Amitron's Leap Into the AI Frontier

Estimated reading time: 2 minutes

In a conversation with Aidan Salvi, Amitron’s chief transformation officer, he spoke of the interaction of machine learning on registration. Amitron has been modernizing much of its manufacturing equipment, and Aidan points out that improving registration is a key objective. He sees registration as a holistic system. To make smart improvements, you need data. To get data, you need equipment and sensors which capture the data. To make sense of the data, you need analysis and, eventually, predictive tools.

Nolan Johnson: Aidan, you’re leading a transformation at Amitron, and you’ve shared that registration improvement is one of the objectives of your modernization. Where do you see machine learning and AI contributing?

Aidan Salvi: There are steps to making AI impactful. It all starts with gathering the data. We operate our process with nearly 60 different pieces of equipment that do different functions. When looking at what the future of AI could be, the biggest challenge is how we consolidate that data. How do we create standards? How do we create scalable hardware that can store and retrieve the data? This is a tremendous amount of data; companies need to look at their data infrastructure and assess how to integrate and pull it all together. That is the first hurdle I see for AI in the future.

Johnson: When it comes to bringing AI into the PCB factory, is this a hardware processing horsepower and data storage challenge, or maybe one with software tools or people and skills? Where do you start using machine learning and AI appropriately for registration?

Salvi: There are at least three things I see as the cornerstones of this challenge. First, it's integration with the equipment. You must deal with quite a few manufacturers to extract the necessary data out of the equipment in a consistent way, and that doesn't exist in our industry as well as it could. Second, it's not necessarily computing horsepower, but more of an infrastructure design issue. We're dealing with quite a bit of real-time data that needs to be transported over networks, stored, and processed. In other words, it’s designing infrastructure that can handle the future needs of AI computing. The last piece is cloud integration, where much of AI is built. That starts moving us away from the physical component, and into looking at how we can start processing the data at scale and building data systems in a virtual cloud architecture. These are challenging things to put in place because of the overall cost and knowledge gaps that exist.

To read this entire conversation, which appeared in the November 2023 issue of PCB007 Magazine, click here.