Apple’s silicon development leader on processor design

“Disruptive products cannot be built from legacy chips,” Apple Senior Vice President of Hardware Technologies Johny Srouji said when he spoke at ITF World in Belgium in May. He was there to receive the prestigious 2025 IMEC Innovation Award and shared some of the lessons Apple has acquired since it began building its own processors with the A4 chip inside the iPhone. 

During his speech, he explained how the scalable architecture Apple chose for its chip development is enabling the company to transform all its hardware. And he shared a few tips for intelligent hardware design, such as to always use the best available tools and technologies during development. 

“I always tell my team, we want to be gated by physics, nothing else,” he said.

iPhone and the future of computing

Apple’s processor journey really began years before Apple Silicon. But it was only when it began work on the iPhone that it committed to taking complete control of the processor development process. 

“The iPhone was seen as a full-blown computer,” he said. “To support its full feature set, it required an incredible amount of performance — but needed to deliver that within a constrained power.”

The company understood that while the easy path might be to use third-party processors, doing so made it hard to design products that stood out. Shifting to custom silicon was difficult, but it enabled the company to design in chip efficiency and gave it the chance to “control our destiny.”

Mobile tech in a Mac

What that means in terms of actual hardware is now reflected in the power and performance advantage characteristic of Apple silicon. And it’s a direct result of the original design decisions made about to the iPhone — decisions such as building chips optimized in such a way that the most “efficient part of the performance versus power curve coincides with the maximum sustained thermal envelope of the device its going into.”

In other words, Apple’s chips are built to deliver their most efficient performance when the devices are being used for quite demanding tasks. This can scale up for short-term higher-end chores, and tick down when all you want to do is check email or pick up some more cookies and adware while browsing. That efficiency by design, in conjunction with the hardware and software Apple also designs, makes for longer battery life and higher performance per watt.

It also makes for the sort of power efficiency you need to run multiple operations on the GPU, which makes Apple silicon highly suitable for on-device AI. It might seem ironic, but the design decisions made for the iPhone directly contributed to the processor choices Apple has made for its fleet of private cloud compute servers.

The fundamental building blocks

Transistor density is fundamental. “We rely on transistors as the chips get more complex,” Srouji said. “As our chips are getting more complex, the transistor as the fundamental building block is getting more important than ever.”

The 190 million transistors crammed inside the A4 and A4X chips that powered the iPhone and iPad 15 years ago would occupy just one tiny corner of the M3 Ultra chip’s 183 billion transistors today. Those numbers are why, for example, GPU performance on the M3 Ultra is 17,000 times that of the A4.

It is also why innovations such as support for unified memory and better performance per watt feeds into the Mac. It means the M1 MacBook Air delivered equivalent performance to that of an Intel MacBook Pro, with system performance scaling up with every model after that. Progress has been so fast that Apple is expected to introduce a Mac that uses an iPhone chip later this year.

The ability to design technologies in which Apple is only constrained by physics enables better design, thanks to the integration of software, hardware, and processors. “And thanks to our vertical integration any innovation here we can transfer to millions of products reaching millions of customers,” Srouji said.

The future remains unwritten

Apple hasn’t finished with chip design yet. It’s likely in a decade’s time we’ll see more progress, more performance, and more computational ability squeezed into Apple’s processors — particularly as we move to whatever the successor may be to the 2nm chips the company is now expected to use. It means you can expect more transistor density moving forward, which means more performance. 

Can we anticipate 17,000-fold increases? 

I don’t know. But from what Apple’s silicon leader is saying, the only barrier will be the laws of physics. Of course, as development gets close to the finite constraints of physics itself, we’ll see ever more focus on small wins that lead to better efficiency. Seeking such wins, don’t be too surprised to see Apple’s chip designers lean even more deeply into AI than they already do as they work to identify and test new solutions.

“Gen AI techniques have high potential to get more design work done in less time,” Srouji said. 

You can follow me on social media! Join me on BlueSky,  LinkedIn, and Mastodon.