Integrated circuit

• 1. Introduction

  ›

  A modern integrated circuit can pack many billions of transistors into an area the size of a human fingernail. That single fact captures a transformation that began in the 1960s, when the first chips held only a handful of transistors. The integrated circuit, also called a microchip or simply a chip, is a compact assembly of electronic components built onto a thin, flat piece of semiconductor material, most commonly silicon. Transistors, resistors, and capacitors sit together with their interconnections on that tiny slab. Today these devices live inside computers, smartphones, and televisions, handling data processing, control, and storage. How did engineers learn to print an entire circuit as a single unit rather than wiring it together one transistor at a time? Who first proved it could work, and who turned the prototype into something the world could mass-produce? And why does a single fabrication facility now cost more than twelve billion dollars to build? The answers run from a quiet July in 1958 through Apollo, calculators, and a law named after a man called Moore.

• 2. Two Inventors, One Idea

  ›

  In July 1958, newly employed at Texas Instruments, Jack Kilby recorded his initial ideas about the integrated circuit. On the 12th of September 1958 he demonstrated the first working example. His patent application of the 6th of February 1959 described the device as a body of semiconductor material in which all the components of the electronic circuit are completely integrated. The first customer for the invention was the US Air Force. In 2000 Kilby won the Nobel Prize in physics for his part in the work. Kilby's creation, though, was not a true monolithic chip. It depended on external gold-wire connections, which made large-scale production impractical. About six months later, Robert Noyce at Fairchild Semiconductor built the first practical monolithic IC chip. Noyce's version was fabricated from silicon using the planar process, the work of his colleague Jean Hoerni, allowing reliable on-chip aluminum interconnections. The monolithic chip drew on two further inventions: the planar process by Hoerni and p-n junction isolation by Kurt Lehovec. Hoerni's work itself built on earlier study of surface protection and passivation by silicon dioxide masking, and on research into the diffusion of impurities into silicon. Where Kilby had used germanium, Noyce used silicon, and modern chips descend from Noyce's monolithic design rather than Kilby's early prototype. A precursor concept had pointed the way: small ceramic substrates called micromodules, each holding one miniaturized component, assembled into a compact grid. That idea, considered promising in 1957, was proposed to the U.S. Army by Kilby himself, leading to the short-lived Micromodule Program before he abandoned it for something fundamentally new.

• 3. Bipolar Versus MOS

  ›

  Transistor-transistor logic, known as TTL, was developed by James L. Buie in the early 1960s at TRW Inc. It became the dominant technology for digital integrated circuits from the 1970s into the early 1980s. Dozens of TTL chips, wired together, were the standard way to build the processors of minicomputers and mainframes. The IBM 360 mainframes, the PDP-11 minicomputers, and the desktop Datapoint 2200 of 1970 were all built from bipolar integrated circuits, either TTL or the faster emitter-coupled logic. A different transistor would eventually win. The metal-oxide-semiconductor field-effect transistor, or MOSFET, was developed at Bell Labs between 1955 and 1960. Unlike bipolar transistors, which needed extra steps for p-n junction isolation, MOSFETs could be isolated from one another easily, an advantage first highlighted by Dawon Kahng in 1961. The earliest experimental MOS integrated circuit was a 16-transistor chip built by Fred Heiman and Steven Hofstein at RCA in 1962. General Microelectronics introduced the first commercial MOS integrated circuit in 1964, a 120-transistor shift register developed by Robert Norman. By that year, MOS chips had reached higher transistor density and lower manufacturing cost than bipolar chips. A turning point came in 1967, when Robert Kerwin, Donald Klein, and John Sarace at Bell Labs developed the self-aligned silicon-gate MOSFET. The following year, Federico Faggin at Fairchild Semiconductor built the first silicon-gate MOS IC technology with self-aligned gates, the basis of all modern CMOS chips. Engineers soon realized a complete computer processor could fit on a single MOS LSI chip, which led to the microprocessor and the microcontroller by the early 1970s. Yet for a time the old guard held its lead: computers built entirely from TTL, like the Datapoint 2200, stayed faster and more powerful than single-chip MOS microprocessors such as the 1972 Intel 8008, until the early 1980s.

• 4. Moore's Law And Scaling

  ›

  Gordon Moore originally stated that the number of transistors would double every year, then revised the claim to every two years in 1975. This doubling, driven by smaller features and larger chips, became known as Moore's law. The extra capacity has been spent in two directions: lowering cost and adding function. As feature sizes shrink, nearly every aspect of a chip's operation improves at once. The cost per transistor and the switching power per transistor fall, while memory capacity and speed rise, through the relationships defined by Dennard scaling. Because users feel those gains in speed, capacity, and power, manufacturers compete fiercely to reach finer geometries. Transistor sizes dropped from tens of microns in the early 1970s to 10 nanometers in 2017, a million-fold increase in transistors per unit area. As of 2016, typical chip areas ran from a few square millimeters to around 600 square millimeters, holding up to 25 million transistors per square millimeter. The pace of this shrinking was forecast for years by the International Technology Roadmap for Semiconductors. The final such roadmap was issued in 2016, replaced by the International Roadmap for Devices and Systems. Today the vast majority of all transistors are MOSFETs laid down in a single layer on one side of a silicon chip in a flat, two-dimensional planar process, but researchers have built prototypes that stack transistors into three dimensions.

• 5. Generations Of Integration

  ›

  The term large scale integration was first used by the IBM scientist Rolf Landauer to describe a theoretical concept. From it came the whole family of names: small-scale integration, medium-scale integration, very-large-scale integration, and ultra-large-scale integration. Each marked a leap in transistor count. Small-scale integration arrived in 1964 with chips holding one to ten transistors. Early linear ICs such as the Plessey SL201 or the Philips TAA320 had as few as two transistors. Medium-scale integration followed in 1968, with tens to hundreds of transistors per chip. In 1964 Frank Wanlass demonstrated a single-chip 20-bit shift register carrying a then-incredible 120 MOS transistors. Large-scale integration appeared in 1971, ranging from 500 to 20,000 transistors. The masks for these early devices were mostly created by hand, often using Rubylith tape, by specially hired layout professionals working under teams of engineers. Calculator chips and the first microprocessors of the early 1970s had under 4,000 transistors. True LSI circuits, approaching 10,000 transistors, began to appear around 1974 for computer main memories and second-generation microprocessors. Very-large-scale integration started with hundreds of thousands of transistors in the early 1980s. One-megabit RAM chips arrived in 1986, each carrying more than a million transistors. Microprocessors passed the million-transistor mark in 1989 and the billion-transistor mark in 2005. As of 2023, maximum transistor counts have grown beyond 5.3 trillion per chip. Some early designs refuse to retire: the 7400 series of TTL chips became a de facto standard and remains in production for old equipment and simple new devices.

• 6. Funded By Defense And Space

  ›

  Both the Minuteman missile and the Apollo program needed lightweight digital computers for their inertial guidance systems. NASA's Apollo program was the largest single consumer of integrated circuits between 1961 and 1965. Although the Apollo Guidance Computer led and motivated the technology, it was the Minuteman missile that forced it into mass production. Minuteman and various United States Navy programs accounted for the entire $4 million integrated circuit market in 1962. By 1968, government spending on space and defense still made up 37 percent of the $312 million total production. That guaranteed demand sustained the young market until costs fell far enough to reach industrial and then consumer buyers. The average price per integrated circuit dropped from $50 in 1962 to $2.33 in 1968. Chips began appearing in consumer products by the turn of the 1970s. A typical early use was FM inter-carrier sound processing in television receivers. The first practical application of MOS small-scale integration chips was not a household gadget at all, but NASA satellites.

• 7. Inside The Fab

  ›

  A semiconductor fabrication facility can cost over twelve billion US dollars to construct, and that cost climbs over time, a trend known as Rock's law. Inside, semiconductors are fabricated in a planar process built on three key steps: photolithography, deposition such as chemical vapor deposition, and etching, supplemented by doping and cleaning. Monocrystalline silicon is the main substrate, though III-V compounds such as gallium arsenide serve specialized roles in LEDs, lasers, solar cells, and the highest-speed circuits. Perfecting methods to grow crystals with minimal defects took decades. A chip is built from many overlapping layers, each defined by photolithography and normally shown in different colors. Some layers diffuse dopants into the substrate, some implant ions, some lay down conductors of doped polysilicon or metal, and some define the connections between conducting layers. A transistor forms wherever the gate layer crosses a diffusion layer, the self-aligned gate. Visible light cannot pattern features this small, since its wavelengths are too large, so shorter ultraviolet photons expose each layer, and electron microscopes become essential tools for debugging. Each device is tested before packaging by automated test equipment, a step called wafer probing. The wafer is then cut into rectangular blocks called dice, and each working die is connected into its package with aluminium or gold bond wires, attached by thermosonic bonding, a method first introduced by A. Coucoulas. Test cost can run over 25 percent of total fabrication cost for low-cost products. Some companies handle all of this in-house as integrated device manufacturers, like Intel and Samsung, while fabless companies such as Nvidia only design their chips and send the manufacturing to pure-play foundries such as TSMC.

• 8. Packaging And Hidden Art

  ›

  The earliest integrated circuits were packaged in ceramic flat packs, kept by the military for years because of their reliability and compact size. Commercial packaging shifted quickly to the dual in-line package, first in ceramic and later in plastic. In the 1980s the pin count of VLSI circuits outgrew the practical limit of that package, bringing in pin grid array and leadless chip carrier designs. Surface-mount technology emerged in the early 1980s, and one common form, the small-outline integrated circuit, occupies about 30 to 50 percent less board area than an equivalent dual in-line package and is roughly 70 percent thinner. Ball grid array packaging has existed since the 1970s, and the flip-chip ball grid array of the 1990s allowed far higher pin counts by mounting the die upside-down and spreading its connections across the whole die. Intel moved away from the pin grid array to land grid array and ball grid array beginning in 2004, releasing its last pin grid array socket in 2014 for mobile platforms. A packaged chip usually carries identifying marks: the maker's name or logo, the part number, a batch and serial number, and a four-digit date code. A part bearing the code 8341 was made in week 41 of 1983, roughly October of that year. Beyond these official marks, chip designers have long hidden surreptitious, non-functional images or words in the silicon, work sometimes called chip art or silicon graffiti. That instinct to sign one's work points to the people behind the parts. The descendants of Noyce's monolithic design now include the Intel 4004, generally regarded as the first commercially available microprocessor, whose line ran on through the 8008, the 8080, and the 8088 used in the original IBM PC.