Apple’s Johny Srouji during a hardware presentation – Image Credit: Apple
While getting an industry award, Apple Silicon chief Johny Srouji said that he wants generative AI to design chips, which has been blown out of proportion to mean that Apple hasn’t ever used AI to do so. This is, of course, nonsense.
On Thursday, reports started to circulate about Apple and an intention to use generative AI to design the chips used in its products. Stemming from a speech at Imec in May, Apple Senior Vice President of Hardware Technologies Johny Srouji made comments about the use of machine learning to design chips.
A video of the speech, seen by Reuters has Srouji explaining Apple’s design of custom chips, going back to the A4 and up to modern-day chips like the M4 processor line.
Srouji’s comments have been taken to mean that Apple wants to start using artificial intelligence for the first time to make it quicker for its engineering teams to design the chips. In fact, it’s something Apple has already been doing in some capacity for many years.
The Apple executive was at the ITF World conference in Antwerp, Belgium in late May. He was being given an innovation award from Imec, an electronics and technology research group, for his role in developing Apple’s chip technology roadmap.
The video was of a speech at the conference by Srouji, and are characterized by Reuters as private remarks.
Srouji is said to have covered the main beats of Apple’s custom chip development. A key lesson Apple learned along the way, likely decades ago, was that it needed to use the most cutting-edge tools available to design its chips.
For this, Srouji said this included using the most up-to-date chip design software, provided by electronic design automation (EDA) firms.
“EDA companies are super critical in supporting our chip design complexities,” remarked Srouji. “Generative AI techniques have a high potential in getting more design work in less time, and it can be a huge productivity boost.”
Those two comments together could be conflated to mean that Apple wants to use generative AI to design its chips. The reality is somewhat different.
The modern reality
Part of the problem with the reporting of Srouji’s comments is the awareness of how chips are designed and made. While it is generally understood that Apple designs the chips which are then made by foundries owned by supply chain partner TSMC, the actual design element is still wrapped up in mystery.
Outside of the benefits of new chip designs being explained each fall during an iPhone launch, a map of a chip’s sections, and some vague hardware lab scenery, Apple doesn’t go into the detail of how a chip is produced internally.
As Srouji’s comments demonstrate, part of this process involves software from EDA companies. Humans are involved, but EDA software is needed for a lot of the process.
Apple’s chip designs use humans for the major notes, EDA for details – Image Credit: Apple
Engineering teams start by determining the specifications of the chip, followed by the architecture design that maps out where elements like processor cores and memory are placed and how data should flow through the chip.
Where the EDA software comes in is to create the actual circuit design for the chip. This can include the layout of connections, as well as simulating testing before committing to a physical version.
Given the M4 chip has 28 billion transistors made using a 3-nanometer process, mapping this out by hand would require an army of engineers working together for years — the proverbial infinite monkeys on infinite keyboards. Using automation tools means that software works on the design, and repeatedly do so if there are any issues or changes requested by engineers.
This on its own is using a machine learning system to design chips. Apple has already been using machine learning — literally AI — to design its chips. And, it has done so for years.
However, companies like EDA software supplier Synopsys are working to make AI more of a driver of chip design. This involves an expansion of what is already being performed, such as by allowing EDA software to come up with new solutions to reduce power consumption.
Synopsys thinks that using generative AI could help create brand new ways to design chips that conventional thinking wouldn’t come up with.
To a company like Apple, which relies on such innovations, this would be an improvement it would be willing to try.
Less fear, more speed
While the reporting may tap into the continual fear of AI taking creative agency away from people and replacing jobs, it’s not the case here. By using EDA software to design chips, Apple is already using AI as part of its production process.
There aren’t going to be significant job losses or restructuring of hardware lines to introduce generative AI. Apple is already in a position that enables the increased use of AI in chip designs.
Not just from it being better than humans, but from necessity prompted by the sheer scale of the task.
Full AI design certainly is a thing that Apple has already started to explore, and has done so for years. The generative AI talk is merely being open to expanding what it already has at its disposal.