Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

IBM's Project SWIFT Ran a Layer in 5 Hours. Modern Fabs Take 19 to 36

Bill Harding's Project SWIFT averaged 5 hours per fabrication layer in the 1970s, IEEE Spectrum reports. The fastest modern fabs take 19 hours, and the industry average is 36.

TechnologyAnalysisRachel NwosuPublished: 27 September 20265 min readSources 6
IBM's Project SWIFT Ran a Layer in 5 Hours. Modern Fabs Take 19 to 36

IBM's Project SWIFT averaged 5 hours to complete each layer of its fabrication process, IEEE Spectrum reported in December 2024. The fastest modern fabs take 19 hours per processing layer. The industry average is 36.

Jesse Aronstein wrote the long history piece. He is an engineer who worked on the programme. In 1970, Bill Harding pitched a fully automated wafer-fabrication line. Harding then managed IBM's Manufacturing Research group in East Fishkill, New York. The line would turn out integrated circuits in less than one day. At the time, IEEE Spectrum writes, ICs such as random-access memory chips typically took a monthlong stop-and-go march through dozens of manual work stations.

Harding was not a typical IBM manager. He grew up in Brooklyn and was wounded three times in combat in World War II. He served in George S. Patton's Third Army. He later earned bachelor's and master's degrees in mathematics and physics, according to the account. Aronstein first met him in 1962. IBM was then gearing up to produce the System/360, its first completely solid-state computer. The encounter was rough. Harding bellowed "What the hell good is that?" at Aronstein. Aronstein was demonstrating how unpackaged semiconductor dice could be handled automatically in bulk for testing and sorting.

The article states the caveats in the comparison to today plainly. Modern ICs are built with many more layers, on larger wafers, and the processing is more complex. IEEE Spectrum adds that those factors do not altogether close the gap. SWIFT's turnaround time, it says, has never been equaled.

What the rest of the supply chain is doing

The capacity question is not only about line speed. Governments and suppliers are spending on volume, materials and upstream inputs at the same time.

Malaysia is pushing into higher-end semiconductor manufacturing. Second finance minister Amir Hamzah Azizan told CNBC the country is unfazed by competition from China or the US. He spoke at the IMF spring meetings in Washington in April 2024. "The semiconductor space is now in the upcycle, and Malaysia will be a beneficiary of that," he said. Malaysia holds 13% of the global market for chip packaging, assembly and testing services. CNBC cited a Malaysian Investment Development Authority report in February for that figure.

"We're seeing a lot of end users now diversifying their supply chain. Our focus, actually, is to provide a very vibrant, strong supply chain connectivity, and make sure that we ride on that."

Amir Hamzah also said Malaysia would not go "head on to the tail end of the high-end competition, where maybe the US is bringing all the parts."

India is taking a different route. TechCrunch reported on 15 July 2026 that New Delhi unveiled a ₹625 billion (about $6.5 billion) Mobile Phone Manufacturing Scheme. It runs five years, with incentives of 2.25% to 5% on eligible sales and an extra 1.5% for sourcing key components and sub-assemblies locally. The same announcement committed a further ₹1.28 trillion (around $13.3 billion) to domestic semiconductor manufacturing. That expands a $10 billion chip incentive programme launched in 2021. India accounted for 18% of global smartphone production in 2025, against China's 63%, according to Counterpoint Research.

On the component side, Texas Instruments said on 24 October 2024 that it had started producing gallium nitride power semiconductors at its factory in Aizu, Japan. The move quadrupled internal GaN capacity across its US and Japanese sites. Mohammad Yunus, TI's senior vice president of technology and manufacturing, said the company aims to grow internal manufacturing to more than 95% by 2030. TI also said it had piloted GaN processes on 300mm wafers.

The input nobody can substitute quickly

Then there is quartz. Spruce Pine, North Carolina, is home to high-purity quartz mines. Sibelco North America and The Quartz Corp operate them. Tom's Hardware reported on 1 October 2024 that Hurricane Helene brought over two feet of rain to the town. The storm cut off roads and left residents without power. Sibelco had confirmed a suspension of operations. The Quartz Corp told The Verge it was too early to say whether its production facility was affected.

Construction Physics examined the claim that all semiconductor manufacturing depends on Spruce Pine. Its November 2024 assessment: quartz crucibles are indeed necessary for most semiconductor manufacturing, and Spruce Pine is where most of the quartz comes from. But the town is not quite an irreplaceable linchpin. Cutting off supply probably would not choke off semiconductors completely. It would mean yields going down and costs going up. New sources of quartz are being developed and new crucible materials investigated.

That piece has details worth keeping. High-purity quartz costs on the order of $10,000 a ton. Crucibles cost several thousand dollars apiece. A crucible is limited to around 400 hours of operation. A Czochralski furnace consumes 20 to 25 crucibles a year. One paper cited by Construction Physics estimated that 30% of the manufacturing cost of monocrystalline silicon ingots came from the crucibles.

Put the numbers together and the picture is less dramatic than the headlines that followed Helene. Spruce Pine is estimated to produce 180,000 to 200,000 tons of high-purity quartz per year. That is up from roughly 30,000 tons of high-purity output in 2015. That is a lot of material for an industry this size, and very little next to a coal mine.

The harder question is the one SWIFT raises. Automation of the kind Harding built has spread across the industry, IEEE Spectrum notes. Many of his line's innovations are now commonplace. The turnaround time is the part that did not travel.

Comments 0

Sources

6
  1. 01IBM invented semiconductor manufacturing automationEN
  2. 02Malaysia pushes into higher-end semiconductor manufacturingEN
  3. 03Hurricane devastates quartz mines critical for semiconductor manufacturingEN
  4. 04India to allocate $20B to semiconductor and smartphone manufacturingEN
  5. 05TI expands internal manufacturing for gallium nitride (GaN) semiconductorsEN
  6. 06Does All Semiconductor Manufacturing Depend on Spruce Pine Quartz?EN

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Rachel Nwosu

Rachel Nwosu

AI, models and technology

Rachel Nwosu covers AI, models and technology for FLASH24, working from public model documentation, benchmark releases and repository histories rather than press summaries, and she skips announcements that arrive without reproducible numbers. She checks training-data claims against dataset cards and reruns reported metrics where code is available. She spends much of her week interviewing researchers and engineers, tracking model launch calendars, and comparing vendor benchmarks with independent evaluations. Outside the desk she runs 3D printers, restores old computers, and tests how models learn from internet junk. She does not publish benchmark figures she cannot trace to a source.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.