Transformative Potential of Generative AI: Alleviating Talent Shortages and Diversifying Chip Design

In the previous posts in this series, we read about the transformative power of generative AI (part 1) and handling possible errors and their impact on chip/system design (part 2). Now, in this third post, our esteemed panel discusses whether generative AI can address talent shortages and facilitate the creation of diverse chip types. The panel moderator, Bob O’Donnell, highlighted the need for more individuals with the required skills for electronic design development. He proposed to the panel whether generative AI-powered tools could potentially be a more accessible entry point for those interested in circuit, chip, or system design, given the volume of companies bringing chip design in-house and the need for more talent. He also asked the panel if these AI-driven tools could enable engineers to create different types of chips and lead to the development of new things.

Will generative AI alleviate talent shortages and facilitate the creation of diverse chip types?

Igor responded first, stating he believes that chip design applications will become simpler and faster in the near future, with a focus on enhancing existing designs rather than creating entirely new ones. He thinks that current generative AI technology might not be capable of facilitating radical changes to design. The potential for radical design changes might be limited due to the current availability of generative AI technologies. According to him, one compelling argument for streamlining the chip design process is the application of language models, which excel in handling sequences of steps or words. By repeatedly going through the same design process, these models can predict and optimize each step more effectively, leveraging their ability to discern subtleties in repetitive tasks. This is an area where human designers might struggle, but language models can excel.

Igor predicts that AI will play a crucial role in shaping the future of chip design. It is possible that we may witness a recursive cycle where AI chips are specifically engineered to support generative AI, empowering them to design even more advanced and efficient chips autonomously. This self-improving process will likely lead to the creation of better and more sophisticated AI chips over time.

Chris noted that, in the future, chips will see a significant shift in their integrated components. While end applications may mostly stay the same, the focus will shift toward algorithms for chip features. “We observe a positive feedback loop, with increasing AI expertise driving AI technology advancements,” he commented. Specialized computational fabrics aid this progress, transforming more problems into AI-related ones. As a result, the computational inference aspect of chips will likely favor ML acceleration, fostering innovation and success in various chip categories.

Paul added that AI has the potential to significantly enhance our productivity by 10X or a lot more than that. According to his viewpoint, two distinct worlds exist, one characterized by a fundamental productivity advantage of 10 times compared to the other. In this context, specific chips or technologies would either be impossible to develop without the remarkable productivity gains brought about by AI or, while technically feasible, would lack economic viability without the accompanying productivity improvements. Paul suggests that AI will catalyze the creation of novel chips, enabling their development for the above reasons.

Prabhal believes in the potential of unlocking a wide range of applications using generative AI that were previously financially and technically unattainable. He thinks there is real potential for a wide range of applications previously considered too expensive to develop or too complex for the workforce to handle. “We can conduct more experiments by accelerating the challenging aspects of the tool flow or those that require extensive experience. This will allow us to conduct interesting experiments, some of which could be worth billions while others may be total flops,” he concluded.

Rob said that as the industry strives for greater efficiency, we can explore new sources for answers and opportunities that were once unimaginable. For example, the intersection of 3D technology and engineering poses numerous challenges that human engineers find difficult to solve. He thinks that with the assistance of generative AI, we could discover solutions that we never would have considered before.

We have heard and seen stories about the hallucinations of large language models. So, what are the possible pitfalls of using generative AI in chip or system design? Read the fourth and final post of the series, Are There Pitfalls to Embracing Generative AI in Chip Design?, to learn more about it.

If you missed the chance to attend the AI Panel discussion at CadenceLIVE Silicon Valley 2023, don’t worry, you can register at the CadenceLIVE On-Demand site to watch it and all other presentations