Sandisk and SK hynix Release First Open Compute Project Specification to Accelerate High Bandwidth Flash Adoption for AI
Sandisk Corporation (NASDAQ: SNDK) and SK hynix Inc. have announced a major milestone in the evolution of AI memory technologies with the publication of the first High Bandwidth Flash (HBF™) technical specification through the Open Compute Project (OCP). The release represents an important step toward establishing an open industry standard for HBF technology, a next-generation memory solution designed to support the growing demands of artificial intelligence inference workloads.
The specification arrives only six months after the HBF consortium was formed in February, highlighting the rapid pace of collaboration among leading technology companies working to create a common framework for high-performance flash memory. Developed under the Open Compute Project’s HBF technology workstream, the specification provides hardware manufacturers, semiconductor companies, cloud providers, and AI accelerator developers with standardized technical guidelines for integrating HBF into future computing platforms.
Sandisk and SK hynix served as the primary contributors to the technical specification, while Google and Tenstorrent joined the consortium during the development process and played significant roles in validating the technology and refining the proposed standard. Their collective efforts are expected to help establish HBF as an open, interoperable technology capable of supporting the rapidly expanding AI ecosystem.
Addressing the Growing Memory Demands of AI
Artificial intelligence applications have evolved rapidly over the past few years, particularly with the widespread adoption of large language models, generative AI platforms, recommendation engines, autonomous systems, and advanced analytics. These workloads require enormous volumes of data to be processed at increasingly higher speeds.
Traditional memory architectures often struggle to provide the ideal balance between bandwidth, storage capacity, power efficiency, and cost. While High Bandwidth Memory (HBM) delivers exceptional performance, it remains relatively expensive and capacity constrained for many inference workloads.
High Bandwidth Flash has been designed to bridge this gap.
HBF combines the scalability and density advantages of NAND flash with substantially higher bandwidth than traditional storage solutions. Positioned close to AI compute engines, HBF provides significantly larger memory capacity while helping reduce latency and improve throughput during AI model inference.
As AI systems continue processing larger datasets and increasingly sophisticated foundation models, demand for high-capacity memory located near processors is expected to grow substantially. HBF aims to satisfy these requirements while improving power efficiency and reducing infrastructure costs for hyperscale data centers.
Open Standard for the Industry
The newly released technical specification establishes a common framework that enables system designers to develop compatible products without relying on proprietary implementations.
Among the key areas covered by the specification are:
- System interface definitions
- Electrical interface requirements
- xPU-to-HBF host interface standards
- Reliability requirements
- Packaging recommendations for HBF die stacks
- Software guidelines for read and write operations
- Performance expectations
- Integration requirements for AI accelerators
By defining these technical parameters, the specification enables equipment manufacturers and semiconductor companies to build interoperable products while simplifying integration into future AI infrastructure.
Industry experts believe that open standards are essential for accelerating innovation because they reduce compatibility issues and encourage broader ecosystem participation.
Collaboration Across the AI Ecosystem
One of the most notable aspects of the initiative is the broad collaboration among leading technology companies.
While Sandisk and SK hynix initiated and led the HBF workstream, Google and AI hardware developer Tenstorrent joined the consortium during the specification’s development.
Their participation helped validate practical deployment scenarios and ensured the specification reflects real-world AI infrastructure requirements.
Rather than designing a technology for a single vendor’s products, the consortium focused on building an open ecosystem that allows multiple companies to innovate around a common standard.
The Open Compute Project has increasingly become a preferred platform for developing open hardware standards across cloud infrastructure, storage, networking, and AI computing. Publishing the HBF specification through OCP makes the documentation freely available to developers and organizations worldwide.
Supporting AI Inference at Scale
Unlike AI model training, which occurs periodically using massive computing clusters, AI inference involves continuously serving trained models to users.
Inference workloads demand rapid access to enormous datasets while maintaining low latency and high throughput.
As organizations deploy increasingly sophisticated language models and AI assistants, inference has become one of the largest consumers of computing resources in modern data centers.
HBF has been specifically designed to address this challenge.
By placing large-capacity flash memory much closer to AI processors, HBF enables systems to access significantly larger datasets without relying exclusively on expensive high-bandwidth DRAM solutions.
This architecture can improve:
- Response times
- Model serving efficiency
- System throughput
- Power efficiency
- Infrastructure scalability
- Total cost of ownership
These improvements are particularly valuable for hyperscale cloud providers operating thousands of AI servers.
Leadership Perspective
According to Alper Ilkbahar, Chief Technology Officer at Sandisk, AI inference is fundamentally changing memory architecture requirements.
He explained that HBF technology has been created specifically to meet these emerging demands by delivering both high bandwidth and high capacity close to compute resources.
Ilkbahar noted that the new specification gives system designers a practical roadmap for integrating HBF into future AI platforms while supporting flexible memory architectures. He described the release as an important milestone not only for the HBF ecosystem but also for the next generation of AI systems focused on improving performance and reducing operational costs at scale.
Coexistence with High Bandwidth Memory
Rather than replacing existing High Bandwidth Memory technologies, HBF is intended to complement them.
The specification outlines architectures where HBF can coexist alongside HBM, allowing system designers to allocate different types of memory according to workload requirements.
This hybrid approach enables organizations to optimize both performance and economics.
For example, HBM may continue handling latency-sensitive computations, while HBF provides significantly larger near-compute storage capacity for inference models requiring extensive contextual data.
Such flexibility gives hardware vendors additional options when designing future AI servers and accelerator platforms.
Driving Industry Standardization
The release of the specification represents one of the first comprehensive technical standards focused specifically on High Bandwidth Flash.
Open standards often play a critical role in accelerating technology adoption because they reduce uncertainty for hardware manufacturers and software developers.
By making the specification publicly available through the Open Compute Project, Sandisk and SK hynix aim to encourage widespread industry participation.
Their long-term objective is to establish HBF as the preferred high-capacity memory layer for AI inference infrastructure.
Early publication also allows technology vendors to begin designing compatible hardware before commercial deployment scales significantly.
Building an Open Ecosystem
The companies emphasized that ecosystem development remains central to their strategy.
Rather than limiting HBF technology to proprietary implementations, they are encouraging broad collaboration among semiconductor manufacturers, cloud providers, AI infrastructure companies, accelerator developers, and software vendors.
An open ecosystem can accelerate:
- Product development
- Interoperability
- Software optimization
- Hardware innovation
- Customer adoption
- Market expansion
As additional organizations join the consortium, HBF standards are expected to continue evolving based on industry feedback and deployment experience.
Featured at the Future of Memory and Storage Conference
Sandisk will showcase HBF technology during multiple sessions at the Future of Memory and Storage (FMS) Conference in Santa Clara, California.
The company’s keynote presentation will examine how NAND flash technologies continue evolving to support AI inference at scale through system-level optimization.
The session will feature Sandisk executives Jim Elliott, Chief Revenue Officer, Khurram Ismail, Chief Product Officer, and Alper Ilkbahar, Chief Technology Officer.
The discussion will explore how advanced flash memory architectures contribute to improving AI infrastructure performance while reducing deployment costs.
Industry Panel on Breaking the Memory Wall
Sandisk, SK hynix, and Google will also participate in a dedicated panel discussion focused on High Bandwidth Flash.
Hosted by Thomas Coughlin, President of Coughlin Associates, the session will examine how HBF could redefine future memory hierarchies by combining near-memory performance with the scalability and persistence of flash storage.
Panelists will discuss architectural integration, technical challenges, standardization timelines, economic considerations, performance expectations, and broader industry adoption.
The discussion is expected to provide valuable insights into how HBF technology may influence next-generation AI infrastructure.
About Sandisk
Sandisk is a leading developer of flash storage solutions and advanced memory technologies serving consumer, enterprise, industrial, and cloud markets. The company focuses on delivering innovative products that improve data storage performance, scalability, and reliability across a wide range of applications. With decades of expertise in NAND flash innovation, Sandisk continues investing in technologies designed to support emerging workloads including artificial intelligence, machine learning, cloud computing, and edge applications.
Forward-Looking Statements
The announcement includes forward-looking statements regarding future expectations for High Bandwidth Flash technology, industry standardization efforts, customer adoption, AI infrastructure growth, and anticipated technological benefits.
These projections are based on current assumptions and involve numerous risks and uncertainties that could cause actual outcomes to differ materially. Factors such as economic conditions, changes in global trade policies, pricing fluctuations, supply chain disruptions, competitive technologies, product development timelines, cybersecurity risks, customer demand, regulatory changes, technological transitions, intellectual property considerations, strategic partnerships, manufacturing challenges, inflation, currency movements, and broader market conditions could influence future results.
Sandisk notes that investors should carefully review the company’s filings with the U.S. Securities and Exchange Commission for additional information regarding these and other potential risks. The company undertakes no obligation to revise or update forward-looking statements except where required by applicable law.
Source link: https://www.businesswire.com/