13.3. Example: using PCA to reduce the dimensionality of semiconductor manufacturing data

Semiconductors are made in some of the most high-tech factories on the planet. The factories or fabrications (fabs) cost billions of dollars and take an army to operate. The fab is only modern for a few years, after which it needs to be replaced. The processing time for a single integrated circuit takes more than a month. With a finite lifetime and a huge cost to operate, every second in the fab is extremely valuable. If there’s some flaw in the manufacturing process, we need to know as soon as possible, so that precious time isn’t spent processing a flawed product.

Some common engineering solutions find failed products, such as test early and test often. But some defects slip through. If machine learning techniques can be used to further reduce errors, it will save the manufacturer a lot of money.