“For decades, autonomous chip design existed only in science fiction.…This pivotal moment in semiconductor history will breathe new life into Moore’s law.”
That’s Aart de Geus, chairman and co-CEO of Synopsys Inc., a provider of electronic design automation and semiconductor IP solutions headquartered in Mountain View, California, speaking about the company’s AI-based design system. In November it was announced that the system, Design Space Optimization AI (DSO.ai), has been used by Samsung to complete a state-of-the-art, high-performance design at an advanced process technology.
According to the company, DSO.ai uses reinforcement learning — an AI technology like that used in self-driving vehicles — to achieve better performance, power and area (PPA). Using the Synopsys Fusion Compiler RTL-to-GDSII solution, the system was applied at every stage of Samsung’s design implementation, resulting in significant time savings, reduced power consumption and an operating frequency over 100 MHz beyond target.
Traditional design-space exploration has been a very labor-intensive effort, Synopsys noted. It typically requires months of experimentation, guided by past experiences and institutional knowledge. The novel approach offered by DSO.ai, enabled by advancements in AI and machine-learning, searches autonomously for optimal solutions within the problem spaces of chip design. That means exploring choices in chip design workflows on a massive scale, while automating a high volume of less consequential decisions. The system can arrive at solutions that translate into faster software performance, extended hours of battery life and even more personalized user experiences.
As the driver behind the technology, reinforcement learning (RL) is focused on learning optimal behavior in an environment to obtain maximum reward. In short, it’s the science of decision-making: Behavior is learned through interactions with the environment and observations of its response, not unlike children exploring the world around them and learning the actions that help them achieve a goal.
Samsung has been an early development partner for Synopsys’ autonomous design technology. Launched in early 2020, DSO.ai started to be deployed into Samsung projects by fall of that same year. The system also received a World Electronics Achievement Award (WEAA) as “Innovative Product of the Year.”
