How AI will change chip design

Moore’s law is ending. Engineers and designers can only do a lot to shorten the transistor and pack many of them as possible in chips. Therefore they are turning to other approaches for chip design, including technologies such as AI in the process.

For example, Samsung is adding AI to its memory chips to enable processing in memory, which saves energy and fasting machine learning. Talking about speed, Google’s TPU V4 AI chip has doubled its processing power compared to its previous version.

But AI still has more promise and capacity for the semiconductor industry. To understand better how to revolutionize AI chip design, we talked with Heather goreSenior Excise Manager for Mathworks‘Matlab platform.

AI is currently being used to design the next generation of Chips?

Heath Gore: AI is an important technique as it is included in most parts of the cycle, including the design and manufacturing process. There are many important applications here, even in general process engineering where we want to customize things. I think the detection of the defect is a large in all stages of the process, especially in manufacturing. But even further thinking in the design process, (AI now plays an important role) when you are designing light and sensors and all different components. A lot of discrepancy detection and mistake is mitigation that you really want to consider.

Heather goreMathworks

Then, thinking about the logistic modeling you see in any industry, downtime is always planned that you want to reduce; But you are also doing unplanned downtime. Therefore, given the historical data when you had the moments where it probably took a little longer than expected to build something, you can take a look at all of that data and use AI to try to identify the proximal cause or be something that can jump out even in processing and design stages. We are once -as a future stating tool, or something as a robot, but many times you get a lot of information from data through AI.

What are the benefits of using AI for chip design?

Gorr: Historically, we have seen a lot of physics-based modeling, which is a very intensive process. We want to do Low order modelWhere instead of resolving such a computationally expensive and comprehensive model, we can do something cheap. You can make a surrogate model, so to speak, use that physics-based model, use data, and then do your own Parameter sweepYour adaptation, your Monte Carlo simulation Using surrogate models. It takes much less time computable than to solve physics-based equations directly. Therefore, we are seeing the benefit in many ways, including efficiency and economy that are the results of quickly recurring on experiments and simulation that will actually help design.

So it is like being a digital twin in a sense?

Gorr: Absolutely. It is too much what people are doing, where you have a physical system model and experimental data. Then, in combination, you have this other model that you can tweek and tune and try different parameters and experiments that allow all those to sweep through different situations and finally come with a better design.

So, it is going to be more efficient and, as you said, cheap?

Gorr: Yes, of course. Especially in use and design stages, where you are trying different things. This is clearly going to achieve dramatic cost savings if you are actually construction and production (chips). You want to use the real process engineering as much as possible, simulation, testing, using.

We have talked about the benefits. How about shortcomings?

Gorr: (AI-based experimental models) are not as accurate as a physics-based model. Of course, that’s why you sweep many simulation and parameters. But it is also the advantage of being a digital twin, where you can keep it in mind – it is not as accurate as the exact model we have developed over the years.

Both chip design and manufacturing systems are intensive; You have to consider every small portion. And it can be really challenging. This is a case where you may have models to predict something and its different parts, but you still need to bring it together.

One of other things also requires data to think that you need data to make models. You have to include data from all types of different sensors and different types of teams, and therefore it increases the challenge.

Engineers how to use AI to preparation and extract insight from hardware or sensor data?

Gorr: We always think of making some prediction or doing some robot tasks to use AI, but you can use AI to come up with patterns and choose things that you may not have seen on your own. People will use AI when they have high frequency data coming from many different sensors, and sometimes this frequency is useful to detect things such as domains and data synchronization or re-start. They can be really challenging if you are not sure to start from where to start.

One of the things I would say is, use available equipment. There is a huge community of people working on these things, and you can find a lot of examples (applications and techniques) Github Or Nausea centralWhere people have shared good examples, even small apps they have created. I think many of us are buried in data and are not sure what to do with it, so definitely take advantage of what is already in the community. You can find out and see what makes you understand, and the domain brings that balance of knowledge and imprisonment obtained from equipment and AI.

What engineers and designers should considerAre you using AI for chip design?

Gorr: Think what problems you are trying to solve or what insight you can expect to find, and try to be clear about it. Consider all different components, and test the documents and each of those different parts. Consider all those involved in this, and explain and hand over in a way that is sensible for the entire team.

Do you think AI chip will affect designers’ jobs?

Gorr: It is going to free a lot of human capital for more advanced works. We can use AI to reduce waste, customize the material, to adapt the design, but still you have this human whenever it comes to making decisions. I think it is a great example of working in the hands of people and technology. It is also an industry where everyone is involved – even on the manufacturing floor – what is happening, some level of understanding is required, so it is a great industry to pursue AI because we tested things and how we think about them before we put them on the chip.

How do you imagine the future of AI and chip design?

Gore, It is very dependent on that human element – involving people in this process and involving that explanatory model. We can do many things with the mathematical minutiae of modeling, but it comes down how people are using it, how in the process everyone is understanding it and implementing it. Communication and participation of people of all skill levels in the process is going to be really important. We are going to see less transparency, more transparency of those superprases predictions and information, and that digital twins – not only using AI, but using our human knowledge and all the tasks that many people have done over the years.