In a Nasscom panel discussion on how AI is reshaping advanced chip design, driving efficiencies, and paving the way for groundbreaking innovation in the semiconductor industry, Jaya Jagadish, Country Head & Senior Vice President, AMD India shared her views on how is AMD making significant strides in developing these advanced processors for the global market, how AMD addresses these complexities by leveraging AI not only as a feature of their products but also as a powerful tool within their own design processes, despite the challenges inherent in designing and manufacturing AI chips.
As the global semiconductor industry heads towards the trillion-dollar mark by 2030, artificial intelligence is positioned to revolutionize chip design, helping to meet the rapidly evolving demands of this dynamic field. While AI might not solve every challenge, it holds the potential to democratize access to chip design and EDA tools, making them more accessible and affordable. As semiconductors continue to miniaturize, AI and GenAI are pushing the boundaries at the nanoscale - bringing precision, optimizing layouts, predicting performance issues, and unlocking new possibilities.
AI has a huge potential to reshape almost every industry and processes that has been carried out in these industries and before looking at AI and its impact on semiconductor design let us take a look at what is happening in the semiconductor industry today. The chips are becoming pervasive and almost every sector needs automation and chips. While the demand is huge and very aggressive there are challenges in chip design that have emerged. We no longer get the type of scaling that we used to get with every new node that we enter. Moore’s law which stated that every 2 years you can double the transistor count in your design with minimal cost is no longer holding good and the customers need chips with the smallest of areas, lowest of powers and highest of performance while keeping the cost at Bay.
Therefore, these are huge challenges that we are facing in the industry today and whenever we are challenged Innovation happens. So that has been the case with the semiconductor industry as well. We have had huge strides of innovation especially in the architectural side with Heterogenous Computing, Chiplet architectures, Advanced Packaging methodologies and also adaptive Computing and many more. Now, AI has come in as the new evangelist and it is touching almost every aspect of the design.
For those of you who are not familiar with the chip design process we have various stages. The first is the architecture and the definition in a natural language but we take that and do the design and description in Hardware description languages which is called the RTL. And, then we have the verification teams which verify the pre-silicon design and then it is implemented into Gates and that is what is sent to fabrication.
We also have DFT - designed for testing, where we introduce test circuits in the design pre-silicon, which aids in Silicon testing. While AI has had a huge impact on each of these sectors it is being leveraged by the design engineers and teams for productivity enhancements and optimizations and not just early warning. A lot of errors down the line can be avoided at an earlier stage with the use of AI. For example, in RTL there could be constructs that are not conducive for implementation and those can be flagged much ahead of time and not wait for the design to get to the implementation phase. Furthermore, AI has also helped in coming up with the most optimized RTL code, which is sometimes used in assisting RTL Engineers to come up with several optimizations.
The next phase which is the verification phase is an extremely labour-intense discipline where you need hundreds of Engineers to verify your chip and designs are extremely complex and that is where we have seen a lot of benefits with Engineers learning to debug with AI-assisted tools and also effective triaging of the failures and then anything unique that you do not want to ignore can be flagged and also when it comes to the coverage, we write thousands of tests to get the coverage, - line coverage for a pre-silicon design and that has become a lot easier and we have cut out a lot of redundancy in the testing.
Next comes the implementation with every process node that we move to and shrinking technology it is difficult to close timing, the tail becomes much longer with every Node change and this is where AI has come in extremely handy in predicting, placing and routing the design in a very optimal manner and also in getting the timing closure much ahead of time and there are some macro placements that can play a very critical role in your design. Therefore, leveraging some of these with AI-assisted methods has helped us in obtaining great results.
Last but not the least comes the test circuits with DFT that we introduced. AI has been able to help us obtain the most optimized test circuits which do not take up too much of the area also we package the scan patterns by using AI methods which, in turn, has saved a lot of tester time and this is a very expensive resource in chip design. Therefore, this is only the start of using AI in chip design but I believe we are already seeing a big impact and the journey should continue and we should witness a lot more optimizations and use case scenarios coming out.
From the design point of view, the design Engineers are the ones who know the pain points. They have an idea about what is taking longer than they would want to or like to. Therefore, they can figure out a lot of optimization points and provide that as feedback to the EDA. And, unless the two are married you will not see that kind of adaption into the industry. Hence, it is a very important relationship or a partnership and that is going well for us. Also, we have partnered with our EDA tool vendors and they are making a lot of strides using AI we are benefiting from it and we have also created a feedback loop with Innovation.