Samsung built artificial intelligence into a memory chip
Samsung Electronics did the development of the first High Bandwidth Memory (HBM) announced module that is integrated with an artificial intelligence module - HBM-PIM. The Processing-in-Memory (PIM) architecture The interior of the high-performance memory modules is primarily intended for accelerating data processing in database centers, high-performance computing (HPC) systems and mobile applications.
Our HBM PIM is the first programmable PIM solutionwhich was developed for various artificial intelligence applications. We plan to expand our partnerships with providers of artificial intelligence solutions so that we can come up with more and more advanced solutions PIM can offer said Samsung Vice President Kwangil Park.
Image source: Pixabay
Most modern computer systems are based on theNeumann architecture, in which data is stored and processed in separate circuits. This architecture requires the constant transfer of information between the chips, which - especially in systems with large amounts of data work - represents a bottleneck that slows down the entire computer system HBM PIM architecturer assumes that the data processing module is used at the location where the data is located. This is done by placing a AI engine in each memory module that is optimized for use with DRAM chips. This enabled parallel processing and minimized the need for data transfer.
Samsung reports that the combination of HBM PIM With the HBM2-Aquabolt solution, which is already on the market, it enables the output to be doubled while reducing the power requirement by 70 percent. The Koreans emphasize that HBM-PIM does not require any hardware or software changes, so it can be easily applied to existing systems. HBM-PIM delivers impressive results performance improvements with a huge reduction in energy consumption for important AI solutions. We are excited to test his performance in solving the problems we are working on in Argonne, "said Rick Stevens, director of computational, environmental and life sciences at Argonne National Laboratory. Details about HBM-PIM will be presented on February 22, 2021 at the International Solid-State Circuits Virtual Conference (ISSCC) vorgestellt.