- Sandisk Top Inclosure Tips to shape a Flash -based memory alternative for AI
- HBF memory supported by RISC and GPU leaders promises a high bandwidth and massive capacity
- Patterson and Koduri join SanDisk to guide the expansion of flash memory beyond the HBM limits
Sandisk has appointed two main IT figures to help shape its high -capacity memory technology for AI workloads.
Professor David Patterson and Raja Koduri joined the new Sankisk technical advisory advice to provide a strategic and technical contribution on a high bandwidth (HBF), an alternative based on a flash with high bandwidth memory (HBM).
Patterson is credited for the co-development of reduced instructions instruction (RISC) and a redundant range of cheap records (RAID), and will direct the Advisory Council. Koduri is known for its leadership in graphic architecture, having supervised GPU conceptions at AMD and Intel.
Decades of experience
Together, they bring decades of experience in IT, memory systems and large -scale architecture.
“We are honored to have two distinguished experts in IT architecture to join our technical advisory advice,” said Alper Ilkbahar, Executive Vice-President and Director of Technology at Sandisk.
“Their collective experience and strategic advice will help shape HBF as a future memory standard for the AI industry, and say that we are not only meeting the expectations of our customers and partners.”
Patterson said: “HBF shows the promise to play an important role in Datacenter’s AI by offering unprecedented memory capacity to a high bandwidth, allowing the workloads to evolve far beyond today’s constraints. It could reduce the costs of new AI applications which are currently not disorders.”
Koduri added: “HBF is set to revolutionize the peak tunes by equipping aircraft with capacity with memory capacity and bandwidth that will support sophisticated models operating locally in real time. This advancement will unlock a new era of intelligent edge applications, fundamentally changing how and where AI inference is carried out. ”
HBF is designed to match the HBM bandwidth while offering up to 8 times the capacity at a similar cost.
Built with BICS Flash, CBA WAFER BONDING and Owner Stacking which allows 16 matrices per package, HBF offers a new way to extend GPU memory without fully leaning on an expensive dram.
Although it is not a direct replacement for HBM, HBF shares the same electric interface and only requires minimum protocol changes.
Sandisk previously demonstrated how an AI GPU using only HBM could support 192 GB of memory, but by combining it with HBF, this figure could reach 3 TB.
In a configuration using only HBF, memory capacity could evolve up to 4 TB.
The technology was revealed for the first time during the SanDisk FWD 2025 investor event in February 2025, in parallel with its roadmap for future HBF generations.
These updates show an increase in capacity and bandwidth over time, with certain compromises in terms of energy efficiency.
By forming an advisory advice and looking for open standard development, Sandisk tries to avoid locking the market in proprietary solutions.
This can help gain ground against rivals like Samsung and SK Hynix which are both strongly invested in the HBM space.
