Editor's note: AI is doing the job humans can't – and it's paying off...
From steel to electronics and semiconductors, AI is reinventing how we spot and prevent defective products. In today's Masters Series, originally published in the March 17 issue of the free Chaikin PowerFeed daily e-letter, Joe Austin, senior analyst at our corporate affiliate Chaikin Analytics, shares the technology that companies are competing over to maintain quality control...
This Successful AI Technology Isn't a Chatbot
By Joe Austin, senior analyst, Chaikin Analytics
California Steel Industries' Hot Strip Mill in Fontana stretches more than half a mile long...
Inside, giant ovens heat steel slabs to about 2,300 degrees Fahrenheit. At that temperature, the steel gets soft enough to roll.
But first, the steel needs cleaning. The furnace leaves a thick crust of "scale" on the surface. If it isn't removed, it gets pressed into the steel and ruins the finish.
A scalebreaker cracks the crust loose. Then high-pressure water jets blast the crust away.
Next, the steel slab passes through five roughing stands that squeeze it down from between 7 and 9 inches thick to as much as 0.0538 inches thick. That's close to the thickness of a credit card.
Each stand uses four stacked rolls. Two smaller work rolls grip the steel directly. And two larger backup rolls drive the force.
After this part of the process, a crop shear trims the ragged ends of the steel before it moves to finishing. Even a slightly uneven edge can throw off the entire process.
The steel then moves through six finishing stands. These roll the steel to its final thickness and set its surface quality.
By now, the steel is moving at about 35 miles per hour. That's too fast to catch defects by eye.
For automotive panels and appliances, the surface needs to be flawless. Defects show right through the paint.
The finished strip winds into a coil. Some coils weigh up to 25 tons.
The whole process can take about five hours. At full capacity, the mill runs 24 hours a day and produces 2 million tons of steel per year.
But at least you can see steel...
In today's most advanced semiconductor fabrication plants (or "fabs"), the defects that matter are invisible to the human eye.
But the consequences of missing them are just as severe. And that means turning to machines to help with quality control...
In semiconductor manufacturing, everything starts with a wafer. These thin, polished discs are sliced from pure silicon. And they range in thickness. Most of them are about 12 inches in diameter. But many are smaller.
These wafers must be flawless. Even a microscopic scratch or contaminant can create defects across hundreds of chips.
The first step is circuit printing using extreme ultraviolet ("EUV") lithography. This process projects circuit patterns using light with a wavelength shorter than any visible color.
A single finished chip can require 20 to 30 passes through this stage.
EUV "masks" – a kind of specialized plate like a 3D stencil – have to be perfect, too. A single defect ruins every chip that the mask touches.
After each EUV pass, the wafer goes through etching, deposition, and chemical treatment to build up transistor layers. Then the cycle repeats.
Today's most complex chips go through a range of about 1,500 to 2,000 individual steps before they become functional.
Each step is a potential failure point. One particle of dust can ruin an entire wafer.
Once the wafer is complete, a diamond blade dices it into individual chips – or "dies." Each die gets tested electrically.
The ones that pass move on to packaging. And the ones that fail get discarded.
Increasingly, packaging means assembling multiple chips into single packages – called "chiplets." That adds another layer of complexity. And inspection expands beyond a single die.
That means checking how multiple dies connect and communicate.
But whether it's steel or semiconductors, human inspectors simply can't do the job...
At around 35 miles per hour, the steel moves too fast to see. In a semiconductor fab, the defects are too small to see.
And with today's most advanced chips, defect control isn't just a quality issue...
It's a competitive one.
A single wafer for some of the most advanced semiconductors can cost between $20,000 and $25,000. Each wafer holds hundreds of chips. So a defective one also wipes out hundreds of products.
The EUV masks cost up to $1 million each. And the fabs where all this happens can cost between $15 billion and $20 billion to build.
Of course, fabs are looking to reduce these costly losses wherever possible...
This is one example of where AI can shine. And even better, it's a job humans can't do well.
Like I said earlier, the human eye can't detect every defect in steel moving at 35 miles per hour. And it can't catch microscopic issues in semiconductors.
AI "deep learning" and "edge learning" take defect control to the next level...
Deep learning is a process where the system looks at hundreds of example images until it learns to make decisions on its own – without a programmer needed.
Edge learning takes it even further. These systems come pretrained. They might need just five to 10 images to get started. And they deploy in minutes.
Quality control with AI is more than just hype...
At carmaker BMW, AI-powered vision systems cut defect rates by 30% at one European plant within a year. Customer satisfaction jumped 15% after the rollout.
At electronics contract-manufacturer Foxconn, AI-powered cameras can catch defects with 98% accuracy... flag 80% fewer false alarms... and inspect each unit 60% faster.
Folks, we've heard plenty of debate over whether the massive investments in AI infrastructure will be profitable...
At the same time, it seems like we hear about new business applications for AI almost every day. Just think of the recent chaos in the software industry surrounding the "SaaSpocalypse."
To find the real winners amid the AI megatrend, look for companies using AI to solve real problems and improve their competitive positions – like by taking quality control to the next level.
Good investing,
Joe Austin
Editor's note: The AI boom isn't slowing down. However, two investing legends say a looming "Dark Chip" crisis could create massive losses for investors.
So on April 29, they're stepping forward to share the one opportunity that could not only protect you from the crisis, but also make you 12 times your money. Click here to learn more.
