The Manufacturing Leap
The transistor. The integrated circuit. The planar process. Three inventions, each independently worthy of reshaping civilization, arrived within a single decade, birthed from the same unlikely womb: a telephone company's research laboratory in New Jersey.
The imperative was physical. Vacuum tubes, essentially glorified light bulbs controlling electron flow, were "big, clumsy, used a lot of power, generated large amounts of heat, and were fragile". AT&T's telephone network needed amplification for transcontinental calls, and tubes couldn't handle the ultrahigh frequencies required. Bell Labs director Mervin Kelly began recruiting solid-state physicists in 1936. Among the first was William Shockley, a brilliant and volatile physicist who proposed amplifier designs based on copper-oxide semiconductors but failed to produce a working prototype in 1939.
The breakthrough came on December 16, 1947. Walter Brattain and John Bardeen, working without Shockley in the room, demonstrated the first point-contact transistor: two gold contacts pressed against a germanium surface, boosting signal power and voltage without a vacuum tube. The name "transistor", transfer plus resistor, was coined by Bell Labs engineer John Robinson Pierce, who also wrote science fiction, in May 1948. Bell Labs announced the invention publicly that June.
Shockley, excluded from the demonstration, responded with furious brilliance. On January 23, 1948, he conceived the junction transistor, a three-layer "sandwich" of germanium or silicon that proved far more robust and manufacturable than the point-contact design. By July 1951, the junction transistor was announced as a commercial product. Bardeen, Brattain, and Shockley shared the 1956 Nobel Prize in Physics. Bardeen later became the only person to win two Nobel Prizes in Physics, also winning in 1972 for the BCS theory of superconductivity.
But the transistor was only the beginning. In December 1957, Jean Hoerni at Fairchild Semiconductor invented the planar process, protecting exposed p-n junctions with a silicon dioxide layer and creating a flat, reliable surface. Gordon Moore recalled: "Where most of us were setting up the facilities and developing the early processes, Hoerni was drawing in his notebook." The results were "fantastic", planar transistors survived the brutal environments of intercontinental ballistic missiles.
Robert Noyce then saw what Hoerni's process truly enabled. By adding a metal layer to interconnect transistors as an additional manufacturing stage, Noyce eliminated hand-wiring entirely, creating the integrated circuit on a single substrate. In September 1958, Jack Kilby at Texas Instruments had demonstrated the first working IC, a germanium flip-flop with wire-bonded components, but Noyce's planar approach proved scalable. After years of legal warfare, Texas Instruments and Fairchild cross-licensed; today both Kilby and Noyce are recognized as co-inventors. Kilby received the 2000 Nobel Prize; Noyce had died in 1990 and so could not share it.
In April 1965, Gordon Moore published his article in Electronics magazine predicting that the number of components on an integrated circuit would double every year at minimum cost. He later revised this to every two years. Moore's Law was born. Moore himself identified the 1959 planar transistor as the true origin point of the trend.
Enabled
Everything. Without the transistor, no computers. Without the IC, no computers small enough to be useful. Without the planar process, no reliable manufacturing at scale. This era created the physical substrate upon which the entire modern world runs.
Why It Matters
As one Bell Labs engineer understood immediately: "The advantage of the transistor is that it is inherently a small-size and low-power device... The significance of the transistor is not that it can replace the tube but that it can do things the vacuum tube could never do!" The transistor didn't just improve amplification. It made possible a category of existence, computation everywhere, in everything, that was physically impossible before.
The Manufacturing Leap
The technical story of this era is MOS technology, the microprocessor, and the emergence of VLSI design. The human story matters just as much: a toxic genius, eight engineers with the courage to mutiny, and the invention of a regional culture that would reshape global capitalism.
In 1955, William Shockley left Bell Labs to establish Shockley Semiconductor Laboratory in Mountain View, California, the first high-tech company in what would become Silicon Valley. He chose the location partly because his mother lived nearby in Palo Alto. He recruited brilliant young talent: Robert Noyce, Gordon Moore, Julius Blank, Victor Grinich, Jean Hoerni, Eugene Kleiner, Jay Last, and Sheldon Roberts, who would become known as the "Traitorous Eight".
Shockley's management was catastrophic. He micromanaged, hopscotched between projects, threatened lie detector tests, and created what engineers described as a "contemptuous, stressful, and stifling" environment. In 1957, all eight left to found Fairchild Semiconductor. Kleiner wrote to his father's investment banker at Hayden Stone; the letter landed on a young Arthur Rock's desk. After approaching roughly 30 companies, Sherman Fairchild of Fairchild Camera and Instrument agreed to put up $1.5 million, in a novel arrangement that gave the founders equity.
This single event "set off a chain reaction of innovation, investment, and entrepreneurship that transformed a quiet region of California into Silicon Valley". By some counts, more than 400 companies trace their lineage directly or indirectly to Fairchild Semiconductor. The term "Silicon Valley" itself was coined in 1971 by columnist Don Hoefler for Electronic News. Stanford engineering dean Frederick Terman had been encouraging students and faculty to start companies; the Fairchild diaspora gave his philosophy an explosive proof point.
Meanwhile, the technology was maturing into product form. By 1964, MOS (Metal-Oxide-Semiconductor) chips had surpassed bipolar chips in transistor density and manufacturing cost. IBM researcher Robert H. Dennard, in autumn 1966, conceived the single-transistor/single-capacitor DRAM, "thinking of MOS technology, he realizes that a stored charge on a capacitor could represent a bit of information, while a transistor could control the writing of that charge". DRAM would become the dominant memory technology for decades.
Then came the microprocessor, and it came from an immigrant. Federico Faggin, a physicist from Italy, developed MOS silicon gate technology at Fairchild in 1968, a technology that was "about five times faster, had about 100 times less junction leakage, and could integrate twice as many random logic transistors" compared to aluminum gate technology. Gordon Moore acknowledged that "a major component of Intel's early success was due to their adoption of the SGT". Faggin joined Intel in 1970 and, in just nine months, designed the Intel 4004, the world's first single-chip microprocessor, containing 2,300 transistors on a 10-micrometer PMOS process. The 4004 and its descendants (8008, 8080) launched an era.
While Faggin was building the microprocessor, another revolution was brewing in design methodology. Lynn Conway, working with Carver Mead at Xerox PARC, launched what became known as the Mead-Conway Revolution. Their textbook Introduction to VLSI Systems and Conway's invention of scalable MOS design rules freed chip design from the control of commercial fabricators. Conway pioneered the first VLSI course at MIT in 1978; by 1983 nearly 120 schools taught the approach. This was more than pedagogy, it was the democratization of a skill that had been locked behind factory walls. Conway had been fired by IBM during her gender transition in 1968 and rebuilt her career in what she called "stealth mode" before emerging as one of the most influential figures in VLSI design.
Grace Hopper, meanwhile, had laid groundwork decades earlier that made modern computing accessible. She developed the A-0 compiler in 1951–1952, the first tool to translate mathematical notation into machine-readable code, and went on to develop FLOW-MATIC and was instrumental in creating COBOL. She was posthumously awarded the Presidential Medal of Freedom in 2016.
The 6502 microprocessor, developed by MOS Technology, drove prices to unprecedented lows and "together with the Z80 was largely responsible for the emergence of the computer hobbyist movement which in turn led to the home computer revolution of the 1980s".
Enabled
The personal computer industry. The fabless design ecosystem. The venture capital model. The concept that engineers with equity could outperform entrenched corporations. And, critically, the notion that semiconductor manufacturing was not just a technical discipline but a cultural one.
Why It Matters
This era established the template for how semiconductor innovation would be organized in America: brilliant immigrants, equity-compensated engineers, venture-backed startups, and a regional culture that treated company formation as a natural career progression. It also established the pattern that would repeat forever: the best people leave the worst managers, and the future belongs to those who let them.
The Manufacturing Leap
Moore's Law worked for decades because it had a physical mechanism: Dennard scaling. In 1974, Robert H. Dennard and his IBM colleagues published a landmark paper showing that as transistors shrink, their power density stays constant, enabling higher clock frequencies without increasing power consumption. For thirty years, this principle was the invisible engine driving the industry. Transistors got smaller, faster, cheaper, and more efficient, all at once.
But this era is defined as much by geopolitical competition as by physics. Japan emerged as a serious challenger. In the early 1970s, the U.S. held 60% of the world semiconductor market. By 1982, that had fallen to 51%; Japan's share had jumped from roughly 15% to 35%. Three factors drove Japanese success: the government-backed VLSI Program (1976–1979), strong domestic demand from a booming consumer electronics industry, and the vertical integration of Japanese electronics firms.
In DRAM, the industry's commodity bellwether, the collapse was devastating. U.S. market share "plummeted from 70 to 20 percent between 1978 and 1986 as the Japanese share jumped from under 30 to about 75 percent". American companies were being outmanufactured, not out-designed. Japanese firms achieved higher yields, lower defect rates, and more aggressive process refinement.
The U.S. industry fought back through politics. In 1985, the Semiconductor Industry Association (SIA) filed trade complaints alleging dumping. On March 14, 1986, Commerce ruled that Japanese firms had indeed dumped 256K and 1 Megabit RAMs, with dumping margins exceeding 100% for some firms. The U.S.-Japan Semiconductor Trade Agreement was signed in 1986, establishing mechanisms to halt predatory pricing and increase U.S. market access in Japan. In 1987, President Reagan imposed 100% tariffs on selected Japanese electronics.
The U.S. industry rebounded over the next decade, regaining global leadership by 1997 with 50% global market share, a position it continues to hold. The lesson: industrial policy, strategically applied, can shape competitive outcomes. Today's CHIPS Act architects studied Japan's challenge carefully.
Meanwhile, Dennard scaling served the industry faithfully until it didn't. "It broke down in the 2005–2006 time period". Around the 90nm node, "shrinking transistor dimensions further did not increase the speed of the transistor", creating what engineers call the power wall. "The power consumption of the chip rapidly increased resulting in thermal dissipation issues, without any performance benefit. Parts of the chip had to be turned off to manage thermal issues which came to be called 'dark-silicon'".
The industry pivoted. IBM's Power4 (2001) was the first dual-core processor, followed by AMD (2004) and Intel (2005). The focus shifted from single-threaded performance to parallel computing. By the 22nm era, about 20% of silicon was "dark," with predictions of 50% to 90+% dark silicon by decade's end.
Andy Grove led Intel through this era with the pivotal shift from memory to microprocessors, a strategic bet that defined the company's dominance for two decades. Grove's philosophy, captured in his book Only the Paranoid Survive, became the industry's operating manual.
Enabled
The personal computer revolution. The Japanese consumer electronics boom. The global DRAM market. The recognition that semiconductor manufacturing was a matter of national industrial policy. The multi-core architecture paradigm. And the dawning awareness that physics, not engineering ambition, would eventually set hard limits.
Why It Matters
This era proved that semiconductor leadership is not guaranteed to any nation. Japan's rise demonstrated that coordinated industrial policy can challenge even the most entrenched technological leaders. Japan's relative decline showed that market dynamics, trade policy, and strategic response matter as much as manufacturing prowess. These lessons are being replayed today with China.
The Manufacturing Leap
The most consequential business model innovation in semiconductor history was not a chip. It was a contract.
Morris Chang, a 25-year veteran of Texas Instruments, proposed a radical idea: a pure-play foundry that would manufacture chips exclusively for other companies, separating chip design from fabrication. TSMC was incorporated on February 21, 1987, in Hsinchu, Taiwan, with initial capital of NT$1.3 billion (approximately US$45 million): Taiwan's Executive Yuan Development Fund (48.3%), Philips (27.5%), and private investors.
The early skepticism was captured by AMD founder Jerry Sanders' famous quip: "Real men have fabs", reflecting the prevailing belief that owning fabrication facilities was essential to semiconductor identity. It was more than a slogan. It was an article of faith. Integrated Device Manufacturers (IDMs) like Intel and Texas Instruments designed their own chips and manufactured them in their own fabs. The idea that you could separate the two, that a company could design world-class chips without ever touching a cleanroom, struck many as absurd.
TSMC proved them wrong. By focusing solely on manufacturing and rigorously protecting client intellectual property, it enabled the fabless revolution. Companies like NVIDIA, Qualcomm, Broadcom, and AMD could focus on design without the enormous capital burden of building fabs that cost billions and required decades of operational expertise.
The irony is exquisite. In 2009, AMD, Sanders' own company, spun off its manufacturing into GlobalFoundries and became fully fabless. The company that had coined "real men have fabs" reversed its own philosophy. Today the most valuable semiconductor company in the world, NVIDIA, has never owned a serious fab. Jensen Huang's empire runs on TSMC's wafers. The foundry model didn't just change the economics, it inverted the ideology.
TSMC went public in 1994 (Taiwan) and 1997 (NYSE). By 2000 it became the world's largest independent foundry by capacity and revenue. Today TSMC commands approximately 55% foundry market share, producing roughly 99% of global AI accelerators by share, with 2024 revenue exceeding $85 billion.
But the foundry model alone wasn't enough. This era also produced the tool ecosystem that made advanced manufacturing possible at all. EUV (Extreme Ultraviolet) lithography research began in the 1980s, with industrialization work kicking off in 1994 with an industry coalition. The journey consumed more than $9 billion in R&D over more than 30 years. Key partners included Carl Zeiss (ultra-flat mirrors) and major semiconductor companies, Intel, Samsung, and TSMC acquired a collective 23% equity stake in ASML in 2012 to help fund the project.
The first commercial EUV machines were delivered in 2010; broader adoption came by 2020. The first EUV production system, the TWINSCAN NXE:3300, shipped in 2013. In early 2020, ASML celebrated its 100th EUV system shipment. ASML is now the sole producer of EUV lithography systems on Earth, a monopoly that makes it one of the most strategically important companies in the world.
Chenming Hu, "Father of the FinFET," developed the 3D transistor architecture that saved Moore's Law at sub-22nm nodes, providing a physical path forward when planar transistors could no longer scale.
Enabled
The fabless semiconductor ecosystem, hundreds of design companies that exist only because they don't need to build fabs. The concentration of leading-edge manufacturing in a small number of extremely capable foundries. The transformation of Taiwan into the world's most critical manufacturing node. And the structural vulnerability that comes with that concentration.
Why It Matters
The foundry model resolved a tension that had been building since Fairchild: the best chip designers are rarely the best manufacturers, and vice versa. By separating the two disciplines, TSMC allowed each to reach heights that would have been impossible under the old IDM model. But it also created a new form of dependency. When the world's most advanced chips are manufactured on an island that China claims as its own, geography becomes destiny.
Related readingMap the physical constraints behind the foundry model.
The Manufacturing Leap
The era of the massive, single-piece monolithic processor is effectively over for high-performance applications. As of late 2025, the industry has pivoted to a new paradigm: "modular silicon components, known as chiplets, [are] 'stitched' together using advanced packaging techniques".
This is not a minor evolution. It is a fundamental restructuring of how chips are designed, manufactured, and assembled, potentially as significant as the transition from IDM to fabless. "The chiplet revolution is the definitive answer to the slowing of Moore's Law. As the physical limits of transistor shrinking are reached, the industry has pivoted to 'More than Moore', a philosophy that emphasizes system-level integration over raw transistor density".
The Universal Chiplet Interconnect Express (UCIe) standard, ratified in August 2025 (UCIe 3.0), provides a "lingua franca" enabling chips from different manufacturers to communicate as if on the same piece of silicon. For the first time, a system designer can combine chiplets from TSMC, Intel Foundry, Samsung, and specialty vendors into a single packaged product, mixing and matching process nodes, architectures, and suppliers.
Advanced packaging technologies, 2.5D interposers, fan-out wafer-level packaging (FOWLP), and 3D stacking, are essential for chiplet integration, making packaging "no longer a 'back-end' task but a co-design activity alongside chip design". The market for semiconductor interconnects is projected to grow dramatically, driven by AI and HPC workloads where "monolithic dies are economically and technically constrained to scale".
High-NA EUV (with numerical aperture increasing from 0.33 to 0.55) represents the next frontier in lithography. The first High-NA system was delivered in December 2023, with high-volume manufacturing expected in 2025–2026. Intel's Mark Bohr stated in 2011: "This change in the basic structure is a truly revolutionary approach, and one that should allow Moore's Law, and the historic pace of innovation, to continue".
AI is the demand engine driving all of this. Data center semiconductor revenue totaled $112 billion in 2024, up from $64.8 billion in 2023, driven by GPU and AI processor demand. NVIDIA's revenue grew 84% in 2024 to $46 billion. Samsung Electronics regained the #1 semiconductor vendor spot in 2024 with $66.5 billion in revenue. Global semiconductor sales reached $627.6 billion in 2024, a 19.1% increase from 2023. Logic products led with $212.6 billion in sales; memory surged 78.9% to $165.1 billion.
Enabled
Continued performance gains despite the physical limits of transistor scaling. A new ecosystem of specialized chiplet designers. The disaggregation of monolithic chips into optimized components, logic, memory, I/O, analog, each manufactured at its ideal process node. And the industrial infrastructure to assemble these heterogeneous components into functioning systems.
Why It Matters
The chiplet revolution acknowledges a truth that the industry has been reluctant to face: the single-die monolith was an artifact of a particular phase in semiconductor development, not the eternal form of computation. By breaking chips into components, the industry gains flexibility, yield, and specialization, but also complexity. The companies that master packaging and integration will be as important as the companies that master transistor design. This is a bet that system-level optimization can substitute for node-level scaling. The verdict is still being rendered.
Related readingWalk through the packaging chapter that makes chiplets work.
The Manufacturing Leap
The leap is not technical. It is structural: the recognition that semiconductor manufacturing capacity is a form of national power, and that controlling who can build what, and where, is a strategic imperative on par with controlling oil reserves or nuclear materials.
Semiconductors sit at the intersection of China's national security and high-tech growth ambitions. As early as 2014, President Xi stated that "semiconductors [are] a core technology that China should produce domestically". China's National Integrated Circuit (IC) Promotion Guidelines (2014) laid foundations for the Made in China 2025 plan, which set an ambitious goal of 70% semiconductor self-sufficiency by 2025. According to the Information Technology and Innovation Foundation, China "has probably invested the equivalent of the US CHIPS Act [in the semiconductor industry] virtually every year since 2014". As of 2025, "43% of registered capital across the Chinese semiconductor industry is either directly or indirectly owned or controlled by the Chinese state".
The U.S. responded with layered, escalating supply chain restrictions: direct sale restrictions on advanced chips; semiconductor manufacturing equipment controls targeting ASML and other tool vendors; export controls on AI chips and manufacturing components; and restrictions on U.S. persons working with Chinese semiconductor companies. The approach has been described as "small yard, high fence", restricting only the most advanced technologies while allowing trade in commodity chips to continue.
The CHIPS and Science Act of 2022 allocated $53 billion to boost U.S. semiconductor production. It is, in many ways, America's own VLSI Program, a recognition that market forces alone cannot guarantee domestic manufacturing capacity for technologies that are both economically critical and militarily decisive. In 2025, the Trump administration announced 100% tariffs on foreign semiconductors, a dramatic escalation that has raised concerns about supply chain disruption and cost increases.
The consequences ripple through every layer of the industry. TSMC is building fabs in Arizona, a $65 billion investment that represents the most advanced semiconductor manufacturing on U.S. soil in decades. Intel is attempting a foundry transition under government pressure and market necessity. Samsung and SK Hynix are expanding U.S. operations. The geography of semiconductor manufacturing is being redrawn in real time, driven not by economics but by geopolitics.
Global semiconductor sales reached $627.6 billion in 2024, with projections exceeding $1 trillion by the early 2030s. Asia Pacific held 52.93% of the market in 2024, with China, Taiwan, and South Korea as manufacturing centers. Who controls this production, and who can be denied access to it, will shape the balance of power for decades.
Enabled
Government-funded domestic manufacturing. A new era of industrial policy in the United States and Europe. The strategic recognition that lithography equipment, AI accelerators, and advanced process nodes are as militarily significant as aircraft carriers. And the beginnings of a multi-polar semiconductor landscape where no single country controls the full stack.
Why It Matters
For seventy years, the semiconductor industry operated as a globalized market: design in the U.S., manufacturing in Asia, equipment from Europe, materials from everywhere. That model is fracturing. The U.S. is subsidizing domestic fabs. China is building its own parallel ecosystem under sanctions. Europe is racing to reclaim manufacturing share. The industry that defined globalization is now its most visible casualty.
This is the defining story of semiconductors in the 2020s. It is not about transistor counts or process nodes. It is about whether the United States and its allies can maintain access to the manufacturing capabilities that underpin AI, military systems, and economic competitiveness, and whether China can build its own alternatives before the technological gap becomes unbridgeable. The lineage of semiconductor manufacturing has become a lineage of industrial power.
Related readingRead how the supply chain is splitting now.