the contact looks solid to the naked eye. Under a profilometer it looks nothing like that. Even a lapped and polished copper surface — the best practical result from manual finishing — has peaks and valleys measured in micrometers. A factory-ground heatsink base has surface roughness (Ra, the arithmetic mean deviation) typically in the 0.4–1.6 µm range. An IHS surface is somewhat better, around 0.1–0.4 µm Ra on most Intel and AMD parts, because it goes through a CMP (chemical mechanical planarization) step during manufacturing. But “better” is relative: even a 0.1 µm Ra surface, when pressed against a 0.8 µm Ra heatsink base, creates genuine gaps.

What fills those gaps when there is no TIM is air — specifically, nitrogen and oxygen at atmospheric pressure, with a thermal conductivity of approximately 0.025 W/mK. That number is the core of the entire problem. Copper conducts heat at 385 W/mK. Aluminum conducts at 205 W/mK. Air conducts heat roughly fifteen thousand times worse than copper. The amount of air in a metal-on-metal dry contact is surprisingly large: typical interfacial contact area, measured carefully, is 1–5% of the apparent geometric contact area. The rest is air voids. The effective thermal resistance across that interface is not determined by the metal at all — it is determined almost entirely by the trapped air.

A good thermal paste fills those voids with a material conducting at 4–14 W/mK depending on formulation — roughly 160 to 560 times better than air, and nowhere near copper, but that framing is wrong anyway. The paste’s conductivity matters far less than its ability to wet the asperities and displace every air pocket. You are replacing 0.025 W/mK air with 4–14 W/mK paste. That is a massive win even if the paste never approaches copper’s 385 W/mK.

The Layer Stack

To reason correctly about TIMs, it helps to be precise about where they appear. A typical desktop CPU has two interfaces, not one:

╔══════════════════════════════════════════╗
║         CPU Cooler (copper/aluminum)     ║
╠══════════════════════════════════════════╣
║  TIM-2: user-applied paste / pad         ║  ← the repaste people talk about
╠══════════════════════════════════════════╣
║  IHS (Integrated Heat Spreader, copper)  ║
╠══════════════════════════════════════════╣
║  TIM-1: factory-applied solder or paste  ║  ← delidding affects this layer
╠══════════════════════════════════════════╣
║  CPU Die (silicon)                       ║
╠══════════════════════════════════════════╣
║  Package substrate (organic laminate)    ║
╚══════════════════════════════════════════╝

TIM-1 is what the CPU manufacturer installs between the die and the IHS. On most Intel consumer CPUs until the 13th/14th generation, this was a mediocre grey polymer paste — a significant and well-documented source of thermal resistance. AMD Ryzen CPUs use solder at TIM-1, which is why Ryzen chips typically show smaller delidding gains. Intel’s newer generations (13th/14th gen Core, Core Ultra) also moved to solder or indium-based materials for TIM-1 on their flagship SKUs. Server and workstation CPUs — Intel Xeon, AMD EPYC — have always used solder or indium at TIM-1, because the cost of a bad thermal design at that price point is unacceptable.

TIM-2 is the layer you apply when you install a cooler. This is the tube of compound on your desk. For laptops and some small form factor systems, TIM-2 may be a pre-applied thermal pad from the factory rather than paste. The physics are the same; the material properties differ substantially.

What Is Actually in the Tube: Filler Chemistry

Almost all commercial thermal compounds follow a similar structural logic: a carrier (which determines viscosity, spreadability, and long-term stability) loaded with filler particles (which carry the thermal conductivity). The carrier and filler together determine everything: how well the compound flows into gaps, how it ages, and what happens if it gets somewhere it should not.

Silicone and Oil-Based Carriers with Ceramic Filler

The majority of consumer compounds use a polydimethylsiloxane (PDMS) silicone base or a synthetic hydrocarbon oil. Into this carrier, manufacturers disperse ceramic particles — most commonly zinc oxide (ZnO, conductivity ~25 W/mK in bulk), aluminum oxide (Al₂O₃, ~30 W/mK), or aluminum nitride (AlN, ~200 W/mK in bulk). The loading fraction and particle size distribution determine the compound’s macroscopic thermal conductivity: more filler and larger particles raise conductivity but also raise viscosity, making the compound harder to spread and more prone to trapping air bubbles in the bond line.

Zinc oxide compounds — the classic Arctic Silver 5 uses an Al₂O₃/ZnO blend as secondary filler — have been around for decades and are well understood. They are electrically nonconductive, making them safe to apply without obsessive precision. They age slowly: the silicone carrier does not evaporate quickly, and the ceramic particles do not degrade. Dry-out (covered below) is the primary aging mechanism.

Aluminum nitride compounds like Thermal Grizzly Kryonaut and Noctua NT-H2 achieve higher conductivities (5–8 W/mK measured) partly through better filler chemistry and partly through particle size optimization that improves packing density. They remain electrically nonconductive and electrically safe.

Silver-Filled Compounds

Silver powder (Ag, bulk conductivity 429 W/mK) is an attractive filler because it conducts heat very well and, at fine particle sizes, packs densely in a carrier. Arctic Silver 5 contains approximately 88% micronized silver by weight and achieves rated conductivity of 8.9 W/mK. Thermal Grizzly Conductonaut (a liquid metal, discussed below) uses gallium alloys rather than silver, but several niche compounds use silver flake in non-silicone carriers.

The important nuance: silver compounds are weakly electrically conductive. The conductivity is low enough that a bridged SMD component is unlikely to cause an immediate short, but not so low that a thin smear across exposed pads is harmless. The guidance to keep silver compounds away from SMD components and CPU socket contacts is legitimate, not paranoia. The practical risk on a typical desktop CPU install — where sockets are covered by the CPU package and the exposed area is just the IHS — is low, but not zero if the compound is dramatically over-applied.

Silver compounds also interact chemically with silver-plated surfaces over time. The old concern about Arctic Silver 5 corroding thin silver plating on some heatsinks is real but affects a small minority of hardware. Copper surfaces are not meaningfully affected.

Carbon-Based Compounds

Graphite is anisotropic: in-plane conductivity exceeds 700 W/mK along the basal plane of the crystal, but cross-plane conductivity (the direction relevant to heat spreading) is only 5–10 W/mK. Compounds using carbon nanotubes, graphene flake, or diamond powder aim to exploit high-conductivity directions while suspending particles in an orientation-agnostic carrier. In practice, the gains from diamond-filled compounds are real but modest at the scale of thermal compound bond lines — measured improvements over top-tier ceramic pastes are often within measurement uncertainty.

Carbon-based compounds are electrically nonconductive (graphite nanoparticles in an insulating carrier are not a continuous conductor) and do not present corrosion risks. Their main drawback is cost: diamond particle compounds are expensive relative to their measured performance advantage.

Liquid Metal

Gallium-based alloys — primarily eutectic gallium-indium (EGaIn, melting point 15.7°C) and galinstan (gallium-indium-tin, melting point −19°C) — are the thermal performance ceiling for TIM-2 applications. Galinstan-based products like Thermal Grizzly Conductonaut rate at approximately 73 W/mK, nearly an order of magnitude above the best ceramic pastes. This is real: liquid metal repastes routinely produce 20–30°C drops on Intel consumer CPUs that use polymer TIM-1, because they address both the TIM-2 interface and (via delidding + relapping) the TIM-1 interface simultaneously.

Liquid metals have exactly two serious problems, and both are genuine.

Electrical conductivity is high: galinstan conductivity is approximately 3.46 × 10⁶ S/m. A drop on exposed motherboard circuitry will short it. A thin film across CPU socket contacts may or may not cause a fault depending on pin geometry, but the downside is total system failure. Application requires care: the spreader surface area must be fully covered but liquid metal must not run off the edge onto the PCB or socket area. Many experienced builders use kapton tape to mask the IHS edges before application. The viscosity is extremely low — galinstan is a true liquid at room temperature — which means it wicks more aggressively than paste.

Aluminum corrosion is the second problem and it is not minor. Gallium reacts with aluminum to form gallium-aluminum intermetallics (GaAl₃ and related phases), a process that proceeds at room temperature and is strongly exothermic once started. The reaction is sometimes called “gallium embrittlement” because it attacks grain boundaries and can mechanically disintegrate aluminum rapidly. A liquid metal compound applied to an aluminum heatsink base will begin dissolving the aluminum within minutes. This is not a slow aging process — it is a fast corrosion reaction. Liquid metal is only safe on copper, nickel, or nickel-plated copper surfaces. All-copper heatsink contact plates (common on premium coolers) are fine. Bare aluminum is destroyed. Many coolers use copper heatpipes pressed into an aluminum base, which means the contact plate material matters: check whether the copper plate extends to the contact surface or whether the aluminum base is what contacts the IHS.

Material Comparison

Material Type	Typical Conductivity (W/mK)	Electrically Conductive	Corrosion Risk	Relative Cost	Risk on Application
Silicone + ZnO/Al₂O₃	4–6	No	None	Low	Very low
Silicone + AlN	6–8	No	None	Medium	Very low
Silver-filled polymer	7–10	Weakly	Minimal	Medium	Low (care near pads)
Carbon/diamond blend	7–11	No	None	High	Very low
Liquid metal (galinstan)	65–85	Yes (high)	Al corrosion	Medium	High
Indium foil (TIM-1 factory)	~80	Yes	Minimal	N/A (OEM)	N/A
Solder (factory TIM-1)	50–60	Yes	None in situ	N/A (OEM)	N/A

Failure Modes That Actually Matter

Pump-Out

Thermal cycling — the expansion and contraction of metal surfaces as a CPU heats under load and cools at idle — creates shear stress across the bond line. A polymer paste has viscoelastic properties; it can flow slowly in response to sustained stress. Over many thermal cycles, this flow tends to be centrifugal: material is progressively squeezed out from the center toward the edges and eventually off the interface entirely. This is thermal paste pump-out, and it is a documented and well-studied failure mode in server and high-cycle-count environments.

Pump-out is not a meaningful concern for a desktop machine that is rebooted daily and runs at moderate temperatures. It is a real and maintenance-scheduled concern in server environments where CPUs run at sustained high temperatures for months or years, and thermal cycling is driven by workload spikes rather than power cycles. Enterprise servers from HP, Dell, and Lenovo include TIM replacement in their maintenance schedules partly for this reason. The pump-out rate increases with higher operating temperatures, larger die sizes, and lower-viscosity compounds. Silicone compounds with engineered thixotropy (higher viscosity at rest, lower under shear) are specifically marketed to server maintenance teams because of their pump-out resistance. Phase-change pads (discussed below) are increasingly specified in server designs partly because solid-at-operating-temperature materials cannot pump out.

Homelabbers running homelab servers continuously — as covered in the homelab hardware guide — should include TIM inspection in their annual maintenance pass. A server that has run 24/7 for three years under sustained CPU load is not guaranteed to have the same thermal performance it had on day one.

Dry-Out and Aging

Silicone carriers eventually lose volatile low-molecular-weight fractions through slow evaporation, especially at elevated temperatures. The compound thickens, loses the flexibility needed to maintain bond line contact, and eventually cracks or separates at the interface. This process is slow — years to a decade under normal conditions — but it is real and irreversible. Compounds based on synthetic oil carriers rather than silicone can behave differently: some oils oxidize at sustained high temperatures, leading to a different failure mode (gummy residue rather than powdery dry paste).

The practical consequence is that a thermal paste applied six or seven years ago on a system that has seen sustained high temperatures deserves replacement. Cleaning old compound off a heatsink and IHS — isopropyl alcohol at 90%+ concentration, or a dedicated TIM remover like Arctic Clean — and applying fresh compound often produces meaningful temperature drops, not because the new compound is chemically better, but because the old compound is no longer performing as specified.

Bond Line Thickness

The optimal bond line thickness (BLT) — the gap between the two metal surfaces filled with TIM — is thin: 25–75 µm is typical for good results. Too thick a bond line increases the thermal resistance proportional to thickness divided by conductivity. Too thin a bond line risks insufficient fill coverage if the surfaces are not perfectly parallel or if the compound was not spread to the edges before seating the cooler. Mounting pressure plays a role: coolers with high clamping force produce thinner BLTs than those with minimal spring tension, which is one underappreciated reason why the same compound produces different results on different cooler mounts.

Thermal Pads vs Paste vs Graphite Pads

These three categories solve the same fundamental problem — filling air gaps at an interface — but have meaningfully different use cases.

Thermal paste is the right choice for CPU-to-heatsink and CPU-to-IHS interfaces where the surfaces are machined flat and clamped under controlled pressure. The ability to conform to the exact gap under pressure and be squeezed to minimal BLT produces the lowest possible thermal resistance for a given surface quality. Paste does not tolerate being repositioned repeatedly: each seating cycle potentially introduces air bubbles, and some compounds (particularly those with curing mechanisms, like Arctic Silver 5’s initial 200-hour cure period) perform differently after repositioning.

Thermal pads — soft polymer sheets loaded with ceramic or metal filler — are appropriate where surfaces are not controlled flat, gaps are large and variable, or where paste application would be impractical or where pads are factory-specified. Common applications: VRAM chips to heatsink on a GPU (the gap varies chip to chip because VRAM dies are not perfectly coplanar), M.2 SSD thermal pads to heatsink, and any application where a user might reassemble the component multiple times. Pad conductivities range from 1 W/mK (cheap generic pads) to 8–15 W/mK for high-specification silicone pads with AlN filler. Thickness selection is critical: a pad that is too thick for the gap produces high thermal resistance because the excess compresses poorly; a pad too thin for the gap leaves voids. Phase-change pads (solid at room temperature, liquid at operating temperature) combine the application convenience of a pad with closer to paste-level performance.

Graphite pads (pyrolytic graphite sheet, PGS) are a distinct category: they are not loaded polymer composites but instead compressed and exfoliated graphite with in-plane conductivities of 700–1500 W/mK. Their cross-plane conductivity is far lower (5–15 W/mK depending on grade), but in applications where lateral heat spreading is the bottleneck rather than cross-plane conduction, they are genuinely excellent. They are reusable (no dry-out, no pump-out), do not require careful application, and do not make a mess. Their thermal performance in a pure CPU-cooler application is typically similar to a mid-tier paste because cross-plane conduction is the relevant direction there. The correct use case for PGS is scenarios like die-direct laptop cooling where lateral die spreading to a vapor chamber matters, or in applications like SSD heatsinks where you want to simultaneously conduct heat to a spreader and spread it laterally across the spreader surface.

Application Method: The Actual Answer

The X pattern versus pea-sized blob versus full pre-spread debate has generated thousands of forum posts and benchmark graphs with error bars wide enough to encompass the differences. The honest answer is that coverage matters, pattern mostly does not — with two caveats.

The goal is complete coverage of the die or IHS surface with a bond line thin enough to minimize resistance but with no voids that would leave air pockets at the interface. A properly sized pea in the center of a small die (8mm × 8mm, typical for a dense high-performance die) achieves this with good cooler clamping pressure. A large IHS (Intel Alder Lake’s die is spread under a ~37mm × 37mm spreader) may genuinely need more paste or a different distribution to ensure full coverage, because the clamping pressure gradient and the compound’s viscosity may not spread a single pea to the edges before the cooler seats.

The practical guidance:

For small to medium dies and IHS: a pea-sized (3–4mm diameter) blob in the center, applied with normal cooler clamping, produces coverage statistically indistinguishable from careful pre-spread in controlled tests.
For large IHS surfaces: a cross or X pattern, or pre-spreading to ensure edge coverage, reduces the small risk of edge voids. The thermal difference if a void does form is measurable.
For liquid metal: pre-spreading with the applicator brush is mandatory. The low viscosity and the need to confirm complete coverage before seating makes center-blob application unreliable.

The variation between careful correct applications of the same compound is typically 1–3°C in well-controlled tests. The variation between a good application and a void-riddled application can be 10–15°C. Getting the application correct matters far more than which correct method you choose.

Die-Direct Interfaces

An increasing number of platforms — Intel’s Arrow Lake Core Ultra 200S desktop CPUs, AMD’s threadripper, and most high-end laptop chiplets — use die-direct cooling: TIM is applied directly between the silicon die and the cooler contact plate, with no IHS in the stack. This eliminates TIM-1 entirely and puts TIM-2 directly on silicon.

Die-direct imposes additional constraints. Silicon is hard and brittle but not as thermally conductive as copper, and its surface finish varies by die geometry (logic dies are effectively flat; chiplet packages have height differences between dies that must be accommodated). The bond line on a bare die is thicker if dies are at different heights — another argument for slightly softer or more conformable pastes on multi-chiplet packages. Liquid metal is a valid choice on bare silicon (no aluminum corrosion risk if the cooler is copper or nickel-plated), but the low viscosity and the risk of the compound running into gap-fill regions between chiplets or under the package adds complexity that makes higher-viscosity premium pastes the more common recommendation.

The electrical conductivity risk of silver or liquid metal is higher on die-direct platforms: exposed contact pads and substrate circuitry are immediately adjacent to the die, rather than hidden under a large IHS with generous clearance. The alignment margin for error is smaller.

The UPS discipline that applies to server hardware — documented maintenance schedules, inspection cycles, and understanding failure modes before they become failures — applies equally to the thermal materials in high-value computing hardware. Power continuity and thermal continuity belong in the same reliability framework. If you are managing rack servers and already thinking carefully about power as described in the UPS sizing and NUT guide, thermal maintenance belongs in the same periodic checklist.

Verdict

Thermal paste is not magic and it is not a performance multiplier. It is a gap-filling material doing exactly one job: replacing 0.025 W/mK air with something 160 to 3000 times more conductive. The best compound for most applications is a quality silicone-based paste with AlN filler (Thermal Grizzly Kryonaut, Noctua NT-H2, or equivalent), applied in sufficient quantity to achieve full coverage, with enough mounting pressure to minimize bond line thickness. The marginal performance gain from liquid metal on a typical desktop CPU-to-IHS interface is real (2–5°C typically) but not worth the electrical and corrosion risk for most users. Liquid metal makes strong sense for delidding builds where TIM-1 is being replaced and the copper IHS is being reseated, and for die-direct platforms where the builder understands the application requirements.

The failure modes that actually end hardware — dry-out after years at elevated temperatures, pump-out in servers with sustained thermal cycling — are more important to understand than the 2 W/mK difference between premium paste tiers. A repaste every three to five years on any system running sustained high CPU temperatures is not excessive caution; it is basic maintenance. The tube of compound costs less than the replacement CPU.

Application method debate is noise. Coverage is signal. Use enough compound, seat the cooler without tilting, and verify with a temperature delta that the application did not introduce void artifacts. Everything else is within measurement noise.

Sources

Prasher, R. (2006). “Thermal Interface Materials: Historical Perspective, Status, and Future Directions.” Proceedings of the IEEE, 94(8), 1571–1586.
Hansson, J., et al. (2018). “Novel nanostructured thermal interface materials: a review.” Electronic Materials Letters, 14(3), 259–275.
Intel Corporation. “Thermal Interface Material (TIM) Characterization Guidelines.” Application Note AN-907, 2019.
Zweben, C. (2012). “Advances in Composite Materials for Thermal Management in Electronic Packaging.” JOM, 50(6), 47–51.
Thermal Grizzly product documentation: Conductonaut application guide (aluminum incompatibility notice), 2023.
Deppisch, C., et al. (2006). “The material optimization and reliability characterization of an indium-solder thermal interface material for CPU packaging.” JOM, 58(6), 67–74.
Narumanchi, S., et al. (2008). “Thermal Interface Materials for Power Electronics Applications.” ITHERM 2008 Proceedings, IEEE.
Chung, D.D.L. (2001). “Thermal Interface Materials.” Journal of Materials Engineering and Performance, 10(1), 56–59.