certain M5 chips will adopt a completely new process to boost their performance

Apple has been designing SoCs (System-on-a-Chip) for almost 15 years, and has always stuck to the same main design principle: closely grouping the CPU and GPU parts of its chips on a single circuit. It is this approach which has notably enabled the firm to deliver high-performance processors for its various iPhones, and it is this same strategy that Apple has been implementing since 2020 on the M chips of its various Macs and iPads.

If we're talking to you about it today, it's because the Californian giant could revise its copy in this area. According to Ming-Chi Kuo, Apple is indeed working on an evolution of its architecture, to achieve a clear separation of the CPU and GPU parts on certain M5 chips.

An evolution permitted by TSMC?

This development would obviously only concern the most high-end versions of the next generation of Apple processors. Its main objective would be to improve their performance, as well as the yields during their production.

This approach would also allow future M5 Pro, M5 Max and M5 Ultra chips to better serve Apple in terms of Artificial Intelligence… by integrating more easily with the PCC (Private Cloud Compute) servers used by the firm for Apple's calculations. Intelligence.

In detail, this novelty would be enabled by the use of the new 2.5D packaging technology “SoIC-mH” (for molding horizontal) signed TSMC.

« The M5 series chips will adopt TSMC's advanced N3P node, which entered the prototyping stage a few months ago. Mass production of M5, M5 Pro/Max, and M5 Ultra chips is planned for H1 25, H2 25, and H1 2026, respectively. », explique Ming-Chi Kuo.

« The M5 Pro, Max, and Ultra processors will use a server-grade SoIC package. Apple will use a 2.5D package called SoIC-mH (horizontal molding) to improve production yields and thermal performance, with separate designs for the CPU and GPU “, he continues.

Offered in addition to the group's third-generation 3nm “N3P” engraving, this manufacturing method makes it possible to encapsulate several chips in a single processor, while improving the thermal efficiency of the whole. This device therefore makes it possible to push the integrated components at full speed for longer before they are confronted with excessive heating, and therefore with thermal-throttling. Performance is therefore better.

As mentioned above, this process also has the advantage of improving yields during production. Reliable, it limits failures and losses during manufacturing, and therefore allows more chips to pass quality controls.

Note that this prediction from Ming-Chi Kuo, relating to a potential separation of the CPU / GPU parts on the M5 Pro, M5 Max and M5 Ultra chips, echoes another rumor that appeared online recently.

We recently learned that Apple would also be working on a separation of two essential elements of its “A” SoCs, this time dedicated to the iPhone. The iPhone 18s would in fact rely on chips with separate RAM… but until now, the RAM was actually integrated into the SoC.