Thermal evaporation is the only deposition technique in a fab where the operator does not set the deposition rate — the operator sets a temperature, and the rate is whatever thermodynamics decides it should be. Every other process has a knob that maps roughly linearly onto the thing you want. Sputter power sets sputter rate. Precursor flow sets growth rate. Evaporation has no such knob. It has a hot source, and above that source sits a vapour pressure that is exponential in temperature, so the relationship between the setting and the outcome is not a proportionality but an amplification. Nearly every characteristic of the technique — its notorious rate instability, its inability to hold an alloy composition, its habit of putting crucible material into the film, and the specific niches where it is nonetheless the only acceptable choice — follows from that single exponential.

The flux leaving a hot surface into vacuum is set by kinetic theory, and the vapour pressure that drives it is set by the enthalpy of vaporisation:

The square-root term in the denominator is a mild correction. The exponential in the numerator is the entire story. Differentiate it and the sensitivity of rate to temperature is the enthalpy of vaporisation divided by the gas constant times temperature squared, which for a typical metal near its useful evaporation point works out to something in the neighbourhood of thirty to fifty times. A one percent error in source temperature is not a one percent error in rate — it is a thirty to fifty percent error in rate. This is why an evaporator is never run open-loop on temperature. It is run closed-loop on a quartz crystal monitor that measures the deposited mass directly and drives the source power to hold the measured rate, with the thermal mass of the source acting as an unhelpfully slow and unhelpfully nonlinear plant. It is also why the shutter exists: the source is brought up, allowed to stabilise against the crystal for a while with the wafer shielded, and only then exposed. Deposition on the wafer starts when the shutter opens, not when the power comes on, because the first minute of any evaporation is unusable.

The same exponential explains why evaporation is described by vapour pressure rather than by melting point, a distinction that trips people up constantly. What matters is not whether the source is liquid but whether it has a useful vapour pressure — conventionally around ten millitorr, the point at which a practical deposition rate is obtained. Some materials reach that pressure while still solid and never melt at all, which is sublimation, and chromium is the standard example: a chromium source is a solid rod or a plated filament that never becomes a puddle. Others must be well above melting before they evaporate usefully. And a few, most of the refractory metals and most of the oxides, cannot be brought to a useful vapour pressure by resistive heating at all, because the boat holding them would fail first. That last case is exactly the boundary where thermal evaporation stops and electron-beam evaporation begins, and it is a boundary about the heater rather than about the material.

Holding the material is a harder problem than heating it, and it is the source of most contamination in evaporated films. A resistively heated source is a refractory metal boat, basket, or filament carrying hundreds of amperes, in direct contact with a molten charge, at a temperature where solubility is not negligible. Molten aluminium dissolves tungsten. It also wets tungsten aggressively, creeping along the filament by capillary action and away from where the heat is, which is why an aluminium filament source has a short and somewhat unpredictable life and why aluminium is more often run from an alumina-coated or boron-nitride-lined boat. Gold alloys with almost every refractory metal it touches. The general pattern is that whatever holds the charge slowly enters the charge and therefore slowly enters the film, at a level that is small, real, and difficult to measure. The escape is to stop touching the melt: a lined crucible heated indirectly by a surrounding element, or a properly designed effusion cell, separates the hot structural element from the material and gives both cleaner films and far better rate stability, at the cost of thermal response so slow that the shutter becomes the only fast actuator in the system.

Compositional control is where evaporation is genuinely weak, and it is worth being explicit because the failure mode is silent. When a binary charge is evaporated from a single source, each component leaves at its own vapour pressure, so the vapour is not the alloy — it is enriched in whichever component is more volatile:

Because the ratio of the pure-component vapour pressures can be orders of magnitude even for metals that seem similar, the film that lands is generally nothing like the ingot that was loaded. Worse, the composition is not merely wrong, it is drifting: the melt depletes in the volatile component as evaporation proceeds, so the film composition changes continuously through the thickness. An evaporated alloy is therefore a graded film with a composition that depends on how much of the charge has been used, which means the first wafer of a charge and the last wafer of a charge are not the same product. The workable answers are to co-evaporate from separate sources with independent rate control, which is how compound semiconductors and doped organic layers are actually done, or to abandon the technique for anything where composition matters and use sputtering, whose central practical virtue is that a steady-state target erodes stoichiometrically and transfers the composition of the target to the film. That contrast is the main reason sputtering displaced evaporation from mainstream metallisation, and it is a compositional argument rather than a coverage one.

Given all that, the interesting question is why thermal evaporation is still in use at all, and the answer is that its weaknesses and its strengths are the same property. The flux is thermal, which means the arriving atoms carry roughly a tenth of an electron volt rather than the several to tens of electron volts of sputtered atoms, and there are no energetic ions in the flux whatsoever. Nothing gets peened, nothing gets implanted, no plasma sits on the substrate, and no ultraviolet or soft X-ray flux illuminates the device underneath. For a gate stack, an organic light-emitting layer, a two-dimensional material, or a delicate contact, that absence of damage is not a minor convenience — it is the reason the process exists. The source is also small and far away, so the flux is close to a point source and highly directional, which gives essentially no sidewall coverage. For most of the fab that is a fatal defect. For lift-off patterning it is the enabling property: the metal must not coat the resist sidewall, or the solvent cannot reach the resist and the pattern will not lift. Evaporation is the standard lift-off metallisation for exactly the reason it is useless for filling a via.

The remaining consideration is thermal, and it is easy to overlook because it is not in any recipe. A source sitting at fifteen hundred kelvin a few tens of centimetres from the wafer is a radiator with a direct line of sight to it, and the substrate absorbs that radiation for the entire deposition. Substrate temperature therefore rises with source temperature, with deposition time, and with how open the geometry is — a coupling that is invisible until it matters, and it matters whenever the substrate is photoresist, an organic layer, a polymer, or anything else that will not tolerate the tens of degrees of unrequested heating that a long high-rate run delivers. Standing off further reduces the radiant load and improves directionality at the same time, which is why long-throw geometry is common in lift-off tooling, and the price is paid in deposition rate falling with the square of the distance and in most of the evaporated charge landing on the chamber walls rather than on the wafer. Every one of those trades is a consequence of the same fact: in evaporation, the source is a hot object obeying its own thermodynamics, and the process engineer is negotiating with it rather than commanding it.