Apple rarely lets three executives sit down together and talk shop about silicon. This week it did exactly that, putting Tom Boger, Kaiann Drance, and Sri Santhanam in front of a microphone at Apple Park to explain why the Apple A20 Pro chip looks the way it does.
The conversation happened on The Circuit, hosted by Ben Bajarin, and it covers Apple’s move of its entire lineup to the 2 nanometer process, from the Apple Watch Series 12’s S11 chip up through the Apple M5 Ultra. The headline point from all three guests is that the manufacturing node is only one piece of the story.
Apple Says Its Doubled Neural Engine Runs as One Unit
The Apple A20 Pro chip carries two separate 16-core Neural Engines instead of one. Sri Santhanam, who leads Apple’s silicon engineering group, described the setup directly.
“These are two discrete 16 core neural engines, but they’re tied together at the hip. So you could actually run different models or workloads simultaneously in parallel, two in parallel across the two ANEs, or one workload split across both.”
The Core AI framework decides how a task gets split between the two engines, so developers don’t have to manually assign work to one Neural Engine or the other. Apple named three workloads that already lean on the dual setup: Siri AI, most of the camera pipeline, and Deep Fusion, which the company called one of the largest neural networks running on an iPhone.
There’s also an efficiency angle. Santhanam said a workload that doesn’t need the full output of both engines can split across them and run at a lower frequency and lower voltage instead, saving power rather than adding speed.
A New Chip Package Moves Memory to the Side for Better Cooling
Apple carried its M-series packaging approach down into a phone chip for the first time. The Apple A20 Pro uses a multi-chip module package that places DRAM to the side of the system on a chip instead of stacking it on top, a design Santhanam said was directly inspired by the Apple M1 chip generation.
The tradeoff is a larger chip footprint on the logic board, which Tom Boger said triggers changes across the rest of the phone’s internals.
Boger said moving the DRAM required Apple to fit a larger vapor chamber into the iPhone 18 Pro Max and iPhone Duo while still increasing battery size, calling it a chain reaction across the device’s internal layout.
Apple Calls Itself a Product Company Building Silicon Years Ahead
Boger drew a clear line between Apple and companies that sell chips to multiple customers. “We’re not a merchant silicon vendor,” he said, meaning Apple designs silicon years ahead against a specific product roadmap rather than choosing from a catalog of existing parts.
Kaiann Drance, who handles iPhone and Apple Watch product marketing, positioned the Neural Engine as one piece of a wider AI setup rather than the only engine doing the work. Her description put the CPU, GPU, and Neural Engine on the same team.
“The neural engine is the most power efficient way to run a workload, but now it’s complemented by having these powerful neural accelerators in the GPU for other tasks that you might not want to be as power efficient. You just want the most performance as possible.”
Drance also gave a concrete example of what the added performance unlocks on iPhone Duo. She pointed to a Netflix app demo that renders video across both displays with reflections shown on the bottom half, something she said the phone’s earlier chips could not handle.
Santhanam also addressed security separately, pointing to a memory tagging technique Apple added to the chip that took, in his words, a lot of effort and internal debate to implement without slowing the CPU down.
Our Take
I think the most telling line in this whole conversation is Boger’s “not a merchant silicon vendor” comment. It’s a polite way of saying Apple will keep building hardware nobody else can copy quickly, because nobody else controls the phone, the chip, and the software stack at the same time. That’s the real moat – not specifications or neural engine core count of its chips.
I’m skeptical of how much credit the dual Neural Engine deserves versus the packaging change. Apple didn’t give a standalone performance number for the second ANE, and the 40% sustained performance gain on iPhone 18 Pro Max sounds like it’s doing more of the actual work for AI and gaming workloads than the extra neural cores are.
