SkyeChip Joins SoftBank Unit SAIMEMORY’s Z-Angle Memory Programme for AI Data Centres
Malaysian chip designer SkyeChip Bhd has signed a design and engineering collaboration with SAIMEMORY Corp, a Tokyo-based subsidiary of SoftBank Corp, to help develop next-generation memory for artificial intelligence, The Star reported on October 21. Under the deal, SkyeChip will provide design and engineering support and custom intellectual property for SAIMEMORY’s Z-Angle Memory (ZAM) programme.
ZAM is being developed for data centres and other settings where large AI models are trained and run. According to the companies’ joint statement, the technology is designed for high-capacity, high-bandwidth data processing, better performance and lower power consumption. The programme sits within a research and development project subsidised by Japan’s New Energy and Industrial Technology Development Organisation (NEDO). SkyeChip will draw on its silicon design and integration experience, and both sides called the engagement a step toward a long-term partnership aimed at commercialising ZAM.
SkyeChip chief executive Datuk Fong Swee Kiang said AI computing is limited by the memory wall and called the company’s memory interface technology the door through that wall. SAIMEMORY president and chief executive Hideya Yamaguchi said SkyeChip’s technical expertise will play an important role in developing ZAM, and that he looks forward to working together on commercialisation and on creating a new market for next-generation memory.
What it means in practice
The deal follows the unveiling of SkyeChip’s MARS1000 edge-AI processor among six homegrown chips and adds another Japanese tie for Malaysia’s AI sector after the SoftBank, Grab and PETROS framework for Sarawak AI infrastructure. It also bears on sovereignty debates such as the choice of chips for a sovereign AI cloud.
For Malaysia’s semiconductor ambitions, design work on memory interfaces is a step up the value chain from assembly and testing, the country’s traditional strength. Memory bandwidth has become a bottleneck for training and running large models, so even a supporting role in a programme like ZAM gives a local design house exposure to one of the most contested parts of AI hardware.
What to watch: whether the collaboration turns into a formal commercial agreement, when ZAM prototypes or performance figures appear, and whether SkyeChip wins similar memory or interface work with other global AI hardware companies.