IBM's 1970s Automated Fab Hit One Day Per Chip. Modern Fabs Take Weeks

That gap is the most concrete thing in the retelling of Project SWIFT, published by IEEE Spectrum in its December 2024 issue. The industry average is 36 hours per processing layer, the article states, against SWIFT's 5. Modern chips use more layers and larger wafers, and the processing is more complex. IEEE Spectrum argues those factors do not close the gap.

Harding ran IBM's Manufacturing Research group in East Fishkill, N.Y. He had been at IBM for three years when the Components Division formed in 1961, and he took charge of the equipment needed to build the System/360's solid-state devices.

The project he led, SWIFT, was built around sectors. Each enclosure held all the wafer-processing equipment needed for one segment of fabrication between lithographic pattern exposures, according to a 1973 patent diagram reproduced by IEEE Spectrum.

What the timeline actually shows

The comparison is not clean, and IEEE Spectrum says so. Today's fabs work on wafers the size of small pizzas and run far more layers. What SWIFT demonstrated was a design principle, not a like-for-like benchmark. It automated the movement and grouping of process steps instead of stacking manual work stations in a monthlong stop-and-go march.

Back then, memory chips moved through dozens of manual stations over roughly a month. Whether that principle can be pushed further today is a separate question, and the article does not answer it. It notes only that many of SWIFT's innovations are now commonplace, while its turnaround time has never been equaled.

The modern economy rests on a single road in Spruce Pine, North Carolina.

That line, quoted by Construction Physics from economist Ethan Mollick, points at the other half of the capacity story: materials. Spruce Pine is where two companies, Sibelco subsidiary Covia and The Quartz Corp, mine the high-purity quartz used to make crucibles for the Czochralski process. In that process, polysilicon is melted above 1,400 degrees Celsius and drawn into single-crystal ingots.

Construction Physics estimates Spruce Pine produces 180,000 to 200,000 tons of high-purity quartz a year, and cites a 2013 figure of 80,000 tons worldwide across all quartz grades. High-purity quartz costs on the order of $10,000 a ton. Crucibles last around 400 hours of operation, and a furnace consumes 20 to 25 a year.

One paper cited by Construction Physics put crucibles at 30% of the manufacturing cost of monocrystalline silicon ingots.

The bottleneck is real, the apocalypse is not

Construction Physics is careful here. Spruce Pine quartz is not an irreplaceable linchpin. Alternatives are not yet developed, not quite as good, and not as cheap. Cutting the supply would not stop semiconductors entirely, but yields would fall and costs would rise. New sources and new crucible materials are under investigation.

Hurricane Helene gave the argument a live test in late 2024. Tom's Hardware reported on 1 October that the storm dropped over two feet of rain on Spruce Pine, cut off roads and left residents without power. Sibelco had confirmed a suspension of operations, the report said. The Quartz Corp told The Verge it was too early to determine whether its production facility was affected. Repeated attempts to reach Sibelco failed because phone service was down.

Capacity, then, is not only a question of how fast a line runs or how much money a government commits.

India's cabinet approved about $6.5 billion for smartphone manufacturing and roughly $13.3 billion more for semiconductors in July 2026, TechCrunch reported. China still held 63% of global smartphone production in 2025 against India's 18%, per Counterpoint Research.

Texas Instruments said in October 2024 that production at its Aizu, Japan, GaN fab would quadruple its internal GaN capacity. Each of those numbers depends, at some point, on a crucible.