Moore's law

• 1. Introduction

  ›

  Moore's law holds that the number of transistors in an integrated circuit doubles about every two years, with minimal increase in cost. That single observation, named after Gordon Moore, has shaped the semiconductor industry for decades and driven much of the technological change that defines modern life.

  Moore was not a physicist announcing a natural constant. He was the director of research and development at Fairchild Semiconductor in 1965, asked to write a short piece for Electronics magazine predicting where the semiconductor industry was headed over the next ten years. His response was a brief editorial titled "Cramming more components onto integrated circuits." In it, he speculated that by 1975, a single quarter-square-inch semiconductor could hold as many as 65,000 components. What began as a wild extrapolation, as Moore himself later called it, has since become a target, a planning tool, and something very close to a self-fulfilling prophecy.

  The questions worth asking are how that prediction became an industry law, what engineering breakthroughs kept it alive for so long, and whether the physical world has finally run out of room.

• 2. One Man, One Forecast

  ›

  Gordon Moore co-founded both Fairchild Semiconductor and Intel, and later served as Chief Executive Officer of Intel. Before any of that, in 1965, he observed that the number of components per integrated circuit had been doubling every year and projected that trend would continue for at least another decade. He later described the 1965 article as "a wild extrapolation saying it's going to continue to double every year for the next 10 years."

  A decade later, at the 1975 IEEE International Electron Devices Meeting, Moore revised the forecast. He predicted semiconductor complexity would continue doubling annually until about 1980 and then slow to a doubling approximately every two years, a compound annual growth rate of 41%. He pointed to several factors driving the exponential trend: the rise of metal-oxide-semiconductor technology, growing die sizes paired with falling defect densities, finer minimum dimensions, and what he called "circuit and device cleverness."

  Shortly after 1975, Caltech professor Carver Mead gave the trend its name. The label stuck, and Moore's prediction transformed into something the industry treated as a goal. Moore himself viewed this with a kind of wry humor, noting: "Moore's law is a violation of Murphy's law. Everything gets better and better."

  One historical detail complicates the tidy story of a lone visionary. A historian of the law cites Stigler's law of eponymy to note that the regular doubling of components was already understood by many in the field. The observation also predates Moore's direct involvement; in 1959, Douglas Engelbart studied the projected downscaling of integrated circuit size and published findings in an article called "Microelectronics, and the Art of Similitude." Engelbart presented his results at the 1960 International Solid-State Circuits Conference, where Moore was present in the audience.

• 3. The 18-Month Myth

  ›

  Intel executive David House, a colleague of Moore, introduced a separate but frequently confused prediction in 1975. House noted that Moore's revised law of doubling transistor count every two years implied that computer chip performance would roughly double every 18 months, with no increase in power consumption.

  The reasoning was grounded in a companion principle called Dennard scaling, formulated by Robert H. Dennard at IBM. Dennard observed that as MOS transistors get smaller, their power density stays constant, so power use remains in proportion with area. Combined with Moore's transistor-count doubling, this meant chips would get faster and more energy-efficient at the same time. According to Dennard's framework, transistor dimensions would shrink by 30% each technology generation, reducing their area by 50%, cutting delay by 30%, and increasing operating frequency by about 40%, all while keeping power consumption flat.

  Dennard scaling ended in 2005-2010, largely because of leakage currents. At small sizes, current leaks through transistors even when they are switched off, generating heat and threatening what engineers call thermal runaway. The breakdown prompted a shift toward multicore processors, but gains from adding cores fell short of what continued Dennard scaling would have delivered.

  The practical result shows up in single-core performance numbers. Single-core speed improved by 52% per year between 1986 and 2003 and by 23% per year from 2003 to 2011, but slowed to just seven percent per year between 2011 and 2018.

• 4. Breakthroughs That Kept the Pace

  ›

  Transistor counts grew by more than seven orders of magnitude in less than five decades, and that did not happen by accident. A series of specific inventions by named engineers at named institutions made it possible.

  The integrated circuit itself is the foundation. Jack Kilby at Texas Instruments invented the germanium hybrid IC in 1958. Robert Noyce at Fairchild Semiconductor followed in 1959 with the silicon monolithic IC chip. The CMOS process, which became the dominant manufacturing approach, was invented by Chih-Tang Sah and Frank Wanlass at Fairchild Semiconductor in 1963. Dynamic random-access memory was developed by Robert H. Dennard at IBM in 1967.

  Two IBM inventions from around 1980 proved especially consequential. Hiroshi Ito, C. Grant Willson, and J. M. J. Frechet invented chemically amplified photoresist, which was five to ten times more sensitive to ultraviolet light than prior materials. IBM introduced it for DRAM production in the mid-1980s. Separately, Kanti Jain at IBM invented deep UV excimer laser photolithography around the same period, a breakthrough later described as one of the major milestones in the fifty-year history of the laser.

  In the early 2000s, Gurtej Singh Sandhu at Micron Technology invented the atomic layer deposition high-kappa film and pitch double-patterning processes, extending Moore's law for planar CMOS technology to the 30 nm class and smaller. By 2019, leading manufacturers TSMC and Samsung Electronics were claiming to maintain the pace of Moore's law with 10, 7, and 5 nm nodes in mass production.

• 5. The Rising Cost of Progress

  ›

  As transistor counts climbed, the expense of building the factories to produce them climbed alongside. The cost of the tools used to manufacture chips, principally extreme ultraviolet lithography equipment, doubles every four years. This inverse relationship between consumer prices and producer costs gave rise to a second observation, also attributed to Moore's circle.

  Moore's second law, also called Rock's law after Arthur Rock, states that the capital cost of a semiconductor fabrication plant increases exponentially over time. It is the mirror image of Moore's original observation: while each transistor becomes cheaper to the buyer, the plant required to make smaller transistors becomes vastly more expensive to build.

  Brian Krzanich, the former CEO of Intel, cited Moore's 1975 revision as a precedent when addressing the slowdown observed since around 2010, calling it "a natural part of the history of Moore's law." In 2015, Intel stated that improvements in MOSFET devices had slowed starting at the 22 nm feature width around 2012 and continuing at 14 nm. That same year, Krzanich acknowledged that Intel's cadence was "closer to two and a half years than two."

  By the end of 2023, Pat Gelsinger, then Intel's CEO, was more candid: "We're no longer in the golden era of Moore's Law, it's much, much harder now, so we're probably doubling effectively closer to every three years now."

• 6. Reaching the Atomic Limit

  ›

  In April 2005, Gordon Moore said plainly that the projection could not last forever, describing the nature of exponentials: "You push them out and eventually disaster happens." He noted that transistors were approaching the size of atoms, calling that "a fundamental barrier," and estimated another ten to twenty years before reaching it.

  The physical limits have become concrete research problems. As device dimensions shrink, controlling current flow in thin channels becomes harder. Engineers have responded by redesigning the shape of the transistor itself. The FinFET, which wraps the gate dielectric on three sides of the channel, became the most common nanoscale transistor. The gate-all-around MOSFET, or GAAFET, offers even better control. A Toshiba research team led by Fujio Masuoka first demonstrated a GAAFET in 1988, calling it a surrounding gate transistor. Masuoka, better known as the inventor of flash memory, later left Toshiba and founded Unisantis Electronics in 2004 to pursue surrounding-gate research with Tohoku University.

  In 2012, a team at the University of New South Wales announced the first working transistor consisting of a single atom placed precisely in a silicon crystal. Moore's law had predicted that milestone for integrated circuits in the lab by 2020, meaning the researchers arrived eight years early.

  In May 2021, IBM announced the creation of the first 2 nm computer chip, with parts described as smaller than human DNA. Samsung outlined plans in 2024 to include production of a 2 nm chip in 2025.

• 7. Beyond Silicon

  ›

  Silicon has powered the integrated circuit era, but at single-nanometer scales its properties start working against engineers. Short-channel effects alter the material behavior that makes silicon useful as a transistor. Several alternative materials are under active investigation.

  Indium gallium arsenide, or InGaAs, is one leading candidate. Because of characteristics shared by III-V compound semiconductors, InGaAs transistors are considered more promising for high-speed, low-power logic than their silicon or germanium counterparts. In 2009, Intel announced 80 nm InGaAs quantum well transistors that the company reported performed equally as well as leading pure silicon transistors of the time while consuming less power. In 2012, a team at MIT's Microsystems Technology Laboratories built a 22 nm InGaAs transistor that was, at that point, the smallest non-silicon transistor ever built.

  Graphene is another candidate. Graphene nanoribbon transistors have appeared in research publications since 2008. Bulk graphene cannot be used in transistors because it has a band gap of zero and therefore cannot turn off, but the zigzag edges of nanoribbons introduce energy states that enable switching. A typical graphene nanoribbon of 10 nm width carries a bandgap energy of 0.4 eV. Further research on sub-50 nm graphene layers is still needed because resistivity increases at that scale.

  Spin-based logic, tunnel junctions, and nanowire geometry are among the other approaches being actively developed in labs. Researchers at Stanford University built a circuit modeled on the human brain in 2014; sixteen Neurocore chips were claimed to simulate one million neurons and billions of synaptic connections, running 9,000 times faster and more energy efficiently than a typical PC.

• 8. A Law That Became a Map

  ›

  Moore's law eventually spread well beyond transistor counts. Several related observations track exponential change across digital technology using similar logic.

  Dennard scaling projected that power per unit area would stay flat as transistors shrank, which held until 2005-2010. David House's 18-month chip performance doubling was a consequence of combining Dennard scaling with Moore's transistor count. Butters' Law of Photonics holds that the amount of data coming out of an optical fiber doubles every nine months, meaning the cost of transmitting a bit over an optical network halves at the same rate. Nielsen's Law says that bandwidth available to users increases by 50% annually. Haitz's law predicts that every ten years the brightness of LEDs increases twentyfold while manufacturing cost drops by a factor of ten. Swanson's law observes that the price of solar photovoltaic modules drops 20% for every doubling of cumulative shipped volume, with costs falling roughly 75% every ten years.

  Eroom's law, written as Moore's Law spelled backward, describes the opposite dynamic in pharmaceutical drug development: the cost of developing a new drug roughly doubles every nine years.

  The International Technology Roadmap for Semiconductors used Moore's law to drive the industry from 1998 until 2016, when it published its final roadmap and shifted to what it called a More than Moore strategy, in which the needs of applications drive chip development rather than a focus on semiconductor scaling. In September 2022, Nvidia CEO Jensen Huang declared Moore's law dead; several days later, Intel CEO Pat Gelsinger argued the opposite. In 2025, researchers Asif Alam and Muhammad Shah Alam proposed an RF analogue to Moore's law based on the historical frontier of transistor maximum oscillation frequency, finding that record values increased by approximately 1.6 times per decade from 1985 to 2025.