Integrated Circuit
An integrated circuit is a complete electronic circuit, with its transistors, resistors, capacitors and the wiring between them, fabricated together on a single small piece of semiconductor, almost always silicon, rather than assembled from separate parts. The piece of silicon is called a die, and the die sealed inside a protective package with metal pins or pads is what people usually call a chip. By building every component and connection in one batch process, an integrated circuit makes electronics smaller, cheaper, faster and far more reliable than hand-soldered discrete parts, which is why one device can contain billions of transistors.
The integrated circuit was invented twice in quick succession, by two engineers solving different halves of the problem. In September 1958 Jack Kilby at Texas Instruments demonstrated a working circuit built on one piece of germanium, though its components were still joined by tiny hand-bonded wires. In 1959 Robert Noyce at Fairchild Semiconductor, building on Jean Hoerni's planar process, described a silicon version in which a flat oxide layer protected the surface and deposited metal lines formed the interconnections on the chip itself, an approach that could be mass-produced. The two companies later cross-licensed their patents; Kilby received the Nobel Prize in Physics in 2000, with Noyce having died in 1990. Gordon Moore, a Fairchild co-founder, observed in a 1965 article that the number of components on a chip was doubling at a steady pace and in 1975 set the doubling period at about two years, the observation known as Moore's law.
Fabrication is a layered printing process, and its economics explain the industry's structure. A silicon wafer, today typically 300 mm across, is coated, exposed to light through a patterned mask (photolithography), etched and chemically doped, and the cycle repeats for dozens of layers including a stack of metal wiring on top. Extreme ultraviolet lithography machines made by ASML are used for the finest features in leading-edge chips. Because a single speck of dust can ruin a die, manufacturers measure yield, the fraction of dies on a wafer that work, and larger dies yield worse, which pushes designers toward chiplets: several small dies in one package. A fabrication plant costs billions of currency units to build, so most chip companies are "fabless", designing chips that foundries such as TSMC manufacture for them.
Integrated circuits come in many kinds, and "chip" does not mean "processor". Memory chips such as DRAM and flash, analog and power-management chips, radio and sensor chips, application-specific integrated circuits (ASICs) built for one purpose, field-programmable gate arrays (FPGAs) that can be rewired after manufacture, and systems on a chip (SoCs) combining processor cores, graphics, memory controllers and radios on one die are all integrated circuits. The Intel 4004 of 1971, the first commercial microprocessor, had about 2,300 transistors; chips such as the large GPUs for AI training now pack tens of billions. Older classifications of small-, medium-, large- and very-large-scale integration (SSI to VLSI) described gate counts, though the terms are rarely used today.
Several misconceptions surround chip technology. Process node names such as "5 nm" or "3 nm" are now marketing labels and do not correspond to the length of any particular physical feature; transistor density, power and performance are the comparisons that matter. Smaller transistors no longer automatically mean a faster clock, because power density limits the speed (the end of Dennard scaling), so gains now come from architecture, specialised accelerators and packaging. Supply chains are also concentrated and fragile, as the shortage of 2020 to 2022 showed, since chips pass through design software, fabrication, packaging and testing across several regions.
For people writing software or training models, the integrated circuit sets the real limits of what can be done: how much memory sits close to the cores, how many operations per watt are available, and whether the task maps to a CPU, GPU or custom accelerator all come down to what was etched into the silicon. The industry's progress, from Kilby's single sliver of germanium to billions of transistors on a thumbnail-sized die, is a story of manufacturing precision as much as of circuit design.
Read an integrated circuit as a whole circuit printed in one piece: instead of wiring parts together, the parts and wires are grown into silicon layer by layer, which is why electronics became small, cheap and everywhere.