Hydrofluoric Acid: The Weak Acid That Eats Glass

Weak Describes Proton Transfer, Not Damage

Acid strength has exactly one meaning: where the equilibrium HF + H₂O ⇌ H₃O⁺ + F⁻ sits. Hydrofluoric acid sits well to the left. Its dissociation constant is Ka = 6.8 × 10⁻⁴ at 25 °C, pKa 3.17 — some nine orders of magnitude short of hydrochloric acid, whose pKa is near −6. It is weak in the same technical sense as acetic acid (Ka = 1.8 × 10⁻⁵), and only about forty times stronger.

Worked example. Let x be the equilibrium concentration of H₃O⁺ in 0.10 M HF. Then x²/(0.10 − x) = 6.8 × 10⁻⁴. Solving the quadratic gives x = 7.9 × 10⁻³ M, so 0.10 M HF is only 7.9% ionised and its pH is 2.1. The same 0.10 M of hydrochloric acid ionises completely — pH 1.0, and 12.6 times more free protons. By pH, by conductivity, by how fast it fizzes on magnesium, HF is the milder reagent.

And yet it is the only common bench acid that dissolves silica. Aqua regia, which eats gold, leaves a glass beaker untouched; HF hollows one out — and the reagent doing it is the 92% of molecules that never ionised.

Entropy, Not the Bond, Keeps the Proton

The usual explanation is the H–F bond: 569 kJ/mol against 431 kJ/mol for H–Cl, the strongest bond any hydrogen halide has. But bond dissociation energy is a gas-phase homolysis number, and ionisation in water is neither. Measure the heat of ionisation directly and it comes out negative: ΔH° ≈ −13 kJ/mol. Splitting HF in water releases energy. Something else stops it.

Worked example. ΔG° = −RT ln Ka = −(8.314 J K⁻¹ mol⁻¹)(298.15 K) × ln(6.8 × 10⁻⁴) = +18.1 kJ/mol. Since ΔG° = ΔH° − TΔS°, the entropy term must be TΔS° = ΔH° − ΔG° = (−13) − (+18.1) = −31.1 kJ/mol, giving ΔS° ≈ −105 J K⁻¹ mol⁻¹. That +31 kJ/mol entropy penalty is more than twice the enthalpy gain: HF is weak because water objects to the products, not because the bond is hard to break.

The products are costly because fluoride is tiny and hard — ionic radius 133 pm against 181 pm for chloride — so its hydration enthalpy reaches about −510 kJ/mol where chloride manages only −367. That energy is bought by freezing a shell of water around the ion, and immobilised water is lost entropy. Worse, the proton never gets away: HF ionises into a contact ion pair, H₃O⁺···F⁻, that hands the proton back before diffusion can separate the charges.

One twist catches people out: above roughly 5 M, HF gets stronger, because the reaction in the next section consumes free fluoride and drags the equilibrium forward. Anhydrous HF reaches a Hammett acidity of H₀ = −15.1, beyond 100% sulfuric acid at −12.

Bifluoride: The Strongest Hydrogen Bond Known

Whatever fluoride does escape immediately finds another HF molecule: HF + F⁻ ⇌ HF₂⁻, association constant about 3.9 M⁻¹ at 25 °C. The product, bifluoride, is a structural outlier — linear and symmetric, F···F 2.27 Å, with the proton sitting in a single broad potential well at the midpoint rather than belonging to either fluorine. Its binding energy is about 163 kJ/mol, the strongest hydrogen bond known: eight times an ordinary water–water hydrogen bond at roughly 20 kJ/mol, and squarely in the range chemists reserve for covalent bonds. It is better described as a three-centre, four-electron bond.

Bifluoride is not a curiosity but the working species. It carries the charge in anhydrous HF, which is why Henri Moissan's 1886 isolation of elemental fluorine electrolysed potassium bifluoride dissolved in anhydrous HF at −50 °C — pure HF barely conducts. In etching, HF₂⁻ is a pre-assembled proton-plus-nucleophile package, precisely what attacking silica requires.

The Etch, Step by Step

Silica is a network of Si–O–Si bridges, each worth 452 kJ/mol, with no weak link. Breaking one needs two things at once, and HF is unusual in supplying both.

• Protonate the bridge. A proton — from H₃O⁺, or handed over directly by an HF molecule hydrogen-bonded to the surface — lands on a bridging oxygen, so that when the bridge breaks the oxygen leaves as a neutral silanol instead of a siloxide anion.
• Attack the silicon. Fluoride, or the fluorine end of HF₂⁻, adds to silicon. As a third-row element with room around it, silicon passes through a five-coordinate, trigonal-bipyramidal transition state — a route carbon cannot take. Bifluoride is the efficient attacker: it brings nucleophile and proton in one encounter.
• Break the bridge. One Si–O bond goes, leaving a surface Si–F on one silicon and an Si–OH on the other; that oxygen only leaves as water when a further HF converts the silanol to Si–F.
• Acceleration in the coordination sphere. Each Si–F formed is worth 565 kJ/mol against the 452 kJ/mol of the Si–O it replaced, and fluorine's inductive pull drains the silicon, making it more electrophilic. The second, third and fourth substitutions are faster than the first: SiF, SiF₂, SiF₃, SiF₄.
• Let it leave. SiF₄ is a small molecular gas with no bond left to the lattice — it sublimes at −86 °C, so it simply bubbles off. In excess HF it is scavenged instead: SiF₄ + 2 HF → H₂SiF₆, hexafluorosilicic acid, which keeps the bath clear.

Overall: SiO₂ + 4 HF → SiF₄↑ + 2 H₂O, or SiO₂ + 6 HF → H₂SiF₆ + 2 H₂O. J. S. Judge's 1971 kinetics (J. Electrochem. Soc. 118, 1772) pinned the thermal-oxide rate law as two parallel attacks, rate = a[HF] + b[HF₂⁻], with bifluoride roughly four to five times more effective per mole than HF and bare F⁻ contributing essentially nothing. That last clause is why strong acids fail: they supply the proton but no nucleophile that can build a bond stronger than Si–O.

Hardware, Standards and How the Etch Rate Is Specified

Concentrated HF is sold at 49 wt%, roughly 28 mol/L of total fluorine. On thermal silicon dioxide it removes about 2 µm per minute — far too fast for a 100 nm film, and aggressive enough to lift photoresist off the wafer, so fabs buffer it. Buffered oxide etch (BOE) at 6:1 is six volumes of 40% ammonium fluoride to one of 49% HF, and takes thermal oxide at roughly 100 nm/min at 25 °C. The ammonium fluoride is a fluoride reservoir: as HF is consumed, F⁻ + HF → HF₂⁻ holds bifluoride and pH nearly constant, so the rate does not drift over the bath's life and the resist survives.

Worked example — why the bath is thermostatted. The etch has an activation energy near 35 kJ/mol, so the rate ratio between 25 °C and 35 °C is exp[(35 000/8.314)(1/298 − 1/308)] = exp(0.46) = 1.6. Ten degrees makes the etch 60% faster, and one degree of drift shifts it by nearly 5% — which is why a BOE bath is held to ±0.5 °C.

Rates are measured on monitor wafers: a blanket thermal oxide is gauged by ellipsometry before and after a timed dip. Selectivity matters as much as rate — 6:1 BOE takes silicon nitride at well under 1 nm/min and silicon essentially not at all. Fluoride in solution is read by a lanthanum fluoride ion-selective electrode, Nernstian at −59.2 mV per decade, in TISAB buffer holding pH 5.0–5.5; below pH 5 the fluoride reverts to HF and the electrode under-reads.

Everything that touches HF is chosen for having nothing HF wants: HDPE, PTFE and PFA, never glass and never a glass thermometer in the same hood, because the vapour frosts any silicate nearby. Electronic-grade HF is bought to SEMI's C-series specifications, whose highest purity grades cap each trace metal at the parts-per-trillion level. OSHA's exposure limit is 3 ppm over 8 hours, ACGIH recommends 0.5 ppm with a 2 ppm ceiling, and 30 ppm is immediately dangerous to life and health.

How It Injures, and the Incidents That Made the Rules

The same neutrality that makes HF a weak acid makes it uniquely dangerous. An undissociated HF molecule is small and uncharged, so it crosses skin and cell membranes that stop H₃O⁺ and Cl⁻ cold, then dissociates in the tissue beneath, where the liberated fluoride precipitates calcium and magnesium as insoluble salts. Stripping the ions muscle, nerve and enzyme chemistry depend on gives deep liquefactive necrosis working inward from an unremarkable surface, plus hypocalcaemia and hyperkalaemia that can stop the heart. Concentrated HF hurts at once, but below about 20% the pain can be delayed up to 24 hours, which is why dilute splashes are the ones that kill — nobody rinses a burn they cannot feel. The response is prolonged rinsing followed immediately by 2.5% calcium gluconate gel, with subcutaneous, intra-arterial or nebulised calcium for deeper injury and continuous ECG and electrolyte monitoring.

At industrial scale: Gumi, South Korea, 27 September 2012, where a hose failed as a tanker was unloaded at the Hube Global plant and released about 8 tonnes of hydrogen fluoride. Five workers died, thousands of residents were treated, crops and livestock in the valley were destroyed and the area became a special disaster zone. On 21 June 2019 the Philadelphia Energy Solutions refinery's HF alkylation unit ruptured at a corroded elbow; roughly 5,000 lb of HF escaped, and only rapid de-inventory and water curtains prevented far worse.

The history is equally pointed. The Nuremberg glass cutter Heinrich Schwanhard was etching glass with fluorspar and acid around 1670, long before anyone knew what the gas was; Scheele made it deliberately in 1771 by heating fluorspar with sulfuric acid, in a glass retort, which corroded. The chemists who then tried to wring elemental fluorine out of it were poisoned — Paulin Louyet and Jérôme Nicklès fatally, the Knox brothers permanently.

Look-Alikes and the Confusions Worth Killing

Corrosive is not strong. Strength is a thermodynamic statement about proton transfer to water; corrosivity is what the whole molecule does to a given material. HF is weak and devastating; perchloric acid is far stronger and will not touch a beaker.

It is not the fluoride ion doing the work. The usual wrong model is that F⁻ is such a good nucleophile it tears silica apart. Bare fluoride is nearly inert here: neutral sodium fluoride solution leaves glass essentially alone, because without acid there is no HF, almost no HF₂⁻ and no protonated bridge.

Frosted glass is not HF's doing. Pure HF etches cleanly and, used properly, polishes glass to optical clarity — it is the reagent for acid-polishing cut crystal. Frosting takes ammonium bifluoride, which on soda-lime glass leaves a crust of insoluble sodium fluorosilicate, Na₂SiF₆, whose crystals scatter light.

HF does not etch silicon. It dissolves silicon dioxide; bare silicon is almost untouched, which is exactly what makes a dilute HF dip useful. Removing silicon needs an oxidant alongside — the HNA poly-etch of hydrofluoric, nitric and acetic acids, where nitric acid makes the oxide and HF carries it away.

It is not a superacid, nor its fluorine relatives. Aqueous HF is nowhere near fluoroantimonic acid, HF plus SbF₅, at H₀ near −28 — nor is it fluorine gas, chlorine trifluoride, or the hexafluorosilicic acid dosed into drinking water, that last a downstream product of this very reaction.

Only fluoride chemistry opens the Si–O–Si bridge at room temperature

Reagent	Acid strength	Attacks silica?	Why
49% HF (aqueous)	Weak — pKa 3.17	Yes, ~2 µm/min on thermal oxide	Undissociated HF and HF₂⁻ deliver a proton and a fluoride in the same encounter; Si–F (565 kJ/mol) beats Si–O (452 kJ/mol)
Concentrated HCl	Strong — pKa ≈ −6	No	Protons in abundance, but Cl⁻ is a poor nucleophile at silicon and Si–Cl is weaker than Si–O, so the protonated bridge simply reverts
1 M NaF at pH 8	Not acidic — F⁻ is a weak base	Essentially no	Bare F⁻ is nearly inert: with no acid there is almost no HF and no HF₂⁻, so the bridging oxygen is never protonated
Ammonium bifluoride paste	Weak, HF₂⁻-rich	Yes — and it frosts the surface	Etches, then leaves insoluble Na₂SiF₆ crystals on soda-lime glass that scatter light
85% H₃PO₄ at 160 °C	Weak — pKa₁ 2.15	Barely, ~0.2 nm/min	Slow hydrolysis, no fluoride chemistry — which is exactly why it strips Si₃N₄ selectively over SiO₂
Hot concentrated NaOH	Strong base	Yes, but slowly	Hydroxide opens the bridge to soluble silicate, but forms no bond to silicon stronger than Si–O, so it needs heat