Diluted 49% hydrofluoric acid vs buffered oxide etch: the buffer is a second purity risk

A process engineer adds buffered oxide etch to a resist-masked oxide step, and your purchase list grows from one fluoride chemical to two: 49% hydrofluoric acid plus 40% ammonium fluoride, or a premixed blend with its own specification.

The engineer has half a point. A buffered bath holds its etch rate as it works, and photoresist adheres much better in it than in concentrated HF. HF diluted past 3:1 holds resist too, so rate stability is what the second chemical buys. The cost is easier to miss: the buffer is a second chemical whose purity you have to control. Nor does the blend spare aluminum: like HF itself, it attacks the metal.

Which 49% hydrofluoric acid etching agent fits: diluted HF or buffered oxide etch?

Hydrofluoric acid diluted with water from 49% stock handles oxide strips and HF dips, and once diluted past 3:1 it no longer lifts photoresist. Buffered oxide etch (BOE) adds ammonium fluoride so the etch rate holds steady as the bath works, which suits resist-masked patterning. Like HF itself, it attacks aluminum, and it brings a second purity specification.

Both start from the same acid. EPA’s 1993 Hydrogen Fluoride Study found that the ultra-high-purity HF used in chip making is typically purchased at its most concentrated form of 49 percent and used at 1 to 10 percent. The same study notes that electronic grade HF is usually 49 percent aqueous HF; for how that strength compares with 70% acid, see hydrofluoric acid grades and concentrations.

Commercial BOE ratios such as 6:1 or 20:1 are normally stated as the volume of 40 wt% ammonium fluoride solution to 49 wt% HF, according to a 2011 national-laboratory etching study.

Diluted HF (DHF) ratios count water instead: 10:1 HF is ten parts water to one part 49% acid by volume. A 1996 etch-rate survey describes 25:1 as a slow etch for dips that strip native oxide without removing much of the other oxide on the wafer, and 10:1 as the mix for oxide stripping and dips, cheaper than buffered HF. It adds that 49% acid tends to peel photoresist, while mixes weaker than 3:1 do not.

Why does the buffer hold the rate?

Both baths etch through molecular HF and the bifluoride ion (HF₂⁻), and the 1996 survey puts bifluoride’s attack on silicon dioxide at about 4.5 times faster. As a bath works it consumes both and loads with etch by-products, among them fluosilicic acid, a strong acid that shifts the pH and with it the etch rate, as patent US5722442A (1998) explains. Ammonium fluoride holds the pH, and with it the HF and bifluoride concentrations.

That is why the 1996 survey finds that 5:1 buffered HF changes rate less with use than unbuffered HF, and calls it often the best choice for controlled oxide etching. Diluted HF is predictable in another way: a 2003 follow-up survey found the thermal-oxide etch rate linear with concentration from 10:1 to 100:1, and measured 10:1 buffered HF at exactly half the rate of 5:1.

Buffering does not mean slower. In Table 1 of patent US6048406A (2000), room-temperature PETEOS oxide etched at 30 Å/min in 0.49% HF against 75 Å/min in the patent’s 0.49% buffered HF, and at 330 against 800 Å/min at 4.9%. That fits a 2012 glass-microfabrication guide, which prefers ammonium fluoride to water as the diluent because the buffer stabilizes and raises the bifluoride content. Patent figures in this post are published data on specific films, not a recommendation to practice them.

Diluted HF and buffered oxide etch side by side

Where does diluted HF do the job on its own?

Wherever the step is a strip or a dip and the rate need not hold through a long patterned etch. After a piranha clean, the 1996 survey describes removing the thin hydrous oxide left behind with a 10-second 10:1 or 25:1 HF dip. Dilute HF can be selective, too: in a 2023 study, LPCVD silicon dioxide etched about 50 times faster than LPCVD silicon nitride in 1:100 HF to water at room temperature.

On metal contamination, diluted HF is a partial tool. Patent US5759971A (1998) states that a DHF clean removes metals held in the oxide film but not copper adsorbed directly on silicon, and leaves a water-repellent surface that picks up particles. A 1993 study found that removing iron from silicon took at least three cleaning steps plus rinses, one of them an HF solution.

HF dissolves most oxides, and that cuts both ways for exposed metals. A materials handbook notes that any alloy relying on a passive oxide film performs poorly in HF. The 2003 survey measured titanium etching faster than silicon dioxide in 10:1 HF and recommends Cr or titanium nitride over titanium as an adhesion layer where HF is present. Aluminum needs protection in diluted HF as well, for the reasons set out under the buffer’s costs below.

The purity input you add to DHF is the water. Patent US6620743B2 (2003) makes its dilute HF from ultrapure water above about 15 megaohm-cm at 25 °C, because dilution water carries metals into the bath. Glass etching and frosting are a separate purchase with their own bath chemistry; see hydrofluoric acid for glass etching and frosting.

What does the buffer cost?

Aluminum comes first, and it is no reason to switch to diluted HF. A 2020 study of wet oxide etching next to aluminum states that BOE etches aluminum, so aluminum structures must be protected. HF itself attacks the metal as well: 49 CFR special provision N34 bars aluminum from any part of a non-bulk HF packaging that touches the contents, and the 2003 survey found aluminum attacked in its HF vapor tests.

Other metals are less of a worry on either route. The 2003 survey found that 5:1 buffered HF and 10:1 HF did not appreciably etch most of the transition metals it tested, with no measured etch of Cr, Au or Pt films. Titanium is the exception: the survey measured it etching faster than silicon dioxide in 10:1 HF, and a materials handbook lists it among the metals readily attacked by HF.

Purity comes second. Patent US5722442A states that with ultrapure chemicals the buffering agent is itself a contamination source and must be equally pure. Copper shows the stakes. In a 2001 spectroscopic study, hydrogen-terminated silicon dipped for five minutes in 0.5% HF carrying copper ions picked up about one-tenth the copper it took from 40% ammonium fluoride at the same copper level.

Bath electrochemistry will not flag it. In a 1998 comparison, silicon corrosion current in buffered HF changed little with copper up to 1000 ppb while the copper deposited on the wafer rose with concentration; in dilute HF, copper also shifted the corrosion current and potential. On a buffered line, look for copper on the wafer surface, with a method such as TXRF.

Preparation comes third. Freshly mixed buffered HF should stand for several hours, the 2003 survey says, because its rate starts much higher than at equilibrium. Ratios also need a stated basis: the 1996 survey defines 5:1 buffered HF by weight (about 33% ammonium fluoride and 8.3% HF), while the 2003 survey and the 2011 study state ratios by volume.

Nitride is the last check. In 5:1 buffered HF, the fastest-etching silicon nitride in the 2003 survey etched 46 times faster than the slowest, against 1.3 in a plasma etch, so qualify the bath on each nitride film you expose.

Which purity standard covers each?

SEMI keeps the chemicals apart. SEMI C28 covers hydrofluoric acid grades used in semiconductor work; its current revision, C28-0618, carries Inactive status but remains valid for use. SEMI C29 covers 4.9% HF designated 10:1 by volume. SEMI C20 covers 40% ammonium fluoride, also Inactive and valid for use, and SEMI C23, a current standard last technically revised as C23-0714, covers buffered oxide etchants themselves. SEMI C2, the etchant specification that once defined buffer oxide etchants, was withdrawn in 2008, so a sheet that still cites it is out of date.

The route sets the paperwork. Dilute in the fab, and one acid plus your water carries the purity budget, with C28 as the reference. Buy BOE premixed, and the reference is C23. Mix BOE yourself, and the references are C28 for the acid and C20 for the ammonium fluoride: two lots, two certificates and two contamination paths into one bath.

The SEMI C1 guide states that a grade validated with the appropriate procedures can be described as meeting SEMI Specifications, with procedures defined for each grade so their sensitivity meets its requirements. When a sheet says semiconductor grade without naming a SEMI standard and grade, judge it on its printed limits and the lot certificate, not the label.

The technical data sheet (PDF) for semiconductor-grade 49% HF is one to read that way. It prints limits for three grades of its own and names no SEMI grade: HF of 48.8 to 49.2 wt% for all three; trace metals, copper and aluminum among them, in ppb at no more than 1, 0.1 and 0.01 per element for grades 1, 2 and 3; and fluosilicic acid (H₂SiF₆), the by-product a working bath builds, in ppm at no more than 5, 1 and 1.

The sheet prints limits only, with no typical values or test methods. The binding figures are those on the lot Certificate of Analysis (CoA) and in the sales contract, and packaging is 235 kg net new poly drums. The IRDS roadmap adds that variation in process-fluid purity can matter as much to wafer yield as absolute purity, so one good certificate is a starting point, not a qualification.

What goes on the purchase order?

1. The etchant as a recipe: DHF ratio by volume from 49% acid, or BOE ratio with its basis (volume or weight).

2. The SEMI standard and grade you expect for each chemical, or the supplier’s own limits, judged on the CoA.

3. On a BOE route, an ammonium fluoride CoA reporting the same trace metals as the HF CoA, copper included.

4. A map of exposed metals: aluminum and titanium need protection on either route. The 2003 survey measured titanium attacked in 10:1 HF and recommends Cr or titanium nitride adhesion layers over titanium wherever HF is present.

5. CoA results across several lots, not one.

The grade page’s specification table and the data sheet download carry the printed limits; the lot CoA and the contract carry the binding ones.

Methodology: figures come from EPA’s 1993 Hydrogen Fluoride Study, two IEEE microfabrication etch-rate surveys (1996 and 2003), a 2011 national-laboratory etching study, peer-reviewed studies from 1993 to 2023, four patents (US5722442A, US6048406A, US5759971A and US6620743B2), SEMI public abstracts (C1, C2, C10, C20, C23, C28, C29), an HF materials handbook, 49 CFR Parts 172 and 173 (eCFR, current to 2026-10-05), the IRDS roadmap and the grade’s technical data sheet (revision 2026-09-30). Statements without a figure are standard etch chemistry.