SK Hynix aims to make chip fabs fully autonomous 

SK Hynix said on Wednesday that the company’s ultimate aim for its semiconductor fabs are to make them fully autonomous.

Company manager Jeong Yu-in said its 300mm wafer fab were currently Level 3 to 4 in autonomy. SK Hynix will need to collaborate with the equipment industry and academia to get close to Level 5 autonomy, Jeong said.

Level 5 indicates full autonomy. Most newly made fabs have Level 3 to 4 autonomy where many processes are done automatically and without requiring operators to monitor them at all times. Measurement of yield rate and productivity is level or higher than that done by people. 

SK Hynix is aiming to reach Level 5 and is trying to make even minute processes autonomous and intelligent, the manager said.

The ultimate challenge is to have the fab make predictions, the manger said. Precise predictions can increase productivity and yield rate and reduce costs.

Predictive solutions based on artificial intelligence and digital twin technologies will grow important to make semiconductor processes autonomous. These technologies will solve various limitations in autonomous processes in the fab.

Some of these limitations include cue-time scheduling and route setting for overhead transports. Many variables that can happen during semiconductor production make it difficult to set time and route automatically, Jeong said.

Cue-time refers to the time when a wafer lot that finished a prior process moves to the next. Precise timing is important as it effects yield rate and quality. Research is being done that uses artificial intelligence to make scheduling autonomous but hasn’t yet yielded a satisfactory result, the manager said.

Digital twin technology is be used to develop autonomous technology to set the route of overhead transports. Simulations of the rail design, routing problem, equipment spending and chamber numbers for the OHT is done and applied to the real world. SK Hynix was also developing a simulator and is attempting to make them have high prediction rates.

Making scheduling and route predictions autonomous not only require equipment and hardware technology but software as well, Jeong said. Applicable AI research and simulation modeling, as well as personnel that can use these technologies is crucial, the manager said.