SK hynix has outlined a technology strategy targeting the next-generation memory market, moving beyond high bandwidth memory (HBM). With the rapid growth of the artificial intelligence (AI) inference market, data movement between memory types has emerged as a new bottleneck. The company aims to maintain its leadership in the AI semiconductor market through customized HBM, high bandwidth flash (HBF), and 3D stacked DRAM.
Ahn Hyun, Chief Development Officer of SK hynix, emphasized during his keynote speech at the 28th Semiconductor Exhibition (SEDEX 2026) in Seoul that "to ensure the continuous growth of AI infrastructure, it is necessary to optimize the entire memory hierarchy beyond improving the performance of individual memory types."
In his talk on the evolution of memory technology in the AI era, Ahn noted that the demand for AI inference is rapidly increasing, revealing the limitations of existing memory structures. He projected that by 2030, the scale of AI infrastructure will expand 19 times compared to this year, driven by the growing complexity of AI models, the volume of information to be processed, and the number of inference requests.
Ahn predicted that while the speed gap between processors and memory was previously a concern, the movement of data between different memory layers, such as HBM, DRAM, and solid-state drives (SSD), will become the new bottleneck. He explained that it is essential to enhance the performance of existing memory while introducing new memory technologies between layers.
To address this, SK hynix plans to evolve HBM into customized products designed to meet the specific needs of AI systems. The company is developing next-generation products like HBM4E and HBM5, while also preparing customized HBM that incorporates requested features on the base die.
Ahn stated, "While standard HBM and customized HBM will coexist, the trend toward customization will expand in the long term. We are equipped with various core technologies to sufficiently satisfy our customers."
Beyond HBM, the company introduced HBF and 3D stacked DRAM as future technologies. HBF is a next-generation memory that applies HBM's stacking technology to NAND flash, with development goals of achieving half the data transfer bandwidth of contemporary HBM while providing over ten times the storage capacity.
Standardization of HBF is also accelerating. SK hynix has collaborated with global partners like SanDisk and Google to unveil initial specifications at the Open Compute Project (OCP) in July, with plans to finalize the specifications by January next year. Ahn emphasized, "It is essential in the AI era to define standards and create an ecosystem together with various customers and partners."
3D stacked DRAM involves directly stacking DRAM on top of semiconductors used for processing, such as graphics processing units (GPUs). The goal is to significantly increase input/output connections compared to existing HBM, enhancing data transfer performance while reducing power consumption.
The next-generation enterprise SSD technology, dubbed 'Storage-Next,' is also under development. This technology aims to redesign the NAND and controller architecture to enhance input/output processing performance by four to eight times compared to existing solutions, with joint development efforts underway with customer companies.
Ahn concluded, "We do not view memory merely as a component. Since the technology requirements vary by customer, we will lead the continuous growth of the AI ecosystem through new memory solutions."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.