Saimemory and SkyeChip Partner to Develop ZAM AI Memory
The firms aim to boost data center efficiency for AI workloads by designing custom Z-Angle Memory hardware.
Updated on Oct. 21, 2026 in Semiconductors
Live Poll
Do you believe partnerships between tech firms will successfully solve AI hardware performance limitations?
Saimemory, a Tokyo-based subsidiary of SoftBank, has announced a design and engineering collaboration with Malaysian firm SkyeChip to develop Z-Angle Memory (ZAM) technology. This research-stage effort aims to create specialized memory hardware to support the high-capacity, high-bandwidth data processing requirements of large-scale AI model training and inference.
Why it matters
As AI model parameters scale, memory bottlenecking has become a primary constraint on computational throughput and power efficiency in data centers. This partnership seeks to address these infrastructure limits by integrating custom silicon IP with new memory architectures.
The ZAM project aims to optimize for high-capacity and high-bandwidth processing compared to traditional data center memory. While the project is in the design phase, the companies have yet to disclose the physical parameters or power-efficiency targets for the silicon.
The players
Saimemory Corp.
A Tokyo-based subsidiary of SoftBank Corp. focused on developing specialized memory technologies for high-performance computing.
SkyeChip Bhd
A Bursa Malaysia-listed company that provides silicon IP and hardware products for semiconductor and high-performance computing applications.
The details
SkyeChip will contribute custom silicon IP (intellectual property) to the ZAM program, utilizing its expertise in chip integration to support Saimemory. The goal is to develop a specialized architecture that improves how data moves between processors and memory, which is critical for the training and inference cycles of large-scale AI models. ZAM (Z-Angle Memory) is intended to reduce overall system power consumption while maintaining the high data rates required by modern AI workloads.
Timeline
October 21, 2026: Saimemory and SkyeChip announce their ZAM memory collaboration.
The Tech Race
This effort places Saimemory and SkyeChip into a crowded field of memory innovation competing against established HBM standards. It reflects a broader industry trend toward custom silicon integration to solve AI-specific bandwidth and power constraints.
The ZAM technology remains in the design and research stage, meaning no commercial product is currently available for data center operators or hardware architects. Stakeholders should watch for future disclosures regarding technical benchmarks or commercialization timelines.
The takeaway
The partnership highlights the intensifying focus on memory-processor integration as a lever for AI efficiency. Industry observers should monitor for future disclosures on ZAM performance benchmarks to determine how this architecture compares against existing high-bandwidth memory alternatives.
Further reading
Explore the evolving landscape of specialized chip architecture in Semiconductors.
Live Poll
Do you believe partnerships between tech firms will successfully solve AI hardware performance limitations?
