How Will EDA Benefit from the AI Revolution?

With the rise in demands for instant gratification, high performance, and smart everything, we are witnessing how AI inclusion in almost every industry revolutionizes productivity, design, and performance. The EDA (Electronic Design Automation) industry is no different; using AI (artificial intelligence) across the EDA design suite is the solution to improve performance and productivity. During the recent CadenceLIVE’23 Europe, I got a chance to witness a panel discussion with an exciting title: How Will EDA Benefit from the AI Revolution? 

The panel was moderated by Rosa Markarian and composed of renowned experts from industry and academia, as mentioned below:

• Rod Metcalfe from Cadence
• Jean-Christophe Glas from Arm
• Hussam Amrouch from TU Munich
• Hammond Pearce from the University of New South Wales in Sydney (UNSW Sydney)

In the hour-long session, the experts delved into the various aspects of applying generative AI in EDA in front of a packed audience. The moderator, Rosa Markarian, confidently stated that EDA is one of the few industries that doesn't see Generative AI as a job killer but rather as a welcoming solution to boost the productivity of engineers and combat the widening engineer gap. Nevertheless, many issues must be resolved, including copyright issues of the created code or hallucination of GPT systems. Apart from this, there were many intriguing discussion topics, such as

• AI, its impact on jobs, and is AI a job killer in the EDA industry?
• How AI is helping to bridge the gap and meet the shortage of skills in the semiconductor industry
• Where are we on Gartner’s research cycle?
• Challenges in verification and increasing cost
• The impact on academia

and many more that I am going to reveal in this blog series. The panel discussion started with an interesting discussion about the impact of generative AI on chip design.

Rosa Markarian: How can generative AI bring chip design to the next level? And to what extent do you already use it?

Hussam: Machine learning has been around for many years and has always been a powerful tool for optimization. With its ability to learn from vast amounts of data, it has been extensively used in analog and physical design. However, the real breakthrough in recent years has been the development of large language models. These models have the potential to revolutionize how we communicate and collaborate. With AI-powered chat engines, designers can now have a virtual assistant to help them write better code and solve bugs. Imagine having an AI assistant that can understand the nuances of language and help you communicate more effectively with your team. The possibilities are endless. With the vast amount of chat data accumulated over the years, the AI engines can learn from real-world conversations and provide better insights and recommendations.

Rod: I find the optimization aspect of design exciting. The challenge from the perspective of EDA is that, before the advent of machine learning and other key technologies, we had to create a single product that could address every design in the world. This was a challenging task to accomplish. However, with the emergence of AI and large language models, we can now learn about the design being created and then customize the EDA tools to that specific design. This is a significant breakthrough that has the potential to revolutionize the industry. The ability to learn through EDA is going to be very powerful. Some technology companies like Cadence have been developing optimization-based AI for a while and have already established some well-known products in that area. However, generative AI and large language models are another breakthrough we can use well. The user interface, being able to interact with the tools in a language-based way, is a big breakthrough. But the ability to generate design data is also a key breakthrough. Overall, a lot of work has already been done, especially in the optimization area with AI. Cadence has invested in that for many years. However, the generative AI stuff is very new, and there's enormous potential for EDA. So, I think it has its place.

The session became exciting when the moderator asked Jean-Christophe how collaboration with Cadence is helping them reap the AI benefits in EDA design.

Jean-Christophe mentioned that collaborating with Cadence has a unique perspective on the advantages of AI for hardware development. However, he emphasizes caution, as the industry is still in the early stages of the innovation cycle. Despite its up-and-coming technology, there is a lot of work to be done to prove the benefits of AI. Demonstrating credibility through value is key. While there is a lot of proof-of-concept work in the industry, there are no known actual products in development for hardware. Christophe hopes that the industry progresses much faster than the 10X revolution of 2013 and believes that proof of concept work is currently the main stage of development.

How do you think generative AI can help close the gap?

Jean Christophe agreed to the Hammond’s response. He mentioned this new energy could help engineers with a background in hardware design and education to do more and expand their capabilities. This, in turn, could help address the shortage of skilled people in the industry. However, I am still skeptical about whether students or people without the right expertise and background can design complex hardware like CPU or GPU. While some designs might seem simple, designing a complex process is not just about making it functional. For instance, generating a CPU with ChatGPT is not just about the functional design; it's about understanding the content that will run on it and the tradeoffs that need to be made for performance, area, and power. Although this idea is inspirational, I think it is more of a long-term goal than a short-term solution.

Rod: In the very long term, it would be great if we could use generative AI to design chips, but in the short term, I believe junior engineers can become much more productive by using generative AI and machine learning. For those of us who have been in the industry for a while, we learn a lot as time goes on. However, if a machine can help a junior engineer with a good understanding of chip design to become more productive in complex designs, it would be immediately beneficial. With the current shortage of engineers, the new engineers coming into the industry who can do more work more productively benefit AI.

“We went through a transition with RTL synthesis, and suddenly, we could generate a few hundred thousand gates in a few minutes. Did we have fewer engineers because of RTL synthesis? No, we have more engineers, and they are all more productive. We are at a similar inflection point with AI, and we are going to be more productive. That's great, but nobody got fired because of RTL design, and I don't think that's going to happen with generative AI either.”

One of the benefits of LLM technology is that it can help improve the quality of the code. There is a lot of bad RTL out there that we write in a hurry, and it's not well thought through. This code comes back to bite us eventually, and sometimes, bugs go unnoticed, regardless of how good our verification plan is. If we can use LLM technology to enhance the quality of the code and verification, it's immediately beneficial. Ultimately, LLM can help improve the quality of the code, productivity, and engineering resources.

Where are we on the Gartner hype cycle?

Hammond does not reply to that. Instead, he starts by talking about the advantages of AI for complex tasks. One of the main advantages of AI is that it makes complex tasks more accessible to non-experts. However, tools like ChatGPT and large language models can also abstract away the complexity of certain tasks, enabling more people to do them. While this can make talented engineers more productive, it can also make non-experts able to do tasks such as design, even if they previously lacked the expertise. For example, someone who has never worked with amber bus technology before can use AI to produce something using it. The result is that AI is making experts more productive and tasks more accessible to non-experts. When we talk about language models and their capabilities to write and interact with EDA tools, we can place them in the hype cycle. Although they still have a long way to go in improving, they have already come a long way in languages like Python. There are many proven areas for growth, particularly in the hardware.

During the discussion, Rosa brought up an interesting topic about the shortage of engineers in the semiconductor design industry. She pointed out that unlike most other industries, where jobs are being replaced by generative AI, the semiconductor industry lacks a skilled workforce. According to the Semiconductor Industry Association, by 2030, there will be a shortage of around 67,000 computer scientists, technicians, and engineers in the United States.

Can generative AI help to accelerate the learning curve to reduce the time to market? CAN Generative AI reduce the rising verification cost? What are the challenges associated with the usage of Generative AI in chip designing? What is the business perspective of using this? I will be covering all these burning questions in the next blog.

If you missed the chance to attend this panel discussion at CadenceLIVE Europe 2023, don’t worry. On demand videos are coming soon. You can register at the CadenceLIVE On-Demand site to watch it and all other presentations.

Learn more about using Generative AI to transform the world of product design from chips to products, drug discovery to life sciences,and specification to manufacturing. 

…to be continued…