An artificial intelligence (AI) technology has been developed that can complete the design of high-performance communication semiconductor circuits in just one day, a task that previously took semiconductor design experts at least several weeks.
According to the Ulsan National Institute of Science and Technology (UNIST) on Wednesday, a research team led by Professor Yoon Hee-in of UNIST's Department of Electrical Engineering and Professor Song Dae-gun of Kyungpook National University has developed an AI model that automatically designs LC voltage-controlled oscillators (LC-VCO), a type of communication circuit, from the circuit design stage through to the physical layout implemented on the actual chip.
LC-VCOs are semiconductor circuits that generate frequencies in high-speed communication systems such as 5G. The model developed by the team integrates circuit diagram design and physical layout, which were previously optimized separately, under AI management. At the circuit design stage, reinforcement learning is applied to adjust design variables and find combinations that meet target frequency and performance requirements. At the layout stage, where the actual chip structure is determined, physical design variables such as wiring width and spacing are iteratively refined in the direction of improved performance.
Experimental results showed that a task that previously took approximately 119 hours using conventional automated design methods was completed in just 28.5 hours, reducing design time by more than 76%. "We can significantly lower design costs while enhancing the performance of frequency generation circuits, which are core components of 5G and 6G communications and AI chips," the joint research team said. "In the mid to long term, this is a tool that can address the shortage of semiconductor design personnel and dramatically accelerate the transition to next-generation processes."
