How a zircon by-product ended up in the transistor

Hafnium reached the centre of the semiconductor industry because silicon dioxide ran out of room — and the mineral it comes from is the same one the tile industry buys.

Reviewed by Peter Uppal

A mirror-bright patterned wafer standing angled in a slotted white carrier on a pale bench, throwing broad blue and gold diffraction across its face. — Illustrative artwork: a patterned silicon wafer of the kind produced in semiconductor manufacturing. Not a facility, equipment or material connected to this project. · Illustration · Osmond Hub

The short version

For decades the insulating layer under a transistor's gate was silicon dioxide, and every generation made it thinner. It stopped working: an Intel author wrote that silicon dioxide 'is running out of atoms for further scaling'(opens in a new tab), and a review puts the physics as gate leakage 'due to direct tunnelling of electrons through the SiO2' exceeding 1 A/cm² at 1 V once the layer is around 1.2 nm(opens in a new tab). The replacement was a hafnium compound, and hafnium comes from zircon — the same mineral that ends up in floor tiles.

The layer that could not get any thinner

A field-effect transistor works by holding a voltage on a gate and letting the resulting field control current in the channel beneath. Between the two sits an insulator, and thinner insulator means stronger control, which for decades meant faster chips. Intel's own description of where that ended up is that in modern transistors 'the gate thickness is about five atomic layers'(opens in a new tab) — undated, on a page the company does not date, and quoted with that caveat.

At that scale the insulator stops being reliably insulating. Electrons tunnel straight through, and the leakage is not a rounding error: the Robertson and Wallace review gives it as exceeding 1 A/cm² at 1 V(opens in a new tab) for a layer around 1.2 nanometres. An undated paper by an Intel author records the same thickness in production, attributing it to an earlier generation: silicon dioxide with a physical thickness of 1.2 nm 'has been successfully implemented in the 90nm logic node'(opens in a new tab), and, in the same document, that the gate oxide leakage is increasing with decreasing SiO2 thickness(opens in a new tab).

The way out was to stop thinning and start substituting: a material with a higher dielectric constant gives the same electrical effect at greater physical thickness. Intel states the baseline it was measured against — silicon dioxide 'has a "k" of 3.9'(opens in a new tab) — and names hafnium oxide among the materials with values above it. This article does not give a numeric constant for hafnium oxide, because no standards-body figure was obtained.

The date it stopped being a laboratory result

On 25 October 2007 Intel announced that production of a new generation of microprocessors 'officially began today' inside its first high-volume 45 nanometre manufacturing factory(opens in a new tab), and stated in the same release that the 45 nm transistors 'use a Hafnium-based high-k material for the gate dielectric and metal materials for the gate'(opens in a new tab). That is a dated, first-party statement of when a hafnium compound entered volume semiconductor manufacturing, and it is the anchor the rest of this rests on.

Every mineral has a moment when a laboratory curiosity becomes a purchase order. Hafnium's is on the record, with a date on it.

Where the hafnium comes from — and what nobody publishes

Hafnium is not mined. The USGS states that zircon 'was also the primary source of hafnium'(opens in a new tab), which puts the semiconductor industry downstream of a heavy-mineral-sands product.

How much hafnium travels with the zirconium is a place where the official record disagrees with itself, and the disagreement is instructive for anyone checking figures. The Mineral Commodity Summaries state that zirconium and hafnium 'are typically contained in zircon at a ratio of about 50 to 1'(opens in a new tab). The Minerals Yearbook, citing a specific study, gives a ratio of about 34 to 1(opens in a new tab). Both are USGS publications; both hedge with 'about'; neither is corrected by the other in the documents read. They are reproduced here rather than averaged.

The larger silence is on production. USGS states flatly that world primary hafnium production data and quantitative estimates of hafnium reserves 'were not available'(opens in a new tab), and separately that quantitative estimates of hafnium resources 'were not available'(opens in a new tab). The most widely cited public statistical source on minerals declines to say how much hafnium the world makes. Any confident tonnage for hafnium is therefore coming from somewhere else, and the sourcing is worth asking about.

Nor does the semiconductor use dominate the official picture of what hafnium is for. USGS lists the principal uses as 'high-temperature ceramics, nickel-base superalloys, nozzles for plasma-arc metal cutting, and nuclear control rods'(opens in a new tab) and states elsewhere that the leading use of hafnium metal is in superalloys(opens in a new tab). Gate dielectrics do not appear in that list — a reminder that a use can be strategically enormous and statistically small at once. The mass of hafnium in a chip is not a figure any source read here provides.

The mineral end

Osmond reports hafnium the way it occurs: as an oxide fraction inside zircon-bearing rock. In the Zone 1 bulk channel samples it reports HfO₂ at 1,178–1,204 ppm alongside ZrO₂ at 5.07–5.57% and zircon at 8.77–9.79%(opens in a new tab), mass-balance estimates rather than direct measurement. The framing that hafnium rides with zircon is this publication's, not the company's; Osmond reports the parts per million and characterises nothing further.

Exploration results and mineralogical estimates only. Orión has no JORC-compliant Mineral Resource or Reserve; maiden MRE and Scoping Study pending, targeted Q3 CY26.

Related