Inverse design proposes brand-new precursors you never listed: it
combines the film's metal with a ligand library, scores every proposal, and ranks them.
For tungsten it re-derives WF₆ (flagged ★) — proof the search is sane.
The interconnect-resistance race
Below ~20 nm, copper/tungsten wiring resistance explodes from electron
scattering. The industry needs a new interconnect metal. Below: materials that already have
precursors (we screen them), and emerging ones that don't (we invent precursors for them).
DensityGen · descriptor scoring by default · opt-in Replicate UMA simulation for capped web runs
Why this matters
The race for the next chip material
Every leap in computing is unlocked by a new material. A modern chip is hundreds of
atom-thin films. The silicon is just the canvas — the deposited films do the work, and each
new generation needs a new one, laid down a single atomic layer at a time (ALD). The
bottleneck is finding the precursor molecule that delivers it.
The last time this happened, America won. In 2007 Intel replaced the transistor's
silicon-dioxide gate with a hafnium "high-k" film (HfO₂, grown by ALD) and a new metal gate.
Gordon Moore called it the biggest change in transistor design since the late 1960s. It
rescued Moore's Law and locked in a U.S. process lead for a decade. (A decade earlier, IBM's
switch from aluminum to copper wiring did the same.) Each win came down to one thing: the
right material, and a precursor to deposit it.
The next race is on now — and it's wide open. Wiring has hit a wall: below ~20 nm,
copper and tungsten interconnect resistance explodes from electron scattering, throttling
every chip. The industry needs a new interconnect metal — ruthenium, molybdenum, or exotic
topological semimetals like NbP. There is no winner yet, and most candidates have no
manufacturable precursor at all.
Whoever finds it first sets the standard. Export controls slow China at the
EUV-tool level, but materials discovery is an open front. The nation that discovers the next
interconnect material — and the precursor to make it — defines the next decade of chips that
run AI, defense, and the economy. China is pouring tens of billions (the state "Big Fund")
into closing the gap.
This is the discovery tool. Screening a precursor used to mean years of lab
trial-and-error and hours-per-candidate of supercomputer DFT. New AI atomistic models (Meta's
UMA) cut each calculation from hours to seconds — so we can screen, and even
invent, precursors for these new materials in minutes.