AI adoption in EDA

 

EDA and Semiconductor industry need to design faster and make design better. A fair amount of hype, surrounds what artificial intelligence (AI) can do for the EDA industry. AI can start designing, verifying, and implementing chip for us. But should AI replace the algorithms in use today, or does it have different role to play? However, it's a win, if AI do something better or make the person more productive. AI has data-hungry techniques. The EDA veterans's opinion about AI is that, once we have more data, and once we pay attention to the quality of data, then sooner or later we will get smarter AI models. 

If AI model is trained based on current design only, then AI’s knowledge is very limited. With another design having slight modification, the previous AI model can be reused. Likewise, some verification engineers are pretty much experienced and they can transfer what they learned from previous projects to a new project. AI can do mistakes because the training data or the model and the solution is new and unproven.  But, AI will give the same answer, right or wrong, consistently which humans don’t do. Once it is proven that the model in AI is accurate enough for a task, it can be assured that it’ll be accurate enough from then on. The problem with the human is that, if people are trained, and they’re accurate enough for the first design, they can still do mistakes for the second or third designs. Also, in future, AI-optimized designs could look drastically different from those created by human engineers. This poses new challenges in downstream tasks of verification and testing with humans in the loop. Generally designs created by engineers are modular in part due to how a project gets broken down to be effectively tackled by a team.