- SanDisk admits Top Computing-mind to shape flash-based memory alternative to AI
- Hbf -Memory supported by RISC and GPU leaders promise high bandwidth and massive capacity
- Patterson and Koduri end up to SanDisk to guide flash memory extension beyond HBM limits
SanDisk has appointed two leading figures in the calculation to help shape the direction of its high capacity to AI-workload.
Professor David Patterson and Raja Koduri are connected to SanDisk’s new technical advisory board to provide strategic and technical input on High Bandwidth Flash (HBF), a flash-based alternative to high bandwidth memory (HBM).
Patterson is credited for co-development of Reduced Instruction Set Computing (RISC) and superfluous series of cheap disks (RAID) and will lead the advisory board. Koduri is known for its leadership in graphics architecture, after monitoring GPU design at AMD and Intel.
Decades of experience
Together, they bring decades of experience across computing, memory systems and large -scale architecture.
“We are honored to have two distinguished computer architecture experts joining our technical advisory board,” said Alper Ilkbahar, CEO Vice President and Heading Technology Officer at SanDisk.
“Their collective experience and strategic councils will be instrumental in shaping HBF as the future memory standard for the AI industry and confirms that we not only meet but surpass our customers and partners.”
Patterson said: “HBF shows the promise of playing an important role in data center AI by delivering unprecedented memory capacity at high bandwidth, enabling the workloads to scale far beyond today’s limitations. It can reduce the cost of new AI applications that cannot be left.”
Koduri added, “HBF is set to revolutionize Edge AI by equipping devices with memory capacity and bandwidth features that support sophisticated models running locally in real time. This progress will lock a new era with intelligent edge applications, which fundamentally changes how and where AI inference is performed.”
HBF is designed to match the bandwidth of HBM while offering up to 8 times the capacity at a similar price.
Built with BICS -FLASH, CBA WAFER bonding and proprietary stacking, allowing 16 matrixes per day. Package, HBF offers a new way to expand the GPU memory without relying on expensive drama.
Although not a direct replacement for HBM, HBF shares the same electrical interface and requires only minimal protocol changes.
SanDisk previously demonstrated in the past how an AI GPU that uses only HBM may support 192 GB of memory, but by combining it with HBF this number could reach 3TB.
In a configuration that only uses HBF, memory capacity could scale up to 4TB.
The technology was first revealed on SanDisk’s Future FWD 2025 Investor Event back in February 2025 together with its roadmap for future HBF generations.
These updates show increases in capacity and bandwidth over time with some balance in energy efficiency.
By forming an advisory board and seeking open standard development, SanDisk tries to avoid locking the market in proprietary solutions.
This can help it get traction against rivals like Samsung and SK Hynix, both of which are heavily invested in the HBM room.
