Apple's 20th-anniversary iPhone RAM upgrade could make it an AI monster
The 20th-anniversary iPhone stands to offer a major jump forward in on-device processing, with memory changes potentially increasing AI response speeds in leaps and bounds.
The 20th-anniversary iPhone's memory could make Siri and Apple Intelligence lightning fast.
Apple is fighting to keep its iPhone lineup relevant in a marketplace infatuated with artificial intelligence. While it has a considerable advantage in working in AI for years already, including the use of its self-developed AI-focused chips, it could soon make other major changes to improve performance.
According to sources of ETNews, Apple is seriously considering the possibility of adding mobile high-bandwidth memory (HBM) to its 2027 iPhone range. This would include the 20th-anniversary iPhone, which is expected to include other display and battery changes.
The use of HBM would be a big change for Apple, as it is a technique that requires the stacking of DRAM layers. The stacking increases the signal transmission speed, and therefore bandwidth, making it a lot quicker and easier for processors to access and secure data from the store.
The HBM is usually connected to the application processor, but there is the possibility of it being connected to the GPU, which can be used for AI processing. This is effectively what Unified Memory does in Apple Silicon Mac models.
A report source insists that Apple will use HBM to the iPhone in 2027, as it is "reviewing a change in the design" of its application processor to make it more useful for AI processing. The source adds that connecting the memory to the GPU is also "likely."
A strong possibility
The report believes that Apple may have already discussed the plan with memory suppliers, such as Samsung and SK Hynix. Both are creating their own mobile HBM modules using their own packaging technologies, with mass production expected after 2026.
SK Hynix is reportedly doing so under the name Vertical wire Fan Out (VFO), while Samsung Electronics is doing so under the title Vertical Cu-post Stack (VCS).
In the case of Apple, reports surfaced in December about Apple working with Samsung on faster memory for the iPhone. The intention was to create a larger DRAM package to increase the number of connectors, increasing the memory bandwidth.
Samsung and SK Hynix will probably fight tooth and nail to secure Apple's custom for their mobile HBM modules. It is anticipated that there will be fierce competition, as each company would be able to expand their existing HBM work in the server market to smartphones.
More anniversary device tech
Memory and improved Apple Intelligence processing won't be the only things expected in the 20th anniversary iPhone. Other more visible changes are also expected for the milestone device.
This includes changes to the display, such as a switch to a 16-nanometer FinFET process for the OLED display driver chip (DDI) to make it considerably more power efficient. A change in tech could also eliminate the bezel, by bending all four sides of the display.
The implementation of an under-display camera will also be tricky, since the screen needs to transmit light without deteriorating the camera image quality.
Current theories to implement the tech includes the use of a transparent polyimide as an OLED substrate material, as well as the use of special lenses to reduce optical loss from the OLED pixels.
Samsung is also expected to be using a new M16 material set for the iPhone 18 displays, and could create a custom material for the 20th-anniversary device.
Other rumors also include the use of 100% silicon material without graphite for the cathode of batteries, which can increase performance and battery life.
With AI having high processing requirements, and therefore power, any savings or other battery improvements will inevitably help make the model's AI more impressive.
Rumor Score: Possible
Read on AppleInsider
Comments
Apple uses mass manufactured components in novel ways to get the memory performance they want while minimizing cost and power consumption. I have no idea how much lower in bandwidth "mobile HBM" will be, but there isn't a free lunch. More bandwidth means more power consumption, and more cost.
Really don't see any reason at all for mobile HBM use in a smartphone, if it is designed like regular HBM. Apple would need to use a silicon interposer, a 4 hi stack (?), and a lot more bumps/pin, for something like 320 GByte/s at 4x the memory cost and power consumption? Makes no sense.
Are there any real advantages over LPDDR5 or LPDDR6 in a mobile device? I don't think there is. Even in the M Max SoCs, where HBM could provide a benefit, Apple choose a novel way of using a gazillion channels of mass market LPDDR to get about 95% of user facing performance HBM would provide for probably half the cost and power consumption, with a much cheaper way to get to 128, 256, 512 GB of memory.
If they are doing anything novel above and beyond LPDDR6 for a 2027 of 2028 iPhone, it's going to be something based on bog standard LPDDR6 with a novel memory interface. Like, the R1 in the AVP is supposedly some kind of wide-IO memory device. Samsung is creating something like that, a wide-IO DRAM device. It likely has 2x, 4x the pin outs as regular LPDDR and comes at a cost, but doesn't need a silicon interposer and a gazillion wire traces. Cheaper.
Samsung's LLW-DRAM might be the 95% solution Apple loves, but who knows if that even provides enough benefit.
Now the iPhone is really just a super mini portable computer (laptop) that takes photos. Oh and it makes phone calls just for fun.
I have the last iPhone Steve would probably approve of fully …iPhone SE 3rd generation. I wouldn’t trade it for anything. Nope. Not even the iPhone 16e. He’ll no.
When I think of “AI monster”, I don’t think of anything positive, haha. Skynet, Cylons, HAL 9000, Ultron, Agent Smith, Omni Consumer Products (OCP) from RoboCop.
OCP especially. The scariest monster out of all of those is the corrupt politically-connected American mega corporation headed-up by a bunch of greedy sociopaths. Quite the monster indeed, lol.