In order for electricity to conduct, electrons need to move from lower energy states to higher ones. Conductors are materials that have very little gaps between low and high energy states and thus, currents can flow pretty easily (e.g. metal). Insulators have larger gaps and therefore need a significant amount of energy to conduct current, often damaging the material (e.g. glass or rubber). Semiconductors are materials that can conduct electricity depending on its environment (hence “semi-conductor”). They can switch between conducting and insulating through changes in electric fields. The ability to control the electric currents at a quick scale is what makes semiconductors very valuable; this allows the chip to efficiently process, store, and transmit binary information. The semiconductor material that is most used for electronics is silicon (which is where Silicon Valley gets its name!).
To manipulate conductivity, there is a process called doping, where impurities are added to the silicon to create n-type (i.e. negative charges moving in between the atoms) and p-type (i.e. positive charges moving in between the atoms) semiconductors, which are laced together to create a transistor.
Transistors are devices of small semiconductors and other components used to amplify signals or switch power. They produce binary code that provide logic, communication, and memory components. A very, very simplified diagram of a transistor (specifically a field-effect transistor) is shown below.
The gate controls whether the semiconductor will conduct currents, which will signal 0s and 1s in computing. Billions of transistors are organized into circuits to create integrated circuits, or more commonly referred to as the chip (moving forward, semiconductors, integrated circuits, and chips will be used synonymously).
There are several different types of chips that can be categorized by their function:
There is also a distinction between leading edge vs mature/legacy nodes. Leading edge chips are more expensive, require advanced technologies that prioritize speed and efficiency, and are tiny in size (typically 5 nanometers or less). Mature or legacy chips use an older technology and are much cheaper to manufacture; the Chips and Science Act of 2022 defines these chips as being 28 nm or larger. The former usually function as logic or memory chips, and the latter are usually analog, MCUs, and power chips.
Draft: AI Accelerator Stack
Chips are manufactured on wafers, which are extremely thin slices of purified silicon, and are coated in photoresist, which makes them sensitive to light. A lithography machine will etch the chip designs, which are produced by engineers, on these wafers with a very precise light (much like a stencil). Multiple circuits are typically printed on a single wafer, and each one of these circuits are referred to as a die. Other materials are laid into these etches to build the transistors. Lastly, all the transistors are connected via microscopic wiring, known as interconnects.
Each die is then tested before being cut from the wafer. When everything works and the die is cut from the wafer, each die needs to be packaged so that it becomes less fragile and can connect to a broader circuit board.
The semiconductor value chain, which embodies all the processes outlined in the “How are they made?” section, consists of three broad components: design, fabrication, and assembly, packaging, and testing. These components are typically decentralized, but there are integrated device manufacturers, like Intel or Samsung, that are vertically integrated for the whole chip-making process. These are legacy firms and have fallen out of popularity given the high level of specialization at each process in the value chain. Around the 1980s, the vertically integrated business model gave way to the modern version of the supply chain when Taiwan Semiconductor Manufacturing Company (TSMC) developed the first dedicated semiconductor foundry.