# History of Semiconductors The semiconductor is the physical substrate of all power in the twenty-first century. Control the chip supply and you control artificial intelligence, military capability, economic production, and the global order itself. The story of semiconductors is not a story about electronics — it is a story about how a material property of silicon became the most strategically consequential resource on Earth, more important than oil, more contested than uranium, and concentrated in a manufacturing geography that makes Taiwan the most dangerous flashpoint in great-power competition. Every node in the vault's tech-politics web — [[NVIDIA]]'s GPU dominance, [[DeepSeek]]'s frontier-AI-on-older-chips proof of concept, the [[GPU Export Controls]] battle, the [[CHIPS and Science Act]]'s industrial policy, [[Palantir]]'s defense-tech contracts — sits on top of semiconductor manufacturing. This page traces the arc from nineteenth-century physics curiosity to twenty-first-century geopolitical weapon. ## Theoretical Foundations Semiconductor properties — materials that conduct electricity under some conditions but not others — were observed in the nineteenth century but remained laboratory curiosities for decades. Michael Faraday noted the temperature-dependent resistance of silver sulfide in 1833. Karl Ferdinand Braun discovered rectification (one-way current flow) in crystal-metal contacts in 1874. The practical application came with "cat's whisker" detectors — fine wire contacts on galena crystals — used in early radio receivers from the 1900s through the 1920s. These were semiconductors in the wild before anyone had a theory to explain them. Quantum mechanics, developed in the 1920s and 1930s, finally provided the theoretical framework. The band theory of solids — explaining why some materials conduct, some insulate, and some sit in between — made semiconductor behavior predictable and, eventually, engineerable. By the late 1930s, researchers at Bell Labs, Purdue, and MIT were deliberately studying semiconductor materials with an eye toward replacing the fragile, power-hungry vacuum tubes that dominated electronics. ## The Transistor (1947) On December 16, 1947, John Bardeen and Walter Brattain demonstrated the point-contact transistor at Bell Labs — a solid-state amplifier that could do the work of a vacuum tube with a fraction of the power, heat, and size. Their supervisor, William Shockley, who had been pursuing a different transistor design, was furious at being sidelined and spent the next several months developing the bipolar junction transistor, a more practical and manufacturable design that he published in 1948. All three shared the 1956 Nobel Prize in Physics. The politics of that prize allocation — Shockley positioning himself as co-equal inventor despite Bardeen and Brattain doing the breakthrough experimental work — foreshadowed the credit disputes that would define [[Silicon Valley]] culture for decades. Shockley's post-Nobel trajectory is where semiconductor history becomes [[Silicon Valley]] history. In 1956, he moved to Mountain View, California — drawn by proximity to Stanford, where [[Frederick Terman]] was building the engineering ecosystem that would become the Valley's academic backbone — and founded Shockley Semiconductor Laboratory. He recruited brilliant young physicists and engineers. Then he managed them with paranoid authoritarianism: lie detector tests, obsessive secrecy, refusal to pursue the silicon research his team believed was most promising. ## Fairchild and the Traitorous Eight (1957) Eight of Shockley's researchers — Julius Blank, Victor Grinich, Jean Hoerni, Eugene Kleiner, Jay Last, [[Gordon Moore]], Robert Noyce, and Sheldon Roberts — defected in 1957, secured backing from Sherman Fairchild, and founded [[Fairchild Semiconductor]]. Shockley called them the "Traitorous Eight." The label stuck as a badge of honor, establishing the founding myth of [[Silicon Valley]]: that betrayal of hierarchy is the highest form of innovation. Two breakthroughs at Fairchild made the modern semiconductor industry possible. Jean Hoerni's planar process (1959) used oxide layers to protect transistor junctions, making silicon chips mass-producible for the first time. Robert Noyce, building on the planar process, invented the monolithic integrated circuit in 1959 — multiple transistors on a single silicon chip. Jack Kilby at Texas Instruments had independently demonstrated an integrated circuit in 1958, using a different approach (germanium, wire bonds). Noyce's silicon-based version proved more manufacturable, and the two are credited as co-inventors. The integrated circuit is the invention that made everything else possible: computers, telecommunications, the internet, artificial intelligence — all of it runs on Noyce's insight that you could put an entire circuit on one chip instead of soldering individual components together. ## The Silicon Valley Explosion (1960s–1970s) [[Fairchild Semiconductor]] became the most productive corporate diaspora in economic history. By the late 1960s, the "Fairchild family tree" was an industry joke — and then a research subject. The major offspring: [[Intel Corporation]] (1968, founded by Noyce and [[Gordon Moore]]), [[AMD]] (1969, founded by Jerry Sanders), and National Semiconductor, plus dozens of smaller companies. The semiconductor industry didn't just grow in [[Silicon Valley]] — it *was* Silicon Valley. Eugene Kleiner, one of the Traitorous Eight, co-founded [[Kleiner Perkins]] in 1972, seeding the venture capital ecosystem that would fund Apple, Amazon, [[Google]], and the entire modern tech landscape. The structural link between semiconductor innovation and venture capital — the idea that technical founders could attract outside risk capital to commercialize research — was established here. This lineage runs directly through to [[Sequoia Capital]], [[Andreessen Horowitz]], [[Founders Fund]], and the VC infrastructure now funding [[NVIDIA]], [[OpenAI]], and [[Anthropic]]. The early chip industry's customers were overwhelmingly military and aerospace. The Apollo guidance computer used integrated circuits. [[NASA]] and the [[Department of Defense]] were the volume purchasers who drove costs down the learning curve. This is the original relationship between the state and [[Silicon Valley]] that the industry's libertarian self-mythology prefers to forget: the foundational technology was government-funded, military-driven, and produced under contract to the [[Military-Industrial Complex]]. The semiconductor industry has never actually been the free-market story it tells about itself. ## Moore's Law (1965–Present) In 1965, [[Gordon Moore]] published his observation that the number of transistors on an integrated circuit was doubling roughly every year (later revised to every two years). [[Moores Law]] is not a law of physics — it is an economic observation that became a self-fulfilling industry roadmap. For sixty years, the semiconductor industry organized itself around maintaining the Moore's Law cadence: each new process node shrinks transistor dimensions, doubling density, reducing cost per transistor, and increasing performance. The economic implications are staggering. A transistor that cost a dollar in the 1960s costs a fraction of a billionth of a dollar today. This exponential cost reduction is the engine beneath every digital revolution — personal computers, smartphones, cloud computing, and now AI training at scale. [[NVIDIA]]'s H100 GPU contains 80 billion transistors; the original Intel 4004 microprocessor (1971) contained 2,300. The ratio is not a metaphor for progress; it *is* the progress. Moore's Law is now approaching physical limits. Transistor gate lengths are measured in nanometers — a few dozen atoms across. Quantum tunneling effects, heat dissipation, and manufacturing complexity are making each successive node more expensive. The cost of a cutting-edge fabrication facility ("fab") has risen from hundreds of millions in the 1990s to over $20 billion today. This economics has driven the most consequential structural change in the industry: the separation of design from manufacturing. ## The Fabless Revolution Beginning in the 1980s and accelerating through the 1990s, the semiconductor industry split into two tiers. "Fabless" companies — Qualcomm, Broadcom, [[NVIDIA]], and later Apple's chip-design division — design chips but own no fabrication facilities. "Foundries" — companies that manufacture chips designed by others — handle the production. This separation was driven by the escalating cost of fabs: only a handful of companies could afford to build and maintain cutting-edge manufacturing, so it made economic sense for designers to outsource production. The fabless revolution produced extraordinary innovation in chip design (NVIDIA's GPU architecture, Qualcomm's mobile processors, Apple's M-series chips) while concentrating manufacturing capability into fewer and fewer hands. By the 2020s, this concentration had reached a geopolitically dangerous extreme. ## TSMC and the Geopolitics of Chips Morris Chang founded Taiwan Semiconductor Manufacturing Company (TSMC) in 1987, pioneering the pure-play foundry model — a company that manufactures chips for other companies but designs none of its own. The model was initially viewed as niche. It became the most strategically important business model in the world. By the 2020s, TSMC manufactured approximately 90% of the world's most advanced semiconductor chips (sub-7-nanometer nodes). [[NVIDIA]]'s GPUs, Apple's processors, Qualcomm's mobile chips, and AMD's server CPUs are all manufactured by TSMC in Taiwan. This concentration means that a single natural disaster, a single military conflict, or a single political decision in Taiwan could cripple the global technology supply chain. Taiwan's semiconductor dominance is sometimes called its "silicon shield" — the theory that China will not invade because doing so would destroy the fabrication capacity that China itself depends on. Whether the shield holds is one of the defining strategic questions of the century. The intersection with the vault's China-tracking work is direct: the chip supply chain is the material substrate beneath the [[GPU Export Controls]] and [[Compute Sovereignty]] debates. ## CHIPS and Science Act (2022) The [[CHIPS and Science Act]], signed into law in August 2022, authorized approximately $280 billion in federal investment to reshore semiconductor manufacturing to the United States — $52 billion in direct subsidies for fab construction plus substantial research-and-development funding. TSMC, Samsung, and [[Intel Corporation]] announced major fab construction projects in Arizona, Texas, and Ohio. The Act is industrial policy by a country whose dominant political ideology has spent four decades opposing industrial policy. The [[Neoliberalism]] consensus that the market allocates capital more efficiently than the government was quietly abandoned when it became clear that the market had allocated cutting-edge chip manufacturing to a single island 100 miles from mainland China. The bipartisan support for the CHIPS Act is a concession that some supply chains are too strategically important for market allocation — a concession that sits uncomfortably next to the same political coalition's insistence that government shouldn't pick winners and losers in other sectors. ## China's Semiconductor Ambitions China has spent over $150 billion attempting to build an indigenous advanced semiconductor industry. The results have been mixed. Huawei's chip-design arm, HiSilicon, produced competitive mobile processors until U.S. sanctions cut off access to TSMC manufacturing. SMIC (Semiconductor Manufacturing International Corporation), China's most advanced foundry, has achieved 7-nanometer production using older DUV lithography equipment — a technical achievement, though still generations behind TSMC's cutting edge. The [[GPU Export Controls]] imposed by the Biden administration in October 2022 (and tightened in October 2023) restrict the export of advanced AI chips and chipmaking equipment to China. The policy thesis: deny China access to the compute necessary for frontier AI development, thereby maintaining a U.S. advantage in AI capability. The logic is straightforward — and [[DeepSeek]] punched a hole in it. In January 2025, DeepSeek released R1, a frontier-competitive AI model reportedly trained on older NVIDIA chips that fell below the export-control threshold. The demonstration that China could achieve near-frontier AI performance with less-advanced hardware undermined the core assumption of the export-control strategy. If compute restrictions don't prevent capability development but merely slow it — while simultaneously motivating Chinese self-sufficiency in chip design and manufacturing — the policy may be accelerating the very decoupling it was designed to prevent. ## Why This Matters for the KB The semiconductor story is the vault's deepest infrastructure layer. Every political-technological dynamic the KB tracks — [[Surveillance State]] data-processing capacity, [[NVIDIA]]'s AI-training dominance, [[Palantir]]'s defense analytics, [[Silicon Valley AI Ecosystem]] power, [[GPU Export Controls]] as geopolitical weapon — depends on semiconductor manufacturing. Three through-lines connect to the broader KB: **The state-market fiction.** The semiconductor industry was born from [[Department of Defense]] contracts, sustained by government R&D funding, and is now being reshored with $280 billion in federal subsidies via the [[CHIPS and Science Act]]. The industry's libertarian self-mythology — and [[Silicon Valley]]'s broader free-market identity — is contradicted by its actual history at every stage. The same political coalition that opposes government intervention in markets treats semiconductor supply chains as a national security priority warranting massive state intervention. **The concentration-of-power problem.** [[Moores Law]]'s economics drove fabrication into fewer and fewer hands until a single company (TSMC) on a single island (Taiwan) manufactures the chips that power the global economy. This is the [[Military-Industrial Complex]] logic applied to civilian infrastructure: the pursuit of efficiency creates strategic vulnerability. The parallel to the vault's tracking of [[Compute Sovereignty]] and the broader concentration of AI capability in a handful of companies is direct. **The export-control arms race.** The [[GPU Export Controls]] are the new containment policy — an attempt to use semiconductor supply-chain leverage to constrain Chinese AI capability. [[DeepSeek]]'s R1 demonstrated the limits of that approach. The policy debate (restrict chips to slow China vs. risk accelerating Chinese self-sufficiency) is one of the defining strategic questions of the 2020s, and it all traces back to the fact that a paranoid Nobel laureate's employees betrayed him in a Mountain View lab in 1957. ## See Also - [[Fairchild Semiconductor]] — the spawning ground - [[Silicon Valley]] — the ecosystem Fairchild created - [[Gordon Moore]] — co-founder of Fairchild and Intel, author of Moore's Law - [[Moores Law]] — the economic engine - [[NVIDIA]] — the AI-era chip-design dominant - [[CHIPS and Science Act]] — the reshoring policy - [[GPU Export Controls]] — the containment policy - [[Compute Sovereignty]] — the strategic framework - [[DeepSeek]] — the export-control stress test - [[Military-Industrial Complex]] — the original customer - [[Silicon Valley AI Ecosystem]] — the downstream power structure - [[Artificial Intelligence Timeline]] — the capability trajectory - [[Frederick Terman]] — the Stanford ecosystem builder - [[Cold War]] — the funding context