Samsung Working on Ultra-Fast RAM for AI Smartphones and Tablets

Samsung is reportedly developing high-bandwidth memory (HBM) RAM for smartphones and tablets, aimed at supporting on-device artificial intelligence workloads. HBM offers faster speeds and lower power consumption than conventional RAM, and is typically used in servers or high-end GPUs to handle heavy computational tasks. Samsung is said to be embedding this technology into smartphones and tablets, which are increasingly being tasked with AI-driven duties, in order to boost the performance of mobile devices into AI-enabled devices.

Traditionally, DRAM used in smartphones or tablets employs copper wire bonding. However, this approach has limitations in the number of input-output (I/O) channels, limiting data transfer efficiency and heat management. HBM is seen as capable of improving thermal resilience while maintaining stable performance under heavy workloads, including AI.

Samsung is also developing a technology called Vertical Copper Post Stack (VCS). With this method, DRAM chips are stacked like steps and connected via very small copper pillars. This approach could significantly increase memory bandwidth even in the constrained space of a smartphone.

ETNews reports that the copper pillar ratio in VCS has risen dramatically from around 3-5:1 to 15-20:1. Additionally, the use of Fan-Out Wafer-Level Packaging (FOWLP) is claimed to further increase the number of I/O terminals and boost bandwidth by up to around 30%.

Nevertheless, this technology also presents new challenges. Copper pillars that are too small risk bending or breaking if their diameter falls below about 10 micrometres. To address this, Samsung is using FOWLP to reinforce the chip structure by extending copper wiring to outside areas.

It is not yet known when mobile HBM might make its official debut. However, the technology is expected to appear in future Samsung chipsets, such as the Exynos 2800 or Exynos 2900.

Although promising, mobile HBM is not expected to arrive soon. One reason is that mobile RAM remains expensive. As a result, improvements in AI capabilities on smartphones over the coming years are likely to depend more on chipsets and storage than on HBM, according to KompasTekno’s summary of Wccftech.