For decades, semiconductor fabs tolerated—and even expected—bad yields. Less than 50%, sometimes as low as 10%. Credible die yield data is rare, but here’s a market analysis from the late 1980s for the US Trade Representative during the US-Japan semiconductor dispute. It implies that when American and Japanese firms began 16K DRAM in 1978, yields were about 2%. Laughably low, but expected. And they stayed low. By 1984, US 16K DRAM yield hit 36%. Japan: 48%—better, still bad. Six years later! Now look at TSMC’s N5/N4 node, about 4–5 years old. Trade secrets, but N4 yields are around 80%. What’s going on? 6-micron process vs. 4-nanometer. The latter is far harder. Yet we’re hitting 80–90% yields. Is it a conspiracy? No—the fabs opened their third eyes. With new tools, they began inspecting and improving. In this video: how automated inspection tools revolutionized chip fabrication.

Get all episodes of Asianometry, Sharp Tech, Sharp China, Stratechery Updates and Interviews, Greatest of All Talk, and Dithering as part of Stratechery Plus for $15/month or $150/year.