All About Circuits is one of the world’s largest and most active independent online communities for electrical engineers. Ingrid Fadelli, All About Circuits contributor, recently interviewed Simon Floyd, industry Director of manufacturing and transportation at Google Cloud, and Jason Ruppert, COO at Phononic, who explained how machine learning is revolutionizing semiconductor factory floors.
Two months ago, Phononic implemented Google Cloud’s new cloud-based tool for machine learning. Google Cloud’s Industry 4.0 solution for manufacturing operations is set to make long-term improvements in semiconductor production and supply chains.
Factory floor equipment
Phononic’s key partnership with Google Cloud is poised to improve productivity, yield, and ROI in semiconductor fabs. Google Cloud’s newest cloud-based tool was designed to connect factory floor equipment to the cloud and provide monitoring, analytics, and AI/ML insights.
Our goal here is to allow people to connect their products and their factory so they can have a single view"
Among the many tools possible through Google Cloud’s new solution, some of the most notable include predictive maintenance, anomaly detection, and vision capabilities. “Our goal here is to allow people to connect their products and their factory so they can have a single view—a single pane of glass end to end—from how something is manufactured to how it operates in the field, and be able to learn from that,” Floyd said.
Hybrid cloud solution
On a lower level, the tool is entirely software-based (i.e., Google Cloud doesn’t provide any hardware), and consists of two major components. The first component, Manufacturing Connect, physically connects factory equipment to the cloud. To make the solution as interoperable as possible, Google Cloud teamed up with Litmus Automation to equip Manufacturing Connect with 250 different types of machine drivers for connection.
The other solution is called Manufacturing Data Engine, the cloud component that performs all the data storage, processing, and analytics. This solution can be deployed as a hybrid cloud solution, where users can select which components to run on the cloud and which to run locally, adding flexibility and customization to the mix.
Thermal electric technology
To be a hybrid cloud, some components run directly on hardware so they can be installed in the factory"
“To be a hybrid cloud, some components run directly on hardware so they can be installed in the factory. Then, other components go to the cloud,” Floyd explained. “The way we balance that is the difference between the scale of data processing and the speed of data acquisition. From here we can make the decision as to where something should run or operate.”
So far, this tool has been used by a select group of companies, including Phononic, a manufacturer of thermal electric technology. Phononic has put Manufacturing Connect and Manufacturing Data Engine to work on its factory floor for two months. “There are a lot of complex processes that we want to be able to see in real-time,” Ruppert said. “For something like a wet etch or a plating process, there are variables like PH balances in assembly. There are pressures on machinery. Whatever the case may be, we want to be able to see those processes in real-time.”
Providing predictive maintenance
With the new tools from Google Cloud, Ruppert said Phononic has turned monitoring and analysis into insights. For example, the solution can provide predictive maintenance, allowing Phononic to shut down and repair a machine before it becomes damaged beyond repair.
There is a significant ROI with just looking at this one part of our fab"
Speaking about the benefits that Phononic has actualized through Google Cloud’s solution, Ruppert noted, “There is a significant ROI with just looking at this one part of our fab. Beyond that, we have a clear line of sight that our yield and throughput in this one particular area are going to improve dramatically.”
Damaged beyond repair
“The worst job in a factory is where you’re the substitute for a machine. You have to do monotonous things by hand, and it’s kind of degrading to the human race in a way,” Floyd added.
“We’d like to give [operators] better instructions for how to perform a function, so that they’re adding their own value in the right way. We want them to be the human in the loop. When it comes down to a machine not quite understanding whether something is good, bad, or indifferent, a person can help the AI be trained using that human knowledge.”