AI for Semiconductor Chips 🔬

How AI can improve semiconductor manufacturing yield.

By: Rohan Saxena

TL;DR

Fortune 500 manufacturing company Applied Materials has created a new inspection system using AI and Big Data that can efficiently identify defects on semiconductor chips and save producers millions while improving their yield rates.

The Breakdown

• Manufacturing company Applied Materials released their New Enlight Optical Wafer Inspection System which leverages Big Data and AI to help manufacturers efficiently identify defects in their semiconductor chips
• The chip manufacturing industry is highly competitive and companies are constantly trying to create smaller and more efficient chips; a feat they achieve primarily by increasing the number of transistors on a chip
• In simple terms, transistors help the chip send signals back and forth; thus increasing the number of transistors makes the chip faster and more energy efficient. Currently, chips range from having 28 to 60 billion transistors on a single chip
• According to Moore’s Law, the transistor count on a microchip will double every two years, which begs the question; “How can companies manufacturing these miniscule chips know if there are any defects if they’re getting smaller and smaller?”

The Tech

• The Enlight system has been in development since 2016 and can provide high-resolution scans to detect any defects; it identifies these abnormalities by shining a light and measuring how it bounces off –if it has an abnormal reflection due to curvature or dents, it’s probably a defect
• Since the system is dependent on light reflections, it’s prone to random errors and potentially outputting false positives (falsely identified defects) which is where their SEMVision G7 system integration comes into play
• SEM stands for Scanning Electron Microscope and essentially uses an electron beam to generate a magnified image and filter out the false-positive defects that are used as training data for their AI system ExtractAI Technology
• Two of the key components of ExtractAI are its reliance on Computer Vision and Supervised Learning
• Supervised learning is when a machine learning model uses labelled training data to correctly classify data or predict outputs
• Computer Vision is a field of AI that aims to help computers see using various forms of image processing, with digital images and videos as training data
• ExtractAI leverages these machine learning principles and architecture created by Applied Material’s Data Scientists to correctly identify defects in real-time after only reviewing 0.001% of the sample pictures provided by the SEMVision G7

The Significance

• The Enlight System is predicted to reduce the cost of capturing defects by three times, while improving the effective yield; a win-win situation
• Yield is an important metric which describes the percentage of chips that are not scrapped due to “killer defects.” With an increasing demand for semiconductor chips, whichever manufacturer can scale their production efforts most cost-effectively will maximize their profit long-term
• To put this in perspective, a semiconductor manufacturing machine with a downtime of one week costs roughly $25 Million in unamortized depreciation costs, signifying the importance of improving yield rates for these machines

Additional Info

Applied Material’s Documentation

TSMC Semiconductor Chips

Manufacturing Semiconductor Chips (Intel)

Full Source List