Thin-film coatings quietly determine whether a solar cell hits 22 % or 26 % efficiency, whether a tool bit lasts 10 000 or 50 000 cuts, and whether a phone screen survives a sidewalk drop. Choosing between evaporation and sputtering is therefore not an academic exercise; it is the moment when a design team locks in yield, cost, and performance for the entire product life cycle.
Both processes live in the same vacuum family, yet they differ in how atoms leave the source, travel through space, and land on the substrate. Those differences ripple outward into film density, stress, adhesion, alloy uniformity, and ultimately the dollars spent per wafer.
Fundamental Physics: How Atoms Escape the Source
Thermal evaporation heats a solid until its vapor pressure exceeds the chamber background, creating a gentle upward wind of single atoms. The energy histogram peaks below 0.2 eV, so arriving atoms skid across the surface like pebbles on ice, leaving voids unless the substrate is hot enough to mobilize them.
Sputtering fires noble-gas ions at 500–1 000 eV into a target; momentum cascades knock loose atoms with typical energies of 5–30 eV. Those ballistic arrivals hammer the growing film, compacting it and disrupting native oxides so the first monolayer welds itself to the substrate.
Higher arrival energy also means resputtering: freshly landed atoms can be knocked off step edges, smoothing sharp corners and widening bottom coverage in trenches. Engineers exploit this “atomic sandblasting” to pinch off void-free barrier layers at 5 nm thickness inside 20:1 aspect-ratio vias.
Hardware Architectures Compared
Evaporation sources are either resistive boats, electron-beam crucibles, or induction-hefted hearths; all pool the material in liquid form and point it upward like a campfire. Substrates face downward in a planetary fixture that rotates to even out the cosine flux distribution and avoid the “coffee-ring” thickness profile.
Sputtering sources are magnetron cathodes where plasma is trapped in a racetrack magnetic field, eroding a narrow groove at 50–200 nm/kWh. Power is supplied by DC, pulsed-DC, or RF generators; the latter two let you poison the target surface with reactive gas without arcing, opening the door to stoichiometric nitrides and oxides.
Chamber footprints diverge: a batch evaporation tool for 200 mm wafers needs 0.8 m² of floor space and one rough pump, while an inline sputter cluster for the same throughput demands 4 m² and three turbo pumps. Clean-room real estate thus becomes a hidden line item in the cost model.
Target and Crucible Utilization Economics
An e-beam hearth loaded with 50 g of platinum costs $1 800 today and coats 3 µm over 500 4-inch wafers before the pit depth triggers shadowing. That is $3.60 per wafer just for material, ignoring yield loss from spit particles when the melt becomes super-heated.
A 200 mm TiAlN sputter target priced at $1 200 delivers 80 µm of useful erosion before the racetrack breaks through the backing plate, enough for 25 000 wafers at 50 nm film thickness. Consumables drop to five cents per wafer, but the cathode itself adds $60 k in capital depreciation that must be amortized.
Film Microstructure and Performance Maps
Evaporated aluminum mirrors for space telescopes show columnar grains 80 nm wide separated by 2 nm voids; their intrinsic tensile stress is 30 MPa, low enough to survive 100 K thermal cycles without cracking. Yet those same voids act as diffusion pipes, letting copper migrate through at 150 °C and kill reflectivity within months.
Sputtered AlSiCu alloy with 1% Si and 0.5% Cu fills the gaps with supersaturated solute, pushing resistivity to 3.2 µΩ·cm versus 2.7 µΩ·cm for pure evaporated Al. The payoff is electromigration lifetime above 300 h at 250 °C and 2 MA/cm², a 10× boost that justifies the extra mask step to remove the native oxide skin.
When you need a barrier that is hermetic at 3 nm, only ion-assisted densification will do. TaN sputtered at 600 W in 20 mTorr Ar/N₂ forms an amorphous film with 12 g/cm³ density, stopping copper diffusion at 400 °C for 100 h, while e-beam Ta evaporated at the same thickness fails within minutes.
Stress Engineering Levers
Evaporation stress is tuned almost entirely by substrate temperature: 200 °C yields 100 MPa tension, 400 °C flips it to 50 MPa compression as adatoms rearrange. The lever is coarse; wafers warp if the stack is thick, and photoresists burn long before the stress sign reverses.
Sputtering gives three independent knobs: gas pressure, ion energy, and substrate bias. Dropping Ar from 5 to 1 mTorr raises ion peening energy and swings TiO₂ stress from −800 MPa compressive to +200 MPa tensile without exceeding 150 °C. A −100 V RF bias adds another 300 MPa compressive increment, letting you match the thermal expansion of GaN on sapphire.
Step Coverage and 3-D Conformality
Evaporation sources follow a cosine emission law; at a 45° sidewall the flux drops to 35 % of the planar value. A 100 nm seed layer on a 1 µm trench floor therefore thins to 35 nm halfway up the wall and pinches to 5 nm at the corner, creating a weak link that electro-migrates first.
Collimated sputtering uses a honeycomb filter 50 mm above the cathode to trap oblique atoms, tightening the arrival angle to ±15°. Coverage on the same trench improves to 65 % at mid-wall, good enough for magnetic tunnel junctions where pinholes kill yield, but still insufficient for 10:1 DRAM capacitors.
High-power impulse sputtering (HiPIMS) delivers 3 kW in 100 µs bursts, ionizing 70 % of the Al flux. A −60 V substrate bias steers those metal ions around corners, yielding 95 % sidewall coverage at 20:1 aspect ratio without the precursor cost of atomic layer deposition.
Alloy and Compound Synthesis Control
Evaporating alloys is like simmering a stew: each element boils at its own temperature, so the melt enriches in the refractory component and the vapor drifts in composition. CoCrPt magnetic media requires 14 at % Cr; maintaining that within 0.5 % across a 200 mm wafer demands two separate e-beam guns whose deposition rates are PID-locked to quartz-crystal feedback updated at 10 Hz.
Sputtering from a single alloy target naturally preserves the bulk composition to ±1 % because the racetrack erodes stoichiometrically. When you need a gradient—say, CoCrPt with 20 % Cr at the bottom interface for corrosion resistance tapering to 10 % at the cap—you cosputter two cathodes and ramp their power ratio in real time using a linear motor shutter.
Reactive sputtering adds O₂, N₂, or C₂H₂ to form oxides, nitrides, or carbides in situ. The hysteresis curve between metal and poisoned target states is steep; a 1 % O₂ spike can drop Ti sputter yield by 60 % and crash the rate. Closed-loop optical emission spectroscopy watching the 777 nm oxygen line keeps the target 5 % oxidized, delivering stoichiometric TiO₂ at 30 nm/min without arcing.
Defect Sources and Yield Killers
Spatter from an e-beam melt ejects micro-droplets 0.5–5 µm across that land as nodular seeds in the film. These seeds shadow the flux, grow into cones, and eventually shed, leaving pinholes that short metal lines. The fix is a swept e-beam raster that avoids local super-heating and a crucible liner wetted by the melt to reduce popping.
Sputter defects come from arcing when the target oxide charges up; the 50 A micro-lightning crater launches 100 nm particles that embed as killer defects in gate dielectrics. Mid-frequency pulsed DC at 100 kHz reverses target voltage every 5 µs, discharging the dielectric layer and cutting particle density from 0.4 to 0.04 cm⁻².
Substage outgassing is another stealth yield thief. Evaporation chambers rarely exceed 150 °C bake, so water desorbs slowly and reads 2 × 10⁻⁶ Torr on the ion gauge. A 30 s plasma clean immediately before sputter deposition knocks that water off the wafer surface, dropping base pressure to 5 × 10⁻⁸ Torr and eliminating the hydroxyl absorption that blisters films during subsequent anneals.
Throughput and Cost-of-Ownership Models
A single-chamber e-beam evaporator can coat ten 200 mm wafers with 100 nm of Au in 25 min including pump-down, giving 24 wafers per hour. Gold costs $60 per gram and utilization is 40 %, so material adds $4.50 per wafer; energy and labor push total CoO to $7.80.
An inline magnetron system with seven cathodes runs 60 wafers per hour at 50 nm TiN, but the target utilization is 80 % and the capital ticket is $4 M depreciated over five years. Even at 20 % uptime the depreciation slice is $9.20 per wafer, yet the throughput gain lets the fab ship 1.2 M wafers annually, beating the evaporation line’s 200 k by 6× and dropping the per-wafer capital cost below $2.
Hidden in the spreadsheet is the clean-room cost of queue length. Evaporation batches queue behind litho and etch because each run is 25 min; sputter inline tools feed every 60 s, so WIP inventory shrinks and cycle time falls from ten days to three, unlocking cash flow that dwarfs the equipment price delta.
Energy Consumption Footprint
Evaporation of 1 µm Al on a 300 mm wafer consumes 0.3 kWh, mostly in the 10 kW e-gun beam. Sputtering the same thickness at 20 kW DC magnetron uses 0.8 kWh because the plasma burns continuously and the turbo pump draws 5 kW. For a fab producing 50 k wafers per month, the annual difference is 300 MWh—enough to power 30 homes, so regions with carbon taxes tilt the model toward evaporation unless film performance is non-negotiable.
Hybrid Workflows: Best of Both Worlds
Some MEMS foundries open with a 10 nm sputtered Ti adhesion layer to nail down the oxide, then transfer the wafer under vacuum to an e-beam chamber that adds 200 nm of Au at 1 nm/s for low stress and mirror finish. The stack retains the 800 MPa compressive strength of ion-bombarded Ti while keeping the gold surface smooth enough for optical waveguides with 0.2 nm rms roughness.
Display makers reverse the order: they evaporate a 50 nm Ag alloy cathode to avoid plasma damage on the organic light-emitting layers, then sputter a 20 nm ITO cap at 150 °C to densify the transparent conductor and block moisture. The hybrid sequence cuts dark-spot growth rate in half compared with all-sputtered encapsulation.
Cluster tools now integrate both sources on a central robot so wafers never see air. A 200 mm tool can deposit Ti/Cu/Ni/Au stack in 8 min with <0.5 nm oxide between layers, enabling 3 µm-pitch Cu pillars for 2.5-D interposers without electroplating baths.
Decision Matrix: Picking the Right Tool Tomorrow
If your spec sheet lists <5 nm thickness tolerance, <100 MPa stress, and aspect ratio <2, evaporation wins on cost and simplicity. Gold pads for wire-bonded RF amplifiers, reflectors for LiDAR, and lift-off metallization for GaAs HEMTs fall squarely in this box.
When the roadmap demands 10:1 vias, hermetic barriers, or ternary alloys tuned to ±0.5 at %, sputtering is the only path. Memory word-lines, copper damascene barriers, and magnetic tunnel junctions all shift the balance to ion-assisted deposition even if the CoO spreadsheet groans.
Start-ups often lease a compact sputter system first because it covers 80 % of possible films, then add an e-beam module later for precious metals. The modular cluster footprint lets them pivot from sensors to photonics without rebuying vacuum hardware, preserving capital while the product-market fit is still fluid.