Nanya Technology discloses its AI memory progress for the first time, with some products already in trial production

Nanya Technology recently disclosed for the first time the progress of its customized AI memory development. The company stated that it is collaborating with multiple logic IC manufacturers and ecosystem partners, and some products have entered the trial production stage. More concrete results are expected to emerge in the second half of this year.

Nanya Technology's main technology is the UltraWIO (Ultra Wide I/O) architecture memory. This architecture is not a JEDEC standard DRAM product, but a customized solution tightly integrated with the customer's AI computing engine. Its concept is similar to the collaboration model between High Bandwidth Memory (HBM) and GPUs, solving the data access efficiency bottleneck in AI computing by significantly increasing the number of data channels and bandwidth.

In terms of physical integration, UltraWIO will adopt 3D stacking and advanced packaging technology, and may incorporate Wafer-on-Wafer (WoW) technology in the future. The first layer of DRAM stacking will be completed by Nanya Technology itself, while the logic chips will be designed by partners.

Regarding market strategy, Nanya Technology points out that currently, AI computing still primarily relies on GPUs paired with HBM. However, with the rapid growth in AI inference demands, not all applications will necessarily adopt HBM in the future; more memory architecture combinations may emerge, such as GDDR, Local Memory, and custom memory. As demand for AI PCs, AI smartphones, and various terminal AI devices increases, Nanya Technology hopes to establish a technological barrier in the emerging AI terminal computing market through the heterogeneous integration of UltraWIO custom memory and ASIC.