Aluminum Process
The aluminum process is the subtractive metallization flow that built interconnect wiring on virtually every chip made before the late 1990s, and it still runs today for bond pads, power devices, MEMS, and legacy-node analog parts. Unlike copper, which is patterned by etching trenches first and filling them (damascene), aluminum is deposited as a blanket film and then etched away everywhere it isn't wanted — a sequence that looks like this:
The alloy is doing real metallurgical work. Pure aluminum interconnect fails two different ways, so the industry never used it pure. Adding about 0.5% copper suppresses electromigration — the slow drift of aluminum atoms under current flow that otherwise opens voids and hillocks in a wire over years of operation. Adding about 1% silicon stops "aluminum spiking," where molten aluminum dissolves silicon at a contact and punches a spike straight through a shallow junction during the post-deposition anneal, shorting it to the substrate. AlCu(Si), sputtered by physical vapor deposition in a single chamber pass, became the default interconnect alloy for roughly three decades of chipmaking for exactly these reasons.
Subtractive etch is the opposite workflow from copper damascene. With aluminum, the metal goes down first as a blanket film, then photoresist defines the wire pattern, and a chlorine-based plasma (Cl2/BCl3) etches away everything the resist doesn't protect, leaving freestanding aluminum lines. Copper can't be dry-etched this way — it doesn't form a volatile halide — so 1997 forced the switch to damascene: etch the dielectric trench first, then electroplate copper into it and polish back with CMP. That single chemistry limitation is why the two metals ended up with entirely different process architectures, not just different atoms.
The alloy's reign ended for electrical, not chemical, reasons. As the industry pushed past the 0.25 µm node, line widths shrank faster than resistivity could be fixed: aluminum's bulk resistivity (~2.7 µΩ·cm) is roughly 70% higher than copper's (~1.7 µΩ·cm), and thinner, narrower lines made RC interconnect delay — not transistor switching speed — the limiter on how fast a chip could actually run. IBM's 1997 copper-damascene announcement, followed quickly by the rest of the industry, retired aluminum from the critical back-end-of-line wiring on leading logic nodes within a few process generations, even though copper itself needed a brand-new barrier (Ta/TaN) and electroplating infrastructure to get there.
Aluminum never left the fab — it just moved to the top of the stack and to other markets. Every modern chip, including copper-wired ones, still finishes with an aluminum (or AlCu) bond-pad layer, because aluminum wire-bonds cheaply and resists the galvanic and corrosion issues that would plague an exposed copper pad. It also remains the default metal for power semiconductors, analog and RF chips, MEMS devices, and any process where the extra resistance is irrelevant but the simpler single-chamber deposition, subtractive patterning, and lower cost are decisive advantages over building out a full damascene/CMP line.
Read the aluminum process as a lesson in how a materials limitation reroutes an entire architecture: the same subtractive deposit-then-etch logic that made aluminum easy to pattern for thirty years is precisely what copper's chemistry forbids, and that one incompatibility — not performance, not cost — is why modern chips carry two entirely different metallization schemes stacked on top of each other, aluminum at the pad and copper underneath.