Subscribe to our monthly newsletter for thought-provoking insights on emerging trends.

From the widespread ownership of smartphones to the rise of AI, 5G and self-driving cars, the signs are all clear that the world is moving deeper into the digitally connected age. And semiconductors are the foundation it’s all built on.

Given the rapid growth of the digital infrastructure, it’s not surprising that the companies that manufacture semiconductors — those at the base level of the digital ecosystem — face an ever-evolving set of competitive challenges. Some are familiar, like the relentless pressure to improve chip performance. But today, chip makers also need to navigate a shifting set of competitive requirements that have a lot to do with the processes they follow, beginning with design.

In the realm of chip manufacturing, packaging — how the parts of a semiconductor are brought together for use in a device — has always been fundamental. Today, however, the issue of packaging has become a major competitive differentiator that affects the power, performance and functionality of chips, not to mention their cost.

So what does it mean for processes? With packaging structures becoming more complex, the entire cycle — from development to prototyping to testing — has lengthened considerably. In the manufacturing process, getting from design to finished product requires the right set of highly detailed instructions around more than 100 parameters, with little to no margin for error.

Under conventional methods, chip makers are forced to make a series of incremental adjustments to identify the optimal “recipe.” In addition to the huge burden on the “cooks” in this equation — the development engineers — the by-product of this trial-and-error approach is increased waste and lower manufacturing yields.

Reduces the development cycle by as much as

30%

by minimizing trial and error in recipe formulation

Reduces equipment maintenance costs for semiconductor manufacturing customers by

50%

through data-driven optimization

Companies like Panasonic Connect — through its Process Automation Business Division — exist to help semiconductor manufacturers navigate these process challenges, and optimize production processes and deliver high quality products. With a 30-year track record of providing chip manufacturers with specialized production equipment, Panasonic recognized it was in a prime position to help them adapt to new semiconductor packaging trends.

Of particular interest, notes Mitsuru Hiroshima, Director of the Semiconductor Process Business Group, was the opportunity to infuse advanced analytics into its equipment solutions to enable truly breakthrough results for its manufacturing customers. “The core of our vision was the idea that the combination of deep learning and automation could bring design and manufacturing operations to a whole new level of optimization,” he says.

At the time, in 2019, Hiroshima and his team knew that realizing this vision — turning it into a concrete solution it could bring to market — would require the company to augment its core equipment competencies. “We were looking to collaborate with a [vendor] that could bring deep industry process expertise, along with a portfolio of advanced analytical technologies in areas like AI and deep learning,” Hiroshima explains. “IBM stood out as the only provider who could bring strength in both these critical domains.”

Panasonic technician working on a computer

Machine learning algorithms and the optimal recipe

The team IBM assembled for the project included AI and deep learning experts from IBM Research® and process experts and industry consultants from IBM Consulting™. In the first intensive months of the engagement, IBM and Panasonic teams collaborated to identify and refine solution opportunities. IBM applied IBM Garage™ methodology — bringing IT and operations people in iterative, high-impact collaboration — to set the tone for the collaboration, determine the overall objective and co-create solutions.

Based on the challenges and an assessment of fastest-time-to-value, the joint team co-defined two process control solutions that emerged as Panasonic’s first smart factory offerings. The first solution involved creating an advanced plasma dicer through fully automating the recipe generation.

Panasonic silicon wafer machine in production

Plasma packaging is a bit like magic. For an engineer trying to find the right plasma recipe, what needs to come out at the end is a wafer with precise cut patterns. That means making the right combination of decisions on variables like vacuum pressure and power, electron energy, ion energies and gases, to name a few.

To develop the proof-of-concept solution, the IBM Research team developed deep learning algorithms that — through huge numbers of calculations — enabled engineers to rapidly derive the optimal mix of variable points. “Rather than relying on intuition or trial and error,” Hiroshima explains, “engineers have an intuitive, visual interface [designed by IBM Consulting] that can simulate the process accurately and in just seconds.”

The second proof of concept co-created by the team addressed a different pain point: the need to optimize plasma cleaner machine performance through smarter, data-driven maintenance practices. “Rather than recipes, the plasma cleaning application uses advanced computation to identify the optimal time to perform cleaning and maintenance,” says Hiroshima. “Too early creates unnecessary costs, while too late risks poor quality and even machine breakdowns.”

Like the dicing solution, the machine status application is underpinned by algorithms developed by IBM Research. Using data from machine-mounted sensors, the application correlates changes in machine operating efficiency with the conditions of various machine parts. The highly intuitive, visual output, notes Hiroshima, is like a wish list for over-taxed technicians. “Technicians on the factory floor get an alert, for example, that one particular plasma cleaner out of many is performing at a suboptimal level, and that dirt on the electrode is the most likely reason,” he says. “This insight enables the technician to take corrective action that maintains high quality and minimizes production disruptions.”

On the path to autonomous factory production

To Hiroshi Benno, Manager of Product Marketing for the Process Automation Business Division and a key figure in the development effort, both applications demonstrate how powerful analytics on the shop floor have the potential to transform the way chips are designed and manufactured. “In the rigorous testing and simulations we ran, the plasma cutting solution reduced the development cycle by as much as 30%,” says Benno. “The shortened cycle reflects how AI-based analytics enables engineers to bypass much of the trial and error in formulating the optimal plasma recipe.” On top of that, AI-driven optimization also significantly reduced waste generated from the process.

As it was designed to do, Panasonic’s plasma cleaner application showed how machine learning insights can provide the foundation for an entirely new, data-driven approach to equipment maintenance decisions. Testing of the application demonstrated that it has the potential — through a combination of less unnecessary maintenance, proactive parts ordering and fewer machine outages — to reduce maintenance costs for manufacturing customers by 50%.

As Panasonic prepares to bring these new solutions to market, these metrics hold a potent message: that AI on the semiconductor shop floor is ready to deliver design and production optimization now. And that by adopting these practices, chip makers can better meet the increasing demands of today’s hyper-competitive global market.

To Hiroshima, Panasonic’s work with IBM thus far has also moved it decisively down the path to its long-term vision. “We’ve shown that by analyzing individual machine status data at the edge, we’ve created the basis for autonomous manufacturing equipment, in which the machine operates under the optimum conditions,” he explains. “The next step in this progression is to integrate multiple machines in the cloud to enable autonomous manufacturing at the factory-wide level. This group of highly autonomous machines — the autonomous factory — is the ultimate form of assistance that we strive for. In this way, collaboration with IBM that transcends the boundaries of a company can take a big step in that direction. We will innovate beyond existing ideas and processes.”

Panasonic Connect logo

About Panasonic Connect

Based in Osaka, Japan, Panasonic Connect (link resides outside of ibm.com) is a unit of Panasonic Holdings Corporation focused on digital transformation. The company’s Process Automation Business Division provides equipment, software and services to manufacturing customers around the world.



Solution components

IBM Consulting™
IBM Research®
 

© Copyright IBM Corporation 2022. IBM Corporation, New Orchard Road, Armonk, NY 10504

Produced in the United States of America, October 2022.

IBM, the IBM logo, ibm.com, IBM Consulting, and IBM Research are trademarks of International Business Machines Corp., registered in many jurisdictions worldwide. Other product and service names might be trademarks of IBM or other companies. A current list of IBM trademarks is available on the web at www.ibm.com/legal/copytrade.

This document is current as of the initial date of publication and may be changed by IBM at any time. Not all offerings are available in every country in which IBM operates.

The performance data and client examples cited are presented for illustrative purposes only. Actual performance results may vary depending on specific configurations and operating conditions. THE INFORMATION IN THIS DOCUMENT IS PROVIDED “AS IS” WITHOUT ANY WARRANTY, EXPRESS OR IMPLIED, INCLUDING WITHOUT ANY WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND ANY WARRANTY OR CONDITION OF NON-INFRINGEMENT. IBM products are warranted according to the terms and conditions of the agreements under which they are provided.