The history of semiconductor manufacturing equipment and data usage

It is no exaggeration to say semiconductor manufacturing equipment and data usage have evolved hand in hand.
Now, each unit of semiconductor manufacturing equipment has hundreds of sensors attached to it, and the data from these sensors are constantly monitored and analyzed to further improve the process accuracy and production efficiency. This process is similar to how the Internet of Things (IoT) is supposed to work: the IoT requires sensors to be attached to everything, and massive amounts of sensor data need to be gathered and analyzed to generate value.
In that sense, it may be claimed that semiconductor manufacturing equipment is the forerunner of the IoT.

Use cases of AI on the semiconductor manufacturing industry

The semiconductor manufacturing industry has a long history of utilizing data in its operations. For example, the industry has been using virtual metrology (methods for predicting the virtual processing outcome based on sensor data) and FDC (fault detection & classification) for nearly 20 years.
Semiconductor manufacturing is mainly divided into three phases; namely, development, ramp up, and volume production. The use of data is essential to each of these phases.
To reduce time to market for new products, semiconductor manufacturers must put priority on shortening development turnaround time and ramp-up period, and on attaining stable product quality in the volume production phase. With regard to state-of-the-art semiconductor manufacturing equipment, in particular, customers are demanding greater use of data science to stabilize chip quality, since process control is becoming increasingly complex and difficult.

Uniqueness of TEL’s data science

Building on its expertise and experience with semiconductor manufacturing equipment, Tokyo Electron Limited (TEL) is now seeking to integrate its full capabilities in such areas as sensor technology, data science (including AI), and data engineering (i.e., handling of massive and diverse data), in order to identify and solve our customers’ real problems.
As new technologies are continually proposed and demonstrated, new AI algorithms are also being put forward on a daily basis. If we are to choose, combine, or invent optimum algorithms for any semiconductor manufacturing equipment, we need to fully understand our customers’ problems as well as know everything about the equipment.
For example, TEL engineers have drawn upon their accumulated knowledge of the equipment to focus on the sensor data that indicate the plasma states within the equipment. The team is now customizing the latest machine learning algorithms by taking the sensor characteristics into account, in the hope of achieving new virtual metrology that is more accurate than a conventional method that relies on simple use of generic algorithms.


The semiconductor manufacturing equipment industry has been using data science from early on to solve customers’ problems.
As the development of semiconductor manufacturing processes becomes increasingly complex and difficult, the role of data science will only become more critical.

TEL’s broad product line has given us a unique set of capabilities to solve customers’ problems through the intense use of data science. We intend to deliver our customers ever greater value by proposing comprehensive solutions.