The preliminary report was published on September 22, 2026and it is the first time that someone outside of TSMC has removed the chip from a real phone to check it. Until now, these speed, consumption and density figures had only been given by the foundry itself: a 15% more speed at the same consumption compared to the N3E, a 30% less consumption at the same speed, and a logical density more than 1.15 times higher. “2 nm” does not measure any physical dimension of the chip, it is a generation label, which is why this external verification is worth more than any commercial figure. The full report is for subscribers, although TechInsights will host a public webinar on October 14.
Nanosheet transistors: this is how Apple leaves FinFETs behind
The most important change is in the architecture. He A20 Pro assembles gate-wrapped nanosheet transistors, known as GAAs, which have replaced the dominant FinFETs since the early 2010s. In a FinFET, the gate covers the silicon fin on only three sides, and that free face loses current the smaller the transistor. With nanosheets, the door wraps around the channel on all four sides and improves electrical control.
TSMC adds another advantage with its NanoFlex technology: if in a FinFET the output current was only adjusted by adding fins, with nanosheets you can vary its width between cells of the same wafer. Performance cores use wide blades and efficiency cores use narrow blades. TechInsights points out internal spacers that reduce leakage and parasitic capacitance, a detail that distinguishes TSMC's implementation from its competitors.
Apple has not used this greater density to shrink the chip, but to put more power in the same space: the silicon matrix of the A20 Pro measures about 98.2 mm²almost the same as that of the A19 Pro. The CPU maintains six cores (two high-performance and four efficiency) and the GPU goes up to seven, one more than before. The chip adds two neural motors 16 cores and additional CPU accelerators, with which Apple ensures that AI performance is doubled compared to the A19 Pro.
Memory moves next to processor to improve cooling
For years, Apple placed LPDDR memory on top of the processor, package-on-package. The A20 Pro breaks with that design: TechInsights detects a WMCM packaging, with the processor and DRAM side by side in the same wafer-level package.
Preliminary analysis points to an interface 96-bit LPDDR5Xa fact that must be confirmed in the final report. The change makes sense: the memory is no longer on the hottest part of the chip, the bus can be widened without stacking penalties, and each component can be tested separately before assembly, improving manufacturing performance.
The new design also helps dissipate heat. Apple declares a vapor chamber with triple the surface area of the iPhone 17 Pro and promises up to 40% more sustained performance: without the memory in the way, the heat reaches the dissipation surface sooner. The disassembly of iPhone 18 Pro Maxpublished on September 18identified the Qualcomm X80 modem and Apple's N1 Wi-Fi chip.
How TSMC competes against Samsung and Intel processes
TSMC is not the only foundry with a 2nm class GAA process. Samsung was the first to bring the wraparound door to production, with its 3nm in 2022and its SF2 entered series production at the end of 2025. Independent reports placed its performance above the 50% in January 2026.
A20 Pro 2nm FAQ The iPhone 18 Pro features the A20 Pro, manufactured using TSMC's 2nm N2 process, TechInsights' independent teardown has confirmed.
They are wrap-around gate nanosheet transistors that surround the silicon channel on all four sides, unlike previous FinFETs, and reduce leakage current.
Yes, the chip uses a WMCM packaging in which the DRAM memory sits next to the processor instead of on top, which improves cooling.
The full report is for subscribers only, but the firm will host a public webinar on October 14, 2026.
What chip does the iPhone 18 Pro have?
What are the GAA transistors used by the A20 Pro?
Has the memory position changed on the A20 Pro?
When will all the details of the TechInsights analysis be known?
