Which materials barely change size when temperature changes?

Expansion, conductivity and stiffness at a glance

Which materials barely change size when the temperature changes?

Among the dated values, fused silica changes size least: 0.55 ppm/K over 20 to 320 °C. Invar 36 (FeNi36) follows at 1.18 to 1.3 ppm/K near room temperature (1.3 from 21 to 93 °C, 1.18 from 30 to 100 °C), then Borosilicate glass (3.25 ppm/K over 20 to 300 °C) and Silicon nitride (Si3N4). CFRP shows 0.18 ppm/K, lower still, but it is one coupon value with no temperature stated, so it is indicative only.

Invar's low number only holds near room temperature. One table gives 2.11 ppm/K from 21 to 149 °C and 4.18 ppm/K from 21 to 260 °C, and a second supplier gives 4.92 ppm/K from 30 to 300 °C, so the figure climbs as the part gets hotter. The only cryogenic entry, 1.8 ppm/K between -240 and -18 °C, is a mean over a much wider window, so it is not comparable with the 21 to 93 °C mean and does not show how Invar behaves when cold. The wide profile range is the same alloy at different temperatures.

The ceramics and tungsten sit in the middle. Silicon carbide (SiC) reads 4.4 ppm/K for a direct-sintered grade over 25 to 1000 °C, Tungsten (W) reads 4.5 ppm/K with no temperature printed, and Gallium arsenide (GaAs) reads 5.73 to 6.86 ppm/K. Aluminium nitride (AlN) reads 4.5 ppm/K for a typical commercial grade with no temperature printed, and 5.5 to 5.6 ppm/K over 25 to 1000 °C for three grades in a second table.

Alumina (Al2O3) reads 8.2 ppm/K over 25 to 1000 °C. YSZ (yttria-stabilised zirconia) reads 9 to 10.3 ppm/K over 25 to 400 °C and 25 to 1000 °C for two grades. Silicon nitride (Si3N4) has three entries: 1.4 near room temperature and 3.2 as a mean to 1000 °C for a gas-pressure-sintered grade, and 2.9 for a hot-isostatically pressed grade. The 1.4 is a single reading and an outlier, so do not rank silicon nitride on it.

Silicon itself is missing. Our profile for Silicon has no expansion value, so the common question of which ceramic matches a silicon chip cannot be answered from numbers here. The practical route is to get silicon's CTE over your own temperature window and compare it with each candidate's.

Which metals and plastics grow the most when heated?

Aluminium alloy 6061 grows the most of the metals, at 23.6 ppm/K over 20 to 100 °C; the datasheet also lists an estimated 25.2 for 20 to 300 °C. Copper (Cu) reads 17.6 ppm/K as a single value at 20 °C, not a 20 to 100 °C mean, and Stainless steel 316 reads 16.5 to 18.2 ppm/K, rising from the 20 to 100 °C window to the 20 to 500 °C window. Ti-6Al-4V is the quiet structural metal at 8.6 ppm/K, so aluminium grows 2.7 times as much. Titanium reads near Alumina (Al2O3) (8.2 ppm/K) and YSZ (9 to 10.3 ppm/K), but their windows are wider, so the closeness is only indicative.

Carbon steel 1018 has no CTE entry. The Invar profile puts ordinary steel at roughly 11 to 15 ppm/K, between titanium and copper, but that is not a steel record, so use it only as a rough guide.

Plastics grow far more. PA 66 (unreinforced nylon 66, Ultramid A3K) reads 98 ppm/K over 23 to 55 °C, which is 4.2 times as much as aluminium (different windows). It also takes up 2.8 % water at 50 % relative humidity and 8.5 % at saturation, and its tensile modulus falls from 3 GPa dry to 1.2 GPa conditioned, so size and stiffness both depend on moisture. Polyetheretherketone (PEEK) and PC have no CTE entry; other plastic profiles in the database do, but are outside this set. General engineering knowledge, not in our data: fillers lower a plastic's expansion, and the value differs by flow direction.

Composites are direction-dependent. The CFRP profile gives 164 GPa along the fibres and 11.7 GPa across them in one laminate, holds no laminate density or conductivity, and its single CTE value does not say which direction it belongs to. Ask for the CTE along and across the fibres before bonding a laminate to metal.

Do low-expansion materials also conduct heat well?

Not as a rule. Among the materials with both values (CFRP has no conductivity entry), every one at or below 3.25 ppm/K near room temperature conducts 29 W/(m·K) or less: Fused silica (SiO2 glass) 1.3 to 1.4 W/(m·K), Borosilicate glass 1.2 W/(m·K), Invar 36 (FeNi36) 10.5 W/(m·K) and Silicon nitride (Si3N4) 24 to 29 W/(m·K) depending on grade. The good conductors start higher: Tungsten (W) pairs 4.5 ppm/K with 174 W/(m·K), and Copper (Cu) and Aluminium alloy 6061 conduct well but grow most. Silicon conducts 150 W/(m·K) and has no CTE here; it is widely regarded as a low-expansion material (general engineering knowledge, not in our data), which would break the pattern.

General engineering knowledge, not in our data: a low-expansion, low-conductivity part such as silica or Invar only stays stable if it is all at one temperature; if one face is warmer, each region grows by its own amount and the part bends.

Silicon carbide (SiC) has one clean pair: the direct-sintered grade reads 4.4 ppm/K and 150 W/(m·K) from the same source. Its conductivity otherwise runs from 140 W/(m·K) for reaction-bonded material up to 490 W/(m·K) for single-crystal 4H, so match conductivity and CTE to the same grade. For scale, copper's top-of-range conductivity is 279 times that of fused silica, across different grades and temperatures.

Silicon nitride (Si3N4) is a middle choice: it conducts less than Aluminium nitride (AlN) (140 to 180 W/(m·K)) but expands less. Gallium arsenide (GaAs) conducts 46 to 55 W/(m·K) with 5.73 to 6.86 ppm/K.

Why does CTE mismatch matter for optics, electronics and space structures?

A mismatch loads the joint. General engineering knowledge, not in our data: for a thin layer fully constrained by a thick partner, stress is roughly E x Δα x ΔT / (1 - ν), where E is stiffness, Δα the CTE difference, ΔT the temperature swing and ν Poisson's ratio, and real joints carry less. The stiff materials need most care: Tungsten (W) is 411 GPa, Silicon carbide (SiC) 370 to 421 GPa and Alumina (Al2O3) 370 GPa, with Aluminium nitride (AlN) up to 330 GPa and Silicon nitride (Si3N4) up to 320 GPa, against 68.9 GPa for Aluminium alloy 6061 and 72.7 GPa for Fused silica (SiO2 glass). PC is only 2.35 to 2.4 GPa (two readings of one grade), so it takes less stress from a given mismatch, but it has no CTE entry.

In optics, a mismatch moves focus and distorts mirrors (general engineering knowledge, not in our data). A 1972 JPL and Itek study used Invar 36 for the structures that control spacing between optical elements and paired Invar rods with shorter aluminium paths so the opposing expansions cancel; it rejected Super-Invar because its expansion was less stable over long periods. Mirror substrates in our data include Fused silica (SiO2 glass), Borosilicate glass (the 200 inch Hale telescope mirror) and Silicon carbide (SiC) mirrors for LIDAR and space telescopes.

In electronics, general engineering knowledge, not in our data, says a mismatch cycles the solder, die attach or substrate bond every time the device powers up. Aluminium nitride (AlN) is sold for IC packages, power transistor bases and heat sinks, and one supplier says its expansion is close to silicon's; that is a vendor statement with no silicon value behind it in our data, so treat it as unverified. For space solar cells, Fused silica (SiO2 glass) is used as cover glass where radiation would discolour cheaper glass, and a 1980 NASA study says cerium-doped Borosilicate glass meets the front-cover need if it bonds to the cell without thermal-mismatch damage.

In space structures, general engineering knowledge, not in our data, says a part swings between sun and shadow, so a mismatch cycles many times, and designers match the CTE of fittings to the structure they clamp. Invar's one cryogenic entry is a wide-window mean, and no thermal-vacuum or radiation test for it is in the profile.

Which low-expansion material fits optics, electronics or a space structure?

For glass optics, choose Fused silica (SiO2 glass) when the lowest expansion matters and Borosilicate glass when moderate expansion (3.25 ppm/K) is enough. The SCHOTT sheet gives two different ratings: thermal shock resistance (RTS) of 140 to 175 K by panel thickness, and resistance to temperature differences (RTD) of 110 K for a 1 hour exposure. A damaged fused silica pane can fail suddenly under tensile stress.

For metal supports of optics, choose Invar 36 (FeNi36) and plan the heat treatment: the profile says to heat to 815 °C, quench, age and do this before final machining, because cold work leaves slow dimensional drift. Keep it out of damp air and magnetic fields, which the profile says hurt corrosion resistance and stability.

For electronics, choose Aluminium nitride (AlN) when conductivity (140 to 180 W/(m·K)) and insulation both matter, Silicon nitride (Si3N4) when strength and thermal shock matter more, and Alumina (Al2O3) (8.2 ppm/K) when a plain insulator with a higher expansion is acceptable. Alumina's thermal shock temperature difference is 200 °C for one grade of at least 99 percent purity. That is a different test from borosilicate's ratings, and AlN has no equivalent figure, so shock numbers are not comparable here.

For space structures, Ti-6Al-4V and Invar 36 (FeNi36) cover the metal side, Silicon carbide (SiC) the stiff ceramic side, and Aluminium alloy 6061 the light default if the mismatch is designed in. Tungsten (W) is brittle and oxidises in hot air, so it is a specialist choice.

Terms used

CTE

coefficient of thermal expansion: how much a material grows per degree of temperature rise, written as a fraction of its length.

ppm/K

parts per million per kelvin, the unit used here. One ppm/K means a 1 metre bar grows 1 micrometre for every 1 K of warming. A change of 1 K equals a change of 1 °C.

Mean CTE

the average expansion over a stated temperature window, such as 20 to 100 °C. Most datasheet values are means, so the window matters as much as the number.

CTE mismatch

the difference in expansion between two bonded or touching materials. The joint has to absorb it as stress or bending.

Thermal conductivity

how well a material carries heat, in W/(m·K). Low values let one side of a part run hotter than the other.

Elastic modulus

stiffness, in GPa. A stiffer material turns a given expansion mismatch into a larger stress.

Thermal shock

cracking caused by a sudden temperature change, because the surface and core change size at different moments.

Anisotropy

a property that differs by direction, as in a carbon-fibre laminate, which is stiffer along the fibres than across them.

What to check before you order

1. Ask for the CTE with its temperature window (for example 20 to 100 °C or 25 to 1000 °C) and whether it is a mean or an instantaneous value.
2. Ask for the CTE of every partner material over the same window, then compare the pair, not each material alone.
3. Ask which grade, form and direction the value belongs to. Ceramics and glass vary by grade; plastics and composites vary by direction and by moisture state.
4. For Invar, ask whether the lot is stress-relieved and stabilised, and whether the part is machined after heat treatment.
5. Ask for the thermal conductivity at operating temperature and think about gradients across the part.
6. Ask for the lowest and highest temperature in service, including storage and shipping.
7. Measure the CTE of a sample from your own lot if the joint is critical, and check the thermal shock limit for fast heating.

What do we not know yet?

• Molybdenum and Kovar-type glass-to-metal alloys, the usual partners for silicon and glass, have no profile, so they cannot be compared with tungsten. Silicon, steel 1018, PEEK and polycarbonate have profiles but no CTE entry, and a pyrolytic graphite profile exists but is below the full-grade bar, so it is not used.
• CTE values are means over different windows and no profile holds a full expansion-versus-temperature curve. Cryogenic CTE exists only for Invar, as one wide-window mean.
• The database holds no joint data (solder, adhesive, die attach, braze) and no stress or fatigue-life data, so mismatch stress can only be estimated with the general formula.
• ULE, Zerodur, Super Invar and other near-zero-expansion materials have no profile. The carbon-fibre-epoxy profile holds one undated coupon value with an unnamed matrix, and no along-fibre versus across-fibre CTE, so composites cannot be ranked reliably.
• Silicon carbide conductivity mixes single-crystal and ceramic grades, so conductivity and CTE cannot always be tied to the same grade.
• No test of expansion change after radiation, vacuum or thermal cycling is entered for any of these materials.

Questions people ask