At the silicon front, companies like Apple, Qualcomm, MediaTek, and Samsung have progressed tremendously in making their SoCs incredibly powerful for on-device AI, but their advancements will continue to remain stunted due to packaging and DRAM limitations. While one of these aspects has been addressed, a new report states that SK hynix is working on the second obstacle through the development of 3D-stacked DRAM-on-logic architecture on mobile phones.
SK hynix is recruiting engineers to accelerate the development of 3D-stacked DRAM-on-logic architecture, removing on-device AI limitations on smartphones
This next-generation memory stacks directly on top of a semiconductor chip and will serve as a key component in edge AI computing. According to ZDNet, SK hynix began hiring design engineers through its U.S. subsidiary in San Jose, California, and has formed a partnership with an unnamed customer to apply this technology. It is likely that this company is Apple, since the latter is expected to be the first to ditch the older InFO-PoP packaging in favor of the newer WMCM (Wafer-Level Multi-Chip Module) packaging for the A20 Pro.
3D-stacked DRAM-on-logic architecture is entirely different from PoP, which only places two chips on top of one another. The next-generation packaging can shorten the distance between chips and provide a higher number of data-transfer channels within the same footprint. In addition to reducing data-transfer latency, smartphones with this technology will also experience greater power efficiency and space utilization.
While SK hynix’s approach is certainly a breakthrough to remove the limitations on smartphones and allow them to truly become on-device AI devices, there are other implementations being worked on as well. Qualcomm, for instance, is developing 3D DRAM for Neural Processing Units (NPUs) and has partnered with CXMT to bring on-device AI capabilities to smartphones in China.
Similarly, Samsung is working on Low Latency Wide DRAM (LLW) that mimics HBM’s integration without the trade-offs of higher temperatures while aiming to deliver up to 150 percent higher bandwidth than traditional LPDDR5X DRAM. Last but certainly not least, Huawei is also working to bring HBM memory to smartphones, but given that this memory is only suited for data centers due to its stacked design, high price, and low yields, it will likely be a customized solution like Samsung did with its Exynos 2600’s memory.
To properly experience on-device AI on smartphones, there needs to be adequate and speedy memory available. Unfortunately, with Apple’s A19 Pro topping out at a measly memory bandwidth of 75.8GB/s, it is clear that new packaging is now mandatory to carve out a new path in the industry. Assuming Apple is one of the first adopters, we doubt we will witness this implementation in the A20 Pro since the iPhone 18 Pro has reportedly entered mass production, so perhaps next year, we will witness a grand surprise.
News Source: ZDNet
About the author: Omar Sohail is a reporter and analyst for Wccftech's mobile section, specializing in the technology and business of the mobile industry. His expertise lies in the intricate hardware supply chain, covering developments in semiconductor manufacturing, chip lithography, and camera sensor technology.
Follow Wccftech on Google to get more of our news coverage in your feeds.
