Semiconductors, from sand to system

At its foundation, a chip repeats one device at enormous scale: the transistor, an electronic switch with no moving parts. Silicon is useful because it is a semiconductor; carefully added impurities, or dopants, allow engineers to control when it conducts. Billions of transistors arranged into logic produce computation.

The basic device is easier to describe than to manufacture. Modern transistors operate at dimensions where atomic-scale variation matters. Producing them requires cleanrooms, vacuum systems, ultrapure materials, precision optics, plasma chemistry, metrology, and process control working as one system. The industry calls the facility a fab. The short name hides the complexity.

Keep one mental model: chips are built layer by layer. Pattern, deposit, etch, measure, and repeat (dozens of times) until a three-dimensional electrical system emerges on a polished silicon wafer. The sequence is repetitive; the tolerances are not.

Every defect on a wafer kills whatever die it lands on. So yield falls roughly as e^−A·D₀ (die area times defect density) which makes die size the most expensive decision in chip design. Drag the sliders: watch a wafer of small dies shrug off defects, then grow the die toward the reticle limit and watch the same wafer bleed out. This one curve explains binning, chiplets, and why AI-scale dies cost what they cost.

The structure matters because it concentrates: one dominant EUV maker, a handful of leading-edge foundries, three EDA firms, a short list of resist and wafer suppliers. Semiconductor geopolitics is mostly the map of these bottlenecks, which is why chips are treated as strategic infrastructure and why fabs anchor the industrial policy of every major economy.