#ResearchWednesdays
This research presents a new way to automatically find tiny imperfections on silicon wafers. These wafers need to be almost perfect, because microscopic flaws can ruin an individual silicon chip. Making computer chips is a delicate process, and even tiny imperfections on the silicon wafers they're built on can cause major problems. To help identify imperfections the researchers used a type of artificial intelligence (AI) called a "deep convolutional neural network," or DCNN for short.
Today we bring you a collaboration between two of our members, Seagate and Ulster University. Contributions from Cormac McAteer, Girijesh Prasad and M. M. Manujurul Islam.
This research presents a new way to automatically find these tiny imperfections. These wafers need to be almost perfect, because even microscopic flaws can ruin the chip. They used a type of artificial intelligence (AI) called a "deep convolutional neural network," or DCNN for short. Making computer chips is a delicate process, and even tiny imperfections on the silicon wafers they're built on can cause major problems.
They started with an existing AI model called LeNet-5 and improved it. This improved model has seven "layers," which are like different steps in the process of analyzing the image of the wafer. Some of these layers, called "convolutional" layers, are like sliding a magnifying glass over the image to find patterns. Other layers, called "pooling" layers, simplify the information to make it easier to process.
To make the AI even better at finding flaws, they used a technique called an "Adam optimizer." This is like fine-tuning a radio to get the clearest signal. It adjusts the AI's settings to improve its accuracy. The results were excellent – the AI correctly identified the flaws over 99% of the time. This means it's a very effective way to automatically find defects and improve the quality of computer chips.
Read more:
https://lnkd.in/gwwM4uWQ
