Quality Control for Chips

Estimated Reading Time: 7 minutes

TL;DR

Today’s issue starts by discussing a few recent news items of critical national importance: first the disturbing shooting in Atlanta and the rise of violence targeted at Asian Americans; and next, the signals coming from Chinese leadership that a potential invasion of Taiwan is being actively planned. The rest of the issue focuses on techniques for quality control in modern chip manufacturing such as automated wafer testing, scanning electron microscopy, and atomic force microscopy. We end by discussing the new trend of applying deep learning for quality control in semiconductor manufacturing.

The Shooting in Atlanta

The disturbing shooting in Atlanta on Tuesday, allegedly by a 21 year old white man, with 8 people dead, 6 of them women of Asian descent, has raised serious fears of racially motivated violence targeting Asian Americans. Heartbreaking stories have come out on Twitter by Asian Americans detailing the racism and assault that they have faced in the US. Read some of these stories if you haven’t already:

America is stronger when we overcome racism and hate. #StopAsianHate. Our previous issue made a few suggestions for steps you can take such as supporting your local Asian restaurant or donating to groups working to stop the hate such as Asian Americans Advancing Justice, AAPI Women Lead, or Stop AAPI Hate.

Aggressive Signaling from CCP Leadership

A recent report from Radio Free Asia notes that Xi Jinping is directing CCP armed forces to get “combat ready,” raising fears of a Chinese invasion attempt of Taiwan.

A hard question for the US is what happens if the Chinese armed forces do indeed launch an invasion? The US has taken some steps to build alliances with the Quad (America, India, Japan, Australia), but as of now, there are no mutual defense agreements as with NATO. India has recently suffered bruises at its border with China and may not be keen to jump into an external dispute, and Japan primarily relies on the US defense shield for its own defenses. Australia, as a smaller power, may not be able to contribute as much directly to the defense of Taiwan.

Historically, the United States has clashed with Communist China directly in the Korean war and to a lesser degree indirectly in Vietnam. During both of these wars, China was much less technologically developed than the US. Today, China is at near technological parity with the United States, and has the advantage that Taiwan is very close to the motherland, making supply chains and logistics much easier. China’s much larger population allows its armed forces to draw on a larger supply of potential recruits as well.

The general populace in the US hasn’t yet grappled with the seriousness of the threat to Taiwan from China. The fall of Taiwan could shatter US alliances in Asia, and lead to dramatic and crippling shortages of chips, potentially bringing the US economy to a grinding halt. Taiwan’s citizens, like Hong Kong’s, will lose their democracy, and the victory could herald the beginning of a Chinese century. The United States needs to strengthen defensive arrangements around Taiwan by building up Taiwan’s military defenses, and partnering with allies in the Quad and in Europe to ensure the security of Taiwan.

Testing Manufactured Wafers

In previous issues we have introduced the steps of semiconductor manufacturing such as front-end-of-line and back-end-of-line manufacturing, but we haven’t yet discussed how manufactured wafers are tested for correctness. Modern chips are extremely complicated, with billions of nanoscale structures. Manual testing can’t hope to catch all the possible errors that can arise in the manufacturing process. Manufacturers have developed sophisticated strategies to catch potential manufacturing errors. Wafer probers, like the one in the image below, can help perform automated testing of electrical contacts using tiny probes which are pressed against the wafer. Note that a large enough wafer by necessity will have some failing circuits. Typically, failing circuits are noted electronically in a file called a “wafermap,” which is used later in the process to package only functional circuits.

A wafer prober is an automated test machine used to test integrated circuits. The machine above is a 8 inch wafer prober with cover panels and tester removed (in a service configuration). Source.

Metrology

Metrology broadly uses scanning techniques to look at the actual nanostructures fabricated onto a chip source. These techniques, which include things like atomic force microscopy or scanning electron microscopes can be used to study the structures on wafers at the nanoscale. Techniques such as scatterometry can be used to inspect the surface by studying how it scatters light. The diagram below illustrates a number of different metrology techniques that can be used to understand a chip.

There are a number of different measurements which can be taken on the wafer. These include microscopy and scatterometry experiments in addition to virtual metrology and deep learning techniques (source).

Deep Learning Techniques for Quality Control

The diagram above hints at virtual metrology techniques which can be used to characterize the wafer. Such methods use raw signals from manufacturing machines to predict wafer quality without any experimental metrology experiments. Virtual metrology can also be used for quality control after etching steps. ASML has similarly developed deep learning methods for photolithography overlay to help alignment between lithography steps.

Deep learning methods can be paired with microscopy to make metrology techniques more capable. One challenge for researchers in the field is that datasets are hard to come by; increased access to open datasets will enable further advances in virtual metrology. At present, deep learning techniques in foundries are primarily limited to quality control (source). Future work could try using techniques like reinforcement learning for full foundry management. Foundry managers should partner more closely with academic institutions and startups to explore these newer possibilities.

Highlights for the Week

Subscription, Feedback and Comments

If you liked this post, please consider subscribing! We have weekly subscriber-only posts on Tuesdays.


Please feel email me directly (bharath@deepforestsci.com) with your feedback and comments! In particular, if you’re currently working in the semiconductor industry, please get in touch! I’d love your input for future iterations in our semiconductor series.

About

Deep Into the Forest is a newsletter by Deep Forest Sciences, Inc. We’re a deep tech R&D company specializing in the use of AI for deep tech development. We do technical consulting and joint development partnerships with deep tech firms. Get in touch with us at partnerships@deepforestsci.com! We’re always welcome to new ideas!

Credits

Author: Bharath Ramsundar, Ph.D.
Editor: Sandya Subramanian