Video by freeCodeCamp.org · taught by Kian Kyars · streamed from YouTube. All rights remain with the creator.
Lesson 1: How AI Chips Are Made
About this lesson
The chips that run modern AI are the product of one of the longest and most delicate supply chains in the world. This lesson follows a single advanced graphics processor along that chain, from raw material to a working rack in a data center, in six parts. It begins with semiconductor physics and scaling: how transistors have changed from flat designs to fin-shaped and then gate-all-around structures, why graphics processors run at lower clock speeds than ordinary processors, and what the end of a scaling rule meant for chip design. The second part covers design, showing how engineers' code is turned into the mask patterns used in manufacture by specialised software from a few suppliers. Part three goes inside fabrication plants and their cleanrooms, looking at wafer handling, ultra-pure materials and the statistical models used to predict how many chips on a wafer will work. Part four is about lithography and the machines that print features using extreme ultraviolet light with a wavelength of 13.5 nanometres. Part five turns to memory and advanced packaging, explaining how stacked high-bandwidth memory is joined to the processor. The final part looks at complete rack-scale systems, their large power demands, and the politics of the chain, including export restrictions and stages where a single supplier could halt everything. No engineering background is assumed. When you finish, read Chapter 1 of the course book, which explains the same journey with a transistor diagram, a yield graph and a worked cost example.