The new architecture will offer twice the system performance
and reduce energy consumption by more than 70%
Samsung Electronics, the world leader in advanced memory technology, today announced that it has developed the first integrated High Bandwidth Memory (HBM) with artificial intelligence (AI) processing power – the HBM-PIM. The new in-memory processing (PIM) architecture brings powerful AI computing capabilities into high-performance memory to accelerate large-scale processing in data centers, high-performance computing systems (HPC) and AI-enabled mobile applications.
Kwangil Park, senior vice president of Memory Product Planning at Samsung Electronics stated: “Our innovative HBM-PIM is the industry’s first programmable PIM solution adapted for a variety of AI-driven workloads such as HPC, training and inference. We intend to develop this innovation by collaborating even more with AI solution providers for even more advanced applications with PIM. “
Rick Stevens, Argonne’s Associate Laboratory Director for Computing, Environment and Life Sciences, commented: “I am very happy to see that Samsung is facing the bandwidth / memory power challenges for HPC and AI computing. The HBM-PIM project demonstrated impressive performance and power gains in important classes of AI applications, so we look forward to working together to assess your performance on additional issues of interest to the Argonne National Laboratory. “
Most of today’s computing systems are based on von Neumann’s architecture, which uses separate processor and memory units to perform millions of complex data processing tasks. This sequential processing approach requires that data move constantly back and forth, resulting in a system slowdown bottleneck, especially when dealing with ever-increasing volumes of data.
Instead, HBM-PIM brings processing power directly to where data is stored, placing a DRAM-optimized AI mechanism within each memory bank – a storage subunit – allowing for parallel processing and minimizing data movement. When applied to Samsung’s existing HBM2 Aquabolt solution, the new architecture is capable of offering twice the system performance, reducing energy consumption by more than 70%. HBM-PIM also does not require any hardware or software changes, allowing for faster integration with existing systems.
Samsung’s HBM-PIM article was selected for presentation at the renowned International Solid State Circuit Conference (ISSCC) held until February 22. Samsung’s HBM-PIM is now being tested on AI accelerators by leading AI solution partners, with all expected to be validated in the first half of this year.