- SK hynix won the IT/Electrical category at the 2023 R&D 100 Awards for its Key Value Computational Storage Drive (KV-CSD)
- Featuring the world’s first application of indexing technology, the KV-CSD assigns key values to data to significantly speed up data classification and processing
- Due to its ultra-fast performance, the product is expected to contribute to AI, big data, and high-performance computing
- SK hynix’s technical team behind the KV-CSD will strive to make its technology a global standard to enhance the company’s competitiveness
SK hynix won the IT/Electrical category at the 2023 R&D 100 Awards held in San Diego, California on November 16 for its next-generation Key Value Computational Storage Drive (KV-CSD).
Referred to as the “Oscars of Innovation” by those in industry and academia, the R&D 100 Awards is an annual science and technology awards ceremony that recognizes the year’s 100 most innovative technologies and products in the world.
▲SK hynix’s KV-CSD featuring the industry’s first indexing technology and a controller that supports its data computation
SK hynix collaborated with the U.S.-based Los Alamos National Laboratory (LANL)1 for the KV-CSD project, as the company handled product development while LANL created the system that applied the KV-CSD to supercomputers.
The product has garnered significant attention in the high-performance computing (HPC) field for its rapid read/write speeds that enables it to process several petabytes (PB)2 of data in minutes, as well as its self-computing capability which ensures swift data analysis.
The KV-CSD also grabbed headlines for its world-first adoption of indexing technology which dramatically increases data retrieval speed. When storing a single file, data is normally split and stored in multiple blocks3. However, indexing technology applies unique keys to the split data so that the indexing criteria can be used to quickly collect and process the data. Consequently, KV-CSD is up to 7.4 times faster than conventional storage devices such as solid-state drives (SSDs) which collect data and search for it block-by-block.
1Los Alamos National Laboratory (LANL): A national laboratory under the U.S. Department of Energy which conducts research on national security, nuclear fusion, and space exploration.
2Petabyte: A petabyte is equal to 1 million gigabytes.
3Block: A space where data is stored in a storage device. The smallest unit of data storage is a cell, and a collection of cells is a page. A group of pages is called a block. A storage device such as an SSD is composed of several blocks.
“As a next-generation storage device that delivers only the desired results to the CPU after completing rapid processing and data analysis based on key values and indexing technology, the KV-CSD will prove to be valuable for HPC, AI, and big data industries that process large amounts of data,” said Project Leader Woosuk Chung of the Computational Storage team at SK hynix. “We will focus on commercializing the product so that it can be used in various fields.”
A Meaningful Step Forward: Presenting Advanced Storage Tech to the World
▲ Members of SK hynix’s Computational Storage team who developed the KV-CSD (from left to right); Woosuk Chung, Soonyeal Yang, Inhyuk Park, and Taejin Oh
The next generation of storage devices is being developed based on several key concepts. These include processing data in a location close to the CPU, having memory itself compute data rather than the CPU, and many more. Storage devices that compute data themselves are called computational storage drives (CSDs), and KV-CSD falls in this category.
The Computational Storage team claims that receiving the award was even more special as it took a lot of trial and error to integrate new technologies into existing CSDs during the development of the KV-CSD.
“We are proud to have received a global award that is coveted by many global research institutes and companies,” said Chung. “Above all, winning the award is significant as it introduced SK hynix’s outstanding storage technology to the world.”
Solving Customers’ Pain Points: Reducing Data Processing Time
The KV-CSD further enhances SK hynix’s reputation as a solution provider as the product solves a crucial pain point for customers by dramatically reducing big data processing time.
▲ Team Leader Woosuk Chung points out the strong features of KV-CSD
“Our customer LANL’s main issue was that the processing speed would slow down when large amounts of data was stored and analyzed due to unnecessary data reading,” explained Chung. “The solution was to increase the processing speed of the CSD, so we proposed a storage device based on key values and indexing.”
▲ Soonyeal Yang talks about the distinct features of the KV-CSD
In particular, indexing technology was instrumental. On top of going through the tough task of implementing key-value storage, the team broke new ground to perform the world’s first integration of indexing technology in a storage device.
“With conventional SSDs, it takes tens of hours to extract desired data from LANL’s own research simulations that generate several PBs of data,” said Technical Leader Soonyeal Yang. “However, the KV-CSD locates and reads key-value data directly from the indexed list, reducing the data processing time to just several minutes.”
“Additionally, as it is the job of the CPU to scan through PBs of data to retrieve desired data, this process puts a strain on the system,” Yang added. “However, KV-CSD’s indexing technology speeds up data processing while minimizing the burden on the system”.
▲ Inhyuk Park explains a core technology of the KV-CSD called secondary indexing
Among the indexing technologies, Technical Leader Inhyuk Park emphasized the importance of secondary indexing. This technology subdivides the indexing list to further categorize data. “For example, if you have a database of products sold by each time of day, general indexing will only allow you to find out which products were sold at specific times,” explained Park. “But with secondary indexing, you can also find out the person who bought the product in addition to how much they paid for it. Ultimately, a range of data can be processed more quickly.”
▲ Taejin Oh recalls the challenges of developing a test application for the KV-CSD
Nevertheless, the team members agree that uncertainty was the biggest challenge in achieving this latest milestone. As they were building a completely new product, understanding new technologies, keeping up with trends, coming up with an architectural design, and everything related to developing the product were all fresh challenges for the team. More notably, the members faced several critical moments during the performance validation that came after the development of the prototype.
“As a new type of storage device, there was no test application that matched the product,” said Technical Leader Taejin Oh. “After spending a long time to reform existing test applications to adopt properties of the KV-CSD, our team eventually succeeded in developing our own test application. In the end, we could finally verify the performance of the product.”
Leading the Future of Storage Device Technology
Chung has high hopes for his team’s creation. “We aim to ensure SK hynix’s technology becomes a global standard to enhance the company’s position in the market,” he said. “We will continue to conduct follow-up research and standardization work, including joint efforts to build an ecosystem for the product.”
The Computational Storage team at SK hynix is determined to use this experience of growth to further excel in its research and development projects, and hinted at more innovations to come: “We will continue to surprise the market with our pioneering next-generation storage devices.”
