The Integrated Circuit: Kilby's Germanium Chip to 5-Nanometer Fabs

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

In the 1950s electronics were built from discrete resistors, capacitors and transistors joined by hand-soldered wires, a limit engineers called the tyranny of numbers, since one failed joint among thousands could crash a computer. The Army's 1957 Micromodule Program shrank the parts but not the wiring problem. In the summer of 1958 Jack Kilby, newly hired at Texas Instruments, carved a whole circuit from one piece of germanium, still connected by tiny gold wires, an idea that later won him the Nobel Prize in Physics. About six months later Robert Noyce at Fairchild Semiconductor built the first practical monolithic integrated circuit in silicon. It relied on Jean Hoerni's planar process, which built chips in flat layers topped with printed aluminum, and Kurt Lehovec's p-n junction isolation, which used depletion regions as electrical moats between transistors. Between 1961 and 1965 the Apollo program was the world's largest single consumer of integrated circuits.

MOS transistors isolated themselves and packed tightly, carrying chips from small scale to very large scale integration while the average price fell from $50 in 1962 to $2.33 in 1968. Moore's Law predicted transistor counts doubling about every two years, and features shrank from tens of microns to five nanometers. Printing lines that small requires extreme ultraviolet light, absorbed even by air, so machines run in a vacuum and fire lasers twice at falling drops of molten tin to make plasma, then bounce the light off ultra-flat mirrors. A modern fab costs over $12 billion, and Rock's Law says that cost doubles every four years. Chips are doped on 300 millimeter wafers, probed, cut into dies and packaged, while the industry splits into manufacturers like Intel and Samsung, fabless designers like Nvidia, and foundries like TSMC.

• Clean room air is filtered thousands of times cleaner than a hospital operating room, since one skin flake can wreck millions of transistors.
• Packaging moved from dual inline packages with centipede legs to ball grid arrays with solder balls under the chip.
• Rampant 1980s reverse engineering led to the Semiconductor Chip Protection Act of 1984, protecting photomasks.
• Designers hide microscopic chip art, such as animals, jokes and their names, on unused edges of the silicon.
• 3D ICs stack chiplets vertically so signals travel shorter distances, cutting latency and heat.