Technical University of Denmark: Turns up digitalization to turn down defects

When a high-speed train speeds by or a wind turbine is connected, this makes great demands on the electronic components. So-called semiconductors constitute an important part of these components. These are materials that have electrical conductivity in the area between metal and perfect insulators like porcelain. If the semiconductors do not meet very precise requirements, a short circuit will occur when the voltage is measured in kilovolts, and this is both expensive and dangerous. Therefore, the produced semiconductor—for example silicon—must be ultra-pure, so that its conductivity is accurately defined.

And that is precisely what Topsil can do: produce the world’s purest silicon, which is the main component in diodes, transistors, and other microchips which form the basis of all electronic circuits. The silicon rods that Topsil produces are cut into paper-thin discs that have completely uniform electrical properties throughout their surface.

“It’s the purest solid on the planet. And that’s a requirement when producing computer chips for heavy current,” explains Topsil’s Head of Production, Christian Hindrichsen.

One crystal becomes two metres long
Topsil produces its silicon through a process known as float zone. Here, they melt the raw product—amorphous silicon—and, through cooling, they make a single silicon crystal grow into a rod that is two metres long and weighs 100 kg, without the molten silicon coming into contact at any point with the sides of the steel container in which the process takes place. And this is absolutely essential, the Head of Production points out:

“If the molten silicon comes in contact with other materials, it will be contaminated.”

The float zone technique is an extremely complicated process, and only five companies in the world are capable of manufacturing silicon semiconductors on a large scale using this technique. This makes Topsil part of a small elite. Less expensive manufacturing methods can also be used to manufacture silicon semiconductors, but they result in a more impure product that cannot be used to conduct heavy current in wind turbines and other high-voltage equipment.

However, the manufacturing process is so complex that it is inevitable that defects will occur in the products, which must therefore be discarded.

Camera-monitored production
“When you make a product without defects, you avoid waste. A manufacturing process optimized to be defect-free also uses less energy. This makes zero-defect production more sustainable.”
Associate Professor Guido Tosello, DTU Mechanical Engineering
To reduce these defects, Topsil has increased the digitalization of its production in the past ten years. And Topsil collects huge data volumes when it produces silicon 24-7.

The entire production of the silicon crystal is monitored by cameras, and data from the cameras are used to regulate the process, depending on what the images show. But a much higher volume of data can be extracted from the images. Topsil therefore asked itself whether these data could not provide even more information to the benefit of the production.

“Can we retrieve further information? Maybe about the purity of the raw materials, the purity of the crystal or fluctuations in the process,” elaborates Christian Hindrichsen.

Advanced image analysis
And this is where DTU comes into the picture. Topsil approached Senior Researcher Matteo Calaon from DTU Mechanical Engineering. They decided to start a research project in which—through pattern recognition in large data volumes—defects are discovered at an early stage and the process is automatically corrected to eliminate these defects. The research project is being conducted as a PhD study by Tingtin Cheng, who makes advanced analysis of images from the camera monitoring of the silicon production.

Tingtin Cheng uses machine learning (a branch of artificial intelligence, ed.) to analyse the large data volumes. Based on a limited number of images of the crystal production in which there have been various defects, she performs advanced image analysis and creates the computer codes which describe the defect signatures in these image data. The codes are then used to train a computer in how to analyse many thousands of images. The computer subsequently divides the images into groups according to the type of computer-recorded production defects. When these data are correlated with other production data, Topsil’s engineers may be able to see some patterns. This may be specific machines or specific suppliers that are typically linked to the occurrence of defects, or it may be that the defects occur at specific times. Topsil can then adjust its production to minimize the defects. The project is currently at this stage.

The next step will be to perform real-time analysis of the images and have computers automatically correct the process during production. In this way, Topsil will be able to stop or adjust a defective production before it has been completed and thus reduce the defect rate.

Digiman4.0
The research project is part of the EU-funded research and education programme Digiman4.0. The name of the programme stands for digital manufacturing, and the figure 4 refers to the concept of industry 4.0—the fourth industrial revolution—in which digitalization plays a crucial role.

“Digiman4.0 is a large training and research programme aimed at the manufacturing industry. The goal is to create experts in integration of digital processes in manufacturing production,” says Guido Tosello, Associate Professor at DTU Mechanical Engineering as well as head and initiator of Digiman4.0.

Fewer defects increase sustainability
Digital manufacturing is not an unknown concept in the manufacturing industry, but the technology has primarily been embraced by the major industrial giants. Small and medium-sized enterprises are lagging behind, and these are the enterprises at which Digiman4.0 is aimed. The goal of digitalization is to improve manufacturing processes and create better products. The ultimate ambitious goal is to achieve zero-defect production.

“The zero-defect goal is a new paradigm in the EU. It is closely linked to the sustainability agenda. When you make a product without defects, you avoid waste. A manufacturing process optimized to be defect-free also uses less energy. This makes zero-defect production more sustainable,” explains Guido Tosello and elaborates:

“The road towards defect-free production goes through analysis of existing process data in productions in which defects regularly occur. If we ascertain in the data that there are patterns that lead to defects, the researcher’s task is to correct the production through new digital processes to avoid the occurrence of the defects.”

This methodology is part of all the projects in Digiman4.0, and the goal under the programme is to define a procedure that can be used in general to achieve zero-defect production.

However, zero-defect production is not the goal for Topsil at this stage. But it is realistic to reduce the defects by several per cent, finds Christian Hindrichsen. The Head of Production especially has faith in the planned automated correction of the production via real-time analysis of the images. Here, digitalization will really make an impact in the production of the purest solid-state material in the world.