Six Million Hertz Minus What the Coating Weighs

Vacuum coating chambers don’t have windows. You can’t see the deposition happening. Magnesium fluoride vaporises at 1263°C in near-perfect vacuum (10⁻⁵ torr), travels ballistically across the chamber, and condenses molecule-by-molecule onto glass you’re trying to coat with a 137.5-nanometre anti-reflection layer. Miss the target thickness by 10 nm and your optic reflects the wrong wavelength—purple fringing instead of clear transmission. You need sub-nanometre precision with no direct measurement.

The solution: measure mass, not thickness. A quartz crystal oscillator sits inside the chamber, collecting the same coating as your optic. As atoms land on its surface, it gets heavier. Heavier crystals vibrate slower. That frequency shift—measured in hertz—tells you exactly how much material you’ve deposited.

It’s called a quartz crystal microbalance. The physics is stranger than the name suggests.

Piezoelectric resonance as a scale

Quartz crystals are piezoelectric: apply voltage and they deform; deform them mechanically and they generate voltage. Cut a thin disc of AT-cut quartz (35°15’ from the optical axis), metallise both faces with gold electrodes, apply alternating current, and it oscillates. Not flexing—shearing. The crystal thickness expands and contracts at a resonant frequency determined by its mass and elastic modulus.

Typical QCM crystals resonate around 5-6 MHz. That’s fundamental mode, thickness-shear vibration. The Q factor exceeds 10⁶, which means the resonance peak is extraordinarily narrow—frequency divided by bandwidth approaches a million. Commercial QCM controllers resolve 1 Hz shifts. At 6 MHz, that’s 0.00002% precision. You’re measuring fractional changes in inertia as single atomic layers accumulate.

The Sauerbrey equation converts frequency to mass:

Δm = -(C × Δf) / n

Where:

Δm = mass change per unit area (g/cm²)
C = crystal sensitivity constant (Hz·cm²/g), typically 56.6 Hz·cm²/µg for a 6 MHz AT-cut crystal
Δf = frequency shift (Hz)
n = harmonic number (1 for fundamental)

For a 6 MHz crystal, 1 Hz of frequency drop corresponds to roughly 0.018 µg/cm² of deposited mass. Convert mass to thickness by dividing by material density. Magnesium fluoride (ρ = 3.18 g/cm³) deposits 0.056 nanometres per hertz. You’re watching atomic layers land in real time.

The equation assumes the deposited film is rigid, thin compared to crystal thickness, and uniformly distributed. Once you exceed ~2% of the crystal’s thickness (roughly 30 µm), viscoelastic effects start distorting the resonance. The film’s own stiffness couples to the shear wave. Below that limit, the approximation holds—frequency drops linearly with mass.

Line-of-sight deposition geometry

At 10⁻⁵ torr, mean free path for vaporised atoms exceeds 10 metres. They don’t collide with residual gas—they fly straight. The coating material travels ballistically from source (thermal evaporation boat, electron-beam hearth) to target in Newtonian trajectories. That’s why vacuum coating is called “line-of-sight” deposition: atoms only land where they can see the source.

Your QCM crystal sits beside your optic, angled to collect the same flux. Geometric tooling factor corrects for slight differences in solid angle. If the crystal sees 0.95× the atom flux your lens sees, you scale the readout accordingly. Commercial systems calculate this from source-to-crystal and source-to-substrate distances. Get it wrong and your thickness calibration drifts.

The chamber walls, pump ports, fixtures, and electrical feedthroughs also collect coating. Everything in line-of-sight gets coated. After 20-30 runs, millimetres of material build up on cold surfaces. The coating delaminates in sheets, falling as particulate contamination onto fresh work. Professional shops acid-strip chambers monthly. Home setups tolerate flaking until yields drop.

Monitoring a quarter-wave coating

A single-layer magnesium fluoride AR coating for 550 nm green light requires 137.5 nm physical thickness. That’s λ/4 in the material: 550 nm ÷ 4 ÷ n, where n = 1.38 (refractive index of MgF₂). Light reflected from the front surface travels 275 nm (round-trip through 137.5 nm) before interfering destructively with light reflected from the glass beneath. Perfect cancellation at 550 nm.

If you deposit 127.5 nm (10 nm too thin), minimum reflection shifts to ~510 nm (cyan). The lens looks purple in white light. Deposit 147.5 nm (10 nm too thick), minimum reflection shifts to ~590 nm (yellow). The lens looks blue. That’s ±7% tolerance for acceptable colour neutrality.

Starting from a 6 MHz QCM crystal, you need a frequency drop of 2446 Hz. The controller displays counts in real-time. Watch the number climb. When it hits 2446, close the shutter between source and substrate. Deposition stops. Coating complete.

Except: the crystal is hotter than your optic. Substrate heaters maintain 250°C to promote dense coating adhesion, but the QCM crystal can’t withstand that temperature—its resonance frequency drifts unpredictably above 100°C. So the crystal sits on a water-cooled holder, maintaining 20-30°C. MgF₂ sticks differently to cold quartz than to hot glass. The density varies. Tooling factor corrects first-order geometric differences; a density correction factor adjusts for thermal deposition conditions. You calibrate by coating test optics, measuring thickness with optical interferometry afterward, and tweaking the mass-to-thickness conversion until it matches.

What goes wrong

Crystals age. Each deposition adds mass, lowering the resonance frequency permanently. After the crystal accumulates ~1% additional mass, Q factor drops and frequency stability degrades. The crystal becomes sluggish. Most operators replace monitoring crystals every 10-15 runs, or when frequency readings start drifting without deposition.

Crystals also crack. Thermal shock during venting (going from vacuum to atmosphere dumps heat into the crystal holder), mechanical stress from thick coatings, and contamination from flaking chamber buildup all cause failures. When a crystal cracks, resonance stops abruptly. The frequency reading flatlines or jumps erratically. If this happens mid-deposition, you’ve lost thickness control. The run is scrap.

Water vapour is poison. At 10⁻⁵ torr, residual atmosphere is mostly nitrogen, with trace water, oxygen, and hydrocarbons from pump oil backstreaming. Water adsorbs onto the growing coating, adding parasitic mass that doesn’t contribute to optical thickness. The QCM reads high. Your coating ends up too thin. Baking the chamber at 150°C for several hours before deposition drives off adsorbed water. Load your optics, pump down, bake overnight, cool, deposit in the morning. Skipping the bake costs you 5-10 nm of error.

Quartz crystals are also frequency-sensitive to pressure. The damping from residual gas molecules shifts resonance slightly. At 10⁻⁵ torr it’s negligible (<1 Hz), but if your vacuum system is leaking and pressure rises during deposition, the frequency shifts for reasons unrelated to mass. You deposit more material than intended, chasing a frequency target that’s moving due to pressure drift.

The compromise nobody mentions

Magnesium fluoride has refractive index 1.38. For a single-layer AR coating on glass (n = 1.5), the optimal index is √1.5 ≈ 1.23. Nothing stable exists at that index. Cryolite (Na₃AlF₆) gets close at n = 1.35, but itNow I’ll write the deep-dive article focusing on the quartz crystal microbalance monitoring system - the counterintuitive physics of measuring mass to determine thickness.Now I’ll write the deep-dive article focused on the technical details of quartz crystal microbalance monitoring.

<<<SLEPP_ARTICLE_BODY>>> When you vaporize magnesium fluoride in a vacuum chamber and need to deposit exactly 137.5 nanometres of it onto a lens, you can’t use a ruler. Optical interference gives you one measurement method—watch the coating change colour as it grows—but that requires constant visual monitoring and works poorly for transparent dielectrics. The standard solution since the 1950s is a quartz crystal microbalance: a piezoelectric resonator that measures mass accumulation by tracking frequency drift. You’re not measuring thickness. You’re measuring inertia.

The physics is counterintuitive enough that it took me three false starts before I understood what the numbers actually meant.

Piezoelectric resonance as a weighing scale

A quartz crystal microbalance is a thin disc of crystalline quartz, typically 6-14mm diameter, with metal electrodes vacuum-deposited on both faces. Apply alternating voltage across the electrodes and the crystal oscillates mechanically at its resonant frequency—usually 5 or 6 MHz for coating monitors. The piezoelectric effect couples electrical and mechanical motion bidirectionally: voltage causes deformation, deformation generates voltage. A properly cut crystal (AT-cut, 35° from the optical axis) oscillates in thickness-shear mode, where layers of the crystal slide parallel to the faces without bending. Q factor can reach 10⁶. That’s a resonance peak sharp enough to resolve frequency changes below 1 Hz.

As atoms condense on the crystal surface during deposition, they add mass. The resonator’s effective thickness increases. Frequency drops. The relationship is nearly linear for thin films, described by the Sauerbrey equation:

Δf = -(2 f₀² Δm) / (A √(μq ρq))

where Δf is frequency shift (Hz), f₀ is the fundamental resonance (5 MHz typical), Δm is deposited mass, A is electrode area, μq is quartz shear modulus (2.947×10¹⁰ kg·m⁻¹·s⁻²), and ρq is quartz density (2648 kg/m³). The negative sign tells you that adding mass decreases frequency—the crystal is vibrating more sluggishly under load.

For a 6 MHz crystal with 1 cm² electrode area, a frequency drop of 1 Hz corresponds to roughly 1.8 nanograms of deposited material. If you’re coating magnesium fluoride (density 3180 kg/m³), that’s about 0.06 nanometres of film thickness per hertz. Ten hertz is half a nanometre. A hundred hertz is six nanometres. You’re reading mass accumulation in real time with sub-nanometre resolution, limited mostly by the frequency counter’s precision and thermal drift.

What the crystal actually feels

The Sauerbrey equation assumes the deposited film is rigid and couples perfectly to the crystal’s shear motion—it vibrates as if the coating were part of the quartz itself. This works well for metals and thin dielectrics under vacuum. The coating adds inertial mass but doesn’t dissipate energy or decouple mechanically. For optical coatings (typically <1 micrometre total), viscoelastic effects are negligible.

But here’s the catch: the crystal measures everything that changes its effective mass. Temperature variations shift the resonance by 10-30 Hz per degree Celsius for an AT-cut crystal. Gas pressure changes loading. Contamination from backstreaming pump oil or outgassing chamber walls adds mass you didn’t intend. The monitoring crystal sits inside the vacuum chamber, exposed to the same vapour flux as the substrate, so it accumulates coating at roughly the same rate—assuming it’s positioned correctly. If the crystal is off-axis or shadowed, it sees less vapour than the substrate, and your thickness reading is wrong by a constant scale factor.

That’s why thickness monitors have a “tooling factor” calibration. You deposit a test coating, measure the actual thickness optically (interferometry, profilometry, ellipsometry), compare to the QCM reading, and compute a correction multiplier. Tooling factors between 80% and 120% are typical. Beyond that range, you’re probably measuring geometry error rather than material properties.

Frequency counting at the edge of what matters

A 1 Hz shift on a 6 MHz carrier is a fractional change of 1.7×10⁻⁷. Modern frequency counters resolve this easily—gate times of 1-10 seconds, direct digital counting, no ambiguity. The challenge is distinguishing real signal (coating deposition) from noise (thermal drift, mechanical vibration, electrical interference). Mounting the crystal rigidly helps. Water cooling the crystal holder to ±0.1°C helps more. Shielding the signal leads from RF pickup matters if you’re operating near radio transmitters or induction heaters.

Some deposition systems use dual-crystal monitoring: one crystal exposed to the vapour flux, one shielded as a thermal reference. Subtract the reference frequency from the active frequency and thermal drift cancels (mostly). The same differential measurement principle as a magnetic gradiometer, where you’re nulling the common-mode signal to isolate what you care about.

Temperature coefficient for AT-cut quartz is approximately zero at 25°C, but “approximately” still means 5-10 Hz/°C if you’re 10 degrees off room temperature. For a 0.1 nm thickness resolution target, that’s ±0.6 nm of thermally induced error per degree. Water-cooled crystal mounts hold ±0.5°C easily. That’s good enough for optical coatings, where you’re targeting ±5 nm tolerances on quarter-wave layers.

The things it won’t tell you

Quartz crystal microbalance measures mass per unit area. It doesn’t measure thickness directly. You infer thickness by dividing accumulated mass by material density: d = m / (ρ × A). That requires knowing the coating density, which isn’t always the density of the bulk material. Vapour-deposited films can be porous, especially at low substrate temperatures. Magnesium fluoride deposited at 200°C typically has 95-98% of bulk density. At room temperature, maybe 85-90%. The porosity depends on deposition rate, substrate temperature, residual gas pressure, and angle of incidence. You calibrate empirically.

The QCM also doesn’t tell you anything about coating uniformity, optical quality, or adhesion. A film with poor adhesion that’s starting to delaminate will still show the correct mass accumulation—until chunks fall off. Haze from surface roughness or columnar grain structure won’t register unless it changes the film density. Index of refraction must be measured optically, usually with a witness sample coated simultaneously on a glass slide.

For multilayer coatings (three to seven layers, alternating high and low index materials), the QCM tracks cumulative mass. You watch for specific frequency targets that correspond to calculated layer thicknesses, then switch materials by opening and closing shutters over the evaporation sources. The monitoring crystal accumulates all layers in sequence. By the end of a seven-layer stack, the crystal might have dropped 5000 Hz from its initial frequency. That’s maybe 300 nm of mixed coating. After twenty runs, the crystal is unusable—too thick, too much mechanical damping, resonance becomes unstable. You swap in a fresh crystal and recalibrate.

When mass is the wrong observable

Not all thin-film processes are compatible with QCM monitoring. Reactive deposition—where titanium vapour reacts with nitrogen gas to form titanium nitride—changes stoichiometry during growth. The crystal measures titanium mass, but the final film is TiN with different density. Sputtering from compound targets (indium tin oxide, aluminium oxide) deposits multiple elements with variable sticking coefficients. Atomic layer deposition uses self-limiting surface reactions where mass per cycle is determined by chemistry, not flux. In all these cases, the Sauerbrey equation gives you an answer, but it’s not the thickness you wanted.

For optical coatings by thermal evaporation—magnesium fluoride, silicon dioxide, titanium dioxide, aluminium oxide—the process is clean. One material, ballistic vapour transport, predictable sticking. QCM works. The frequency drops smoothly at 10-50 Hz per second depending on deposition rate, you hit your target thickness (±0.5 nm if you’re paying attention), you close the shutter, and the reading stops changing. Same discipline as monitoring spectral line positions—you’re measuring a number that shifts in real time, and the skill is knowing when to stop.

The coating you just deposited is optically flat, microscopically uniform, and exactly 137.5 nanometres thick. You measured none of those things directly. You measured mass, assumed density, inferred thickness, and trusted calibration. The lens you just coated will now transmit 99.5% of incident 550 nm light instead of 96%, and you’ll never see the coating with your naked eye. It’s one to two wavelengths of material, thinner than a soap bubble, applied atom by atom in a vacuum by watching a crystal oscillate slightly slower.

That’s the part that still surprises me: precision manufacturing where the primary observable is a radio frequency.