Will AI Chip Design Tools Take Your Job?

The concept of technology advances taking away jobs isn’t new. In 1908, the Model T was developed. In its first year of production the Ford Motor Company sold over 10,000 of them. As a result of this trend, the horse and buggy industry slowly became a specialty item, as it is today. Many jobs were lost in the process. You’ve probably heard a joke or two about comparing something out-of-date to a buggy whip. As is typically the case with a story like this, there is more. First, a huge number of jobs building, selling, and maintaining automobiles was created. The supply chain for gasoline is a whole story by itself. Another aspect of any story like this is adaptation. Some parts of the horse and buggy industry adapted. Some didn’t. You can read more about that here. Next, I’ll explore this phenomenon as it applies to chip design – will autonomous design take your job away? But first, the following quote from the link provided is worth repeating with respect to adaptation:

Timken Company’s signature product, roller bearings, were first used in wagon wheels in the 1890s. They easily adapted to the automobile because they could be applied “to nearly anything that moved.” Timken remains a market leader to the present day.

This is where autonomous driving and autonomous chip design begin to diverge. In this context, a “chip” is actually a collection of chips and complex software all interacting in a highly integrated and dense form factor. Potentially billions of transistors all working together, in some way, to perform incredibly sophisticated tasks that include many algorithms to implement AI and machine learning. The possible configurations and outcomes are vast. A closed-form, fully automated solution to this design problem is very remote. What is real, however, is cutting the problem down to size.

I discussed one example of AI assistance in my prior blog post. In that post I mentioned a solution based on reinforcement learning called DSO.ai. This technology autonomously searches design spaces for optimal solutions. By applying AI to chip design workflows, DSO.ai facilitates a massive scaling in the exploration of choices while also automating a large volume of less consequential decisions. In other words, it helps the designer find the best solution for implementation.  Nobody loses their job in this scenario, they just get to do a better job, which is good for everyone.

In that prior post I also talked about intelligent test selection and machine learning-assisted library characterization. AI also finds application in many other areas, from process optimization to physical design and security analysis and optimization. All of these technologies aim to find problems faster, identify opportunities for optimization, and generally perform highly detailed analysis, allowing the designer to focus on what they do best.