01Executive summary
Three milestones in twenty-two years replaced the vacuum tube, then removed the wiring between components, then removed the assumption that computers must be connected point to point.
Bardeen, Brattain and Shockley demonstrated the point-contact transistor at Bell Labs in December 1947. Jack Kilby built a working monolithic circuit at Texas Instruments in 1958, and Robert Noyce at Fairchild produced a planar version the following year in which the interconnect was deposited rather than wired. In October 1969 two hosts exchanged the first message over the ARPANET, and the session crashed within seconds. Each of these is more interesting for what it made routine than for what it demonstrated.
02The transistor: what it replaced and why it mattered
A vacuum tube amplifies by controlling electron flow through a vacuum from a heated cathode. It works well and made radio, radar and the first electronic computers possible. Its problems are all consequences of that heated cathode: it consumes power continuously, generates heat, takes time to warm up, is physically large and fragile, and eventually burns out. A machine with eighteen thousand tubes will have one fail regularly, which places a hard practical ceiling on system size.
A transistor does the same job in solid material, with no heater, no vacuum and no filament to fail. The immediate advantages were power, size and reliability. But the consequence that reorganised the industry was different, and it was not obvious at the time.
A tube is manufactured individually and costs roughly the same whether you buy one or a million. A transistor is made by processing a wafer, and once the process exists the marginal cost of an additional device on that wafer approaches zero. Complexity stopped being expensive. Every subsequent development in computing follows from that single change in the cost structure, and it is why the integrated circuit was inevitable once the planar process worked.
Bardeen and Brattain built and demonstrated the point-contact device; Shockley, who led the group, developed the junction transistor theory shortly afterwards, and the junction device is what became manufacturable. The three shared the Nobel Prize in Physics in 1956. Field-effect concepts had been patented decades earlier by Julius Lilienfeld and others without a working device being built or the physics being understood. The accurate summary is that the effect had been anticipated, was first demonstrated in 1947, and was made practical over the following four years.
03The integrated circuit and the tyranny of numbers
By the late 1950s the limiting factor in electronics was not the components but the connections between them. A circuit with ten thousand components needs tens of thousands of soldered joints, each an opportunity for a failure. Reliability falls as complexity rises, and the industry called this the tyranny of numbers. It was a genuine ceiling.
Kilby’s insight was that if every component could be made from the same semiconductor material, they could all be fabricated in one piece and the interconnections would not be separate parts at all. His 1958 demonstration used germanium with fine wires bonded between elements — it proved the concept but was not manufacturable. Noyce’s planar version, building on Jean Hoerni’s planar process, isolated devices with a silicon dioxide layer and deposited the interconnect as a metal film patterned photographically. That removed the wires entirely.
- OxidationA silicon dioxide layer is grown on the wafer surface, serving as insulator and as a mask against dopant diffusion.
- PhotolithographyA pattern is projected onto photoresist and developed, defining where the oxide will be removed. Resolution here sets device size.
- DopingDopants are introduced through the openings to form junctions, with the surrounding oxide protecting the rest of the wafer.
- MetallisationA conducting film is deposited and patterned to connect the devices, replacing individually attached wires.
- Test and singulationDies are tested on the wafer, then separated and packaged. Yield — the fraction that works — governs the economics entirely.
The interconnect became part of the process
Deposited and patterned metal scales with the same photolithography that defines the devices. Bonded wires do not scale at all. Making the connection a process step rather than an assembly step is the whole of the advance.
Yield became the governing variable
Each die either works or does not. A defect anywhere kills that die. This makes contamination control, process uniformity and defect density the central engineering disciplines — the same shift toward process control seen in this series repeatedly.
Both men are properly credited. Kilby demonstrated that a monolithic circuit was possible; Noyce and Hoerni demonstrated the process that made it economic. The patent dispute between the two companies ran for years and was ultimately resolved by cross-licensing, which is a reasonable practical verdict on the question of who invented it.
04Packet switching and the first network
Telephone networks establish a circuit: a path is reserved end to end for the duration of the call, and it is held whether or not anyone is speaking. This is efficient for continuous speech and extremely inefficient for computer traffic, which is bursty — long silences punctuated by short transfers.
Packet switching breaks a message into addressed fragments that are forwarded independently and reassembled at the destination. No path is reserved, so links carry traffic from many conversations interleaved, and capacity is shared statistically. If a link or node fails, subsequent packets are simply routed differently, because no state about the conversation is held in the network.
The first transmission attempt in October 1969 was the word “login”, sent from UCLA to SRI. The receiving system crashed after the first two characters. It is a useful corrective to the way first-of-type milestones are usually described: the significant engineering was the years of protocol development, interface hardware and operational debugging that followed, not the first packet. Almost every entry in this series has the same shape, and treating the demonstration as the achievement consistently misplaces where the effort went.
05Takeaways for current practice
- Watch for changes in cost structure, not just performance. The transistor mattered most because it made complexity nearly free.
- When connections limit a system, integrate them into the process. The tyranny of numbers was solved by making interconnect a fabrication step.
- Yield governs any batch-fabricated product. Defect density and process uniformity outrank device performance in determining cost.
- Keep state at the edges where you can. Simple, stateless cores are more robust and far easier to extend.
- The demonstration is not the achievement. Plan and resource the interval between first working instance and dependable operation, because that is where projects actually fail.