When I first joined SK hynix, one phrase on the wall of a meeting room stayed with me.
At the time, I thought that was simply what semiconductor manufacturing was. In an already highly optimized process, you keep looking for tiny differences, changing conditions little by little, and finding a slightly better result. It is almost literally like wringing an already dry towel again and again.
That is in fact how semiconductor manufacturing has advanced. Rather than changing everything at once, large numbers of engineers keep chasing small differences to the end, improving performance and yield step by step.
The difference between manufacturing and software is how knowledge accumulates
These days, when I look at CRM or marketing software, I often think of that phrase. Teams keep breaking down and improving details that look almost trivial: when to contact a customer again, which reaction should move a lead to the next stage, or which message works slightly better.
At first it can feel like overkill. But semiconductor manufacturing is similar. When most things are already done well, the organizations that keep finding the small differences all the way to the end move slightly ahead.
In CRM, a method that works well becomes part of the product or workflow. A small improvement discovered at one company can be used by another, and the resulting experience is fed back into the product again. As many companies solve recurring problems, the software itself gradually contains more and more of those methods.
In manufacturing, this happens much more locally.
When a problem occurs on a specific tool in a specific factory, a few engineers examine the data and look for the cause. They change conditions, check the result, and eventually find a better method. The knowledge earned through that effort often stays in someone's experience, an Excel file, or a report.
When a similar problem appears on another tool or process, someone else often has to solve a large part of it again.
That is why manufacturing's learning rate matters
Productivity is not determined only by how quickly one person solves one problem. It also depends on how far ahead the next person can start after that problem has been solved once.
If good decisions remain in a repeatable form and can be referenced again on other equipment and processes, the organization's starting point moves forward a little every time it solves a problem. If the problem-solving process is scattered across people and documents, every similar problem requires collecting data again, forming hypotheses again, and validating again.
If manufacturing is already an industry that improves down to the “last grain,” this difference can compound over time.
If AI is going to take on more judgment in manufacturing, collecting large amounts of historical data is not enough. To inform the next decision, the system needs to preserve what was observed in what situation, what decision was made, and what result followed.
Manufacturing software therefore needs to preserve not only analytical results, but the process of solving the problem. A method created once by an engineer should be able to become the starting point for the next process or tool.
Don't squeeze the same last grain twice
Semiconductor manufacturing has advanced by squeezing out every last bit of improvement. The effort to find even smaller differences will continue.
But the next productivity gap may not come only from squeezing harder.
This is also what Amously considers important when building manufacturing software. Rather than creating one answer for every process on behalf of engineers, we believe manufacturing's learning rate can rise more when the knowledge gained as engineers solve and operate problems remains in the system and can be reused for the next one.