[Edaily Reporter SOYEON KIM ] As “next-generation packaging” and “customized memory” emerge as the keys to overcoming the performance limitations of artificial intelligence (AI) semiconductors, the global semiconductor ecosystem is turning its attention to Taiwan. Key executives from SamsungElectronics and SK hynix will attend “Semicon Taiwan,” which opens on the 2nd of next month, to announce their memory innovation strategies. Global semiconductor companies such as NVIDIA, TSMC, and Meta will also be in full force to unveil roadmaps for heterogeneous integration technologies, including advanced packaging, aimed at the AI era.

According to industry sources on the 31st, Semicon Taiwan will be held from the 2nd (local time) through the 4th at the Nangang Exhibition Center in Taipei, Taiwan. Major forums and keynote speeches will begin on the 2nd.

SEMICON is a global trade show for the semiconductor industry organized by the Semiconductor Equipment and Materials International (SEMI) and held in major semiconductor-producing countries such as the U.S., Taiwan, South Korea, Japan, and China. This year’s SEMICON Taiwan is expected to attract more than 100,000 industry professionals from 65 countries. Some 1,300 companies will set up more than 4,300 booths to showcase their latest technologies. Global semiconductor companies such as NVIDIA, Meta, TSMC, Micron, Siemens, Broadcom, MediaTek, and Lam Research will also participate in large numbers.

Held under the theme “Shaping Tomorrow,” SEMICON Taiwan 2026 will cover the entire semiconductor industry—from wafer manufacturing to equipment and materials, IC design, system applications, and the startup ecosystem—while discussing the future of AI semiconductors. In particular, interest from global companies is growing, driven by Taiwan’s industrial strengths, which include a concentrated ecosystem of foundries, back-end (OSAT), and advanced packaging companies centered around TSMC.

In the memory-related sessions, SamsungElectronics Executive Vice President Choi Jang-seok and SK hynix Executive Vice President Kim Ho-sik will each present memory innovation strategies for the AI era. As the importance of memory grows as a key factor determining AI performance, its status is rising beyond that of a simple component to become the core infrastructure of AI systems.

On the 1st, Executive Vice President Choi will present on the topic “Powering the AI Era Through 3D Integration,” introducing technologies such as zHBM, zNAND-O, and 3D packaging. zHBM is a technology that goes a step beyond the conventional method of placing high-bandwidth memory (HBM) next to an AI accelerator (xPU) by vertically stacking HBM on top of the AI accelerator. By stacking HBM directly on top of the GPU, the goal is to reduce the distance data must travel between the AI accelerator and the memory, thereby increasing bandwidth and power efficiency.

zNAND-O is a next-generation, high-performance NAND flash solution currently under development by SamsungElectronics. It utilizes the vertical stacking technology of existing V-NAND and combines it with through-silicon via (TSV) technology to create 3D-packaged semiconductor chips with 4 or 8 layers. The “-O” at the end of the product name indicates that it is optimized for on-device AI environments.

As the focus of the AI industry shifts from training to inference, the importance of memory capacity, bandwidth, and power efficiency is growing even more. Vice President Kim points out that, amid surging memory demand driven by the proliferation of AI, KV caches and agent workflows are emerging as new bottlenecks. The company plans to introduce next-generation memory technologies, including in-memory computing, virtual HBM pooling, Compute Express Link (CXL)-based connected memory, and long-term AI memory orchestration.

The role of memory manufacturers is also expected to expand beyond simply supplying high-performance products to include designing AI architectures in collaboration with partners. This is because the importance of integrated technologies—which efficiently allocate and move the vast amounts of data and long-term memory that AI must process across various memory types, such as HBM and CXL—is growing, as they reduce power consumption and increase computational speed. In the advanced packaging session, Yang Seung-taek, Team Leader at SK hynix, will deliver a presentation on heterogeneous integration packaging technologies for the AI era, alongside global companies such as TSMC.

In addition to SamsungElectronics and SK hynix, other Korean companies are also participating. SamsungElectroMechanics is participating in Semicon Taiwan for the first time to introduce glass substrates, package substrates, and advanced packaging technologies—all considered next-generation alternatives for AI semiconductor packaging. SamsungElectroMechanics strengthened its capabilities in this sector last year by recruiting Vice President Kang Doo-an, who previously led Intel’s semiconductor packaging division.

HANMISemiconductorCO.,Ltd. will showcase its technological capabilities centered on the TC bonder, a key piece of equipment for HBM production for AI semiconductors. In addition, numerous domestic small and medium-sized equipment companies will jointly participate in the Korea Pavilion, setting up booths to promote their technological capabilities to global clients.