Jack Kilby and the Circuit That Fit on a Sliver of Germanium

Biographies · Semiconductor engineering and the hardware of AI

On 12 September 1958, Jack Kilby switched on an oscillator in Dallas. It produced a signal at about 1.3 megahertz. An ordinary oscillator could do that. What mattered was where the signal came from. Kilby and his colleagues had cut the circuit elements into germanium, a material better known for making transistors. Fine gold wires still arched above the device. Those wires help explain both why Kilby deserves a place among the inventors of the integrated circuit and why the story also belongs to engineers at another laboratory.

Importance of his work for artificial intelligence

Imagine taking a photo of a bicycle and asking your phone what it sees. To a computer, the photo is a grid of numbers. A neural network can examine small patches of that grid for edges, then combine those clues into shapes such as wheels. It learns which clues matter by adjusting numbers called weights during training. To recognise the bicycle in a new photo, it has to multiply and add numbers over and over again. How a network reads an image.

Now picture having to connect by hand the electronic parts that carry out those calculations. How much of your phone would the wiring take up? Kilby showed that several parts of a circuit could be formed in the semiconductor itself. Later engineers learned to manufacture the connections too. Today a chip small enough to fit inside a phone can carry out many calculations at once. Kilby’s place in the history of AI begins there, with the physical work that makes a decision as simple as “bicycle” possible. Kilby’s experiment · A neural engine inside a phone.

When the telephone lines went down

Kilby grew up in Great Bend, Kansas, where his father ran an electricity company serving rural communities. During an ice storm, telephone and power lines came down. His father turned to amateur radio operators to reach customers cut off by the storm, and the teenage Kilby went along. In his Nobel lecture he recalled what it meant for people who had lost power to hear from someone outside. Electronics had acquired a purpose he could recognise long before he had a degree in it. Kilby’s Nobel lecture · National Academy of Engineering memoir.

Service in the US Army during the Second World War interrupted his studies at Illinois. He returned and graduated in electrical engineering in 1947. At Centralab, a division of Globe-Union in Milwaukee, he learned to make compact circuits for consumer electronics. Conductive silver and resistive carbon inks could be applied to ceramic, although some parts still had to be attached individually. He earned his master’s degree in 1950 while working there. By the time he moved to Texas, he knew from experience which parts of a circuit could be patterned on a surface and which still cost someone time at an assembly bench. The Centralab years · Wisconsin’s graduate record.

Two experiments in Dallas

Kilby joined Texas Instruments in May 1958. As a new employee he had no holiday allowance when most of the staff took their summer break. Engineers already knew how to shrink a transistor. A machine containing thousands of components still needed thousands of reliable connections. On 24 July he set down a proposition in his notebook. The same material that made a transistor could also provide resistance and capacitance. What if a manufacturer formed all three where they were needed? Kilby’s laboratory notebook · His account of the summer.

His supervisor, Willis Adcock, asked for a demonstration. On 28 August Kilby showed him a circuit assembled from separate silicon parts. It answered the materials question. The September experiment tackled the harder problem of building the elements into the semiconductor. Technicians Pat Harbrecht and Tom Yeargan cut the germanium into narrow bars about a centimetre long. In his lecture Kilby remembered three completed oscillators and the frequency of the first. The fine gold wires provided the electrical connections. Each one also represented a job that would have to be repeated if anyone wanted to make the device by the thousand. Kilby’s description of the two tests · Computer History Museum account.

What the gold wires left to solve

The idea of putting a circuit in a solid block was already in circulation. The British engineer Geoffrey Dummer had described that possibility in 1952. Kilby supplied a working demonstration and filed his US patent application in February 1959. Across the country at Fairchild Semiconductor, Robert Noyce had independently been thinking about how to make the connections themselves part of the manufacturing process. Jean Hoerni’s planar process left a protective oxide layer on silicon. Noyce described running strips of metal across that insulating surface, with openings where contact was needed. The connections could now be laid out as a pattern. Dummer’s proposal · Kilby’s patent · Noyce’s patent.

Electrical isolation posed another problem. Neighbouring components had to share a piece of silicon yet remain electrically separate except where the design connected them. Kurt Lehovec proposed using semiconductor junctions for isolation. At Fairchild, Jay Last’s team made a working planar circuit in May 1960 using physical separation and another in September using junction isolation. The later patent dispute between Kilby and Noyce turned on a precise point about a conductor adhering to an insulating layer. Asking who made a semiconductor circuit work and asking who devised the interconnections used in production brings different people into view. Lehovec and the other patent holders · The Fairchild experiments · The court’s opinion.

A calculation you could hold in one hand

In 1967 Kilby could show someone an application of integrated circuits and let them try it. The Cal Tech calculator had keys under the user’s fingers and a thin strip of paper on which the answer appeared. Kilby led the team. Jerry Merryman worked on the processor and memory, and James Van Tassel designed the keyboard. A thermal printer made the digits visible on paper. All three appear on the calculator patent. Over the course of his career Kilby held more than sixty patents, and the National Academy of Engineering also credits him as a co-inventor of the thermal printer. The surviving calculator · The three inventors’ patent · Engineering memoir.

In 2000 the Royal Swedish Academy of Sciences awarded Kilby half the Nobel Prize in Physics for his part in the invention of the integrated circuit. Zhores Alferov and Herbert Kroemer shared the other half for work on semiconductor heterostructures. The wording left room for the several histories inside one device. Kilby died of cancer at his home in Dallas on 20 June 2005, aged 81. The Academy’s award announcement · Wisconsin’s remembrance.

Where the distance matters in AI

A neural network has to multiply numbers and add the products, often repeating the operation many times with the same weights. In a 2017 paper, Norman Jouppi and colleagues described a processor already serving neural networks in Google’s data centres. Its matrix unit contained 65,536 multiply-and-accumulate units working with eight-bit inputs. Putting that arithmetic together on a chip allowed many operations to proceed at once. Their measurements also showed why arithmetic alone was an incomplete guide to speed. Some networks spent much of their time waiting for weights to arrive from memory. The original TPU study.

Six years later the researchers described a machine linking 4,096 chips with a network that optical switches could reconfigure. In November 2026 Google described another choice, placing two accelerator chips in one package so that more data could reach the arithmetic units. The distances have changed since Kilby worked at his bench. Deciding what to build together and what to connect across a gap is still part of designing a computer. Jouppi and colleagues on TPU v4 · Google’s description of the two-chip package.

A chronology

  1. 1923–1950Born in Missouri and raised in Kansas. Earns electrical engineering degrees in Illinois and Wisconsin.
  2. 1958Joins Texas Instruments in May. Records his idea on 24 July. Demonstrates the germanium oscillator on 12 September.
  3. 1959–1960Kilby and Noyce file separate patent applications. Last’s team demonstrates planar integrated circuits at Fairchild.
  4. 1967Kilby, Merryman and Van Tassel build their handheld calculator prototype.
  5. 1969–1970Receives the US National Medal of Science for 1969 at the White House on 16 February 1970. Official award record.
  6. 1993–2005Receives the Kyoto Prize in 1993 and the Nobel Prize in Physics in 2000. Dies in Dallas in 2005.

Further reading

In 1976 Kilby reconstructed the decisions behind his first circuits for readers of IEEE Transactions on Electron Devices. Arjun N. Saxena later examined the patents against a stricter technical definition of a monolithic circuit. Read together, the two accounts show why the choice of a definition changes which achievement a historian calls the first. Kilby’s retrospective · Saxena’s analysis.

Liked this post? Follow this blog to get more.