Add AI to memory? Samsung does this

Add AI to memory? Samsung does this

Recently, Samsung demonstrated its latest progress in processing in memory (PIM) technology at the Hot Chips 33 conference. As an important conference in the semiconductor industry, the Hot Chips 33 conference has high-profile microprocessor and IC innovative products appear every year.

Samsung Semiconductor HBM-PIM

The information released by Samsung includes that Samsung successfully integrated PIM-based high-bandwidth memory (HBM-PIM) into a commercial accelerator system for the first time, and expanded the scope of PIM applications to DRAM modules and mobile memory, thereby accelerating the integration of memory and logic. .

Samsung integrates HBM-PIM into AI accelerator for the first time

In February of this year, Samsung launched its first HBM-PIM (Aquabolt-XL), which integrates AI processing capabilities into Samsung HBM2 Aquabolt to enhance the high-speed data processing capabilities of supercomputers and AI applications. Subsequently, HBM-PIM was tested in Xilinx's Virtex Ultrascale+ (Alveo) AI accelerator. It improved system performance by nearly 2.5 times and reduced energy consumption by more than 60%.

“HBM-PIM is Samsung’s first AI custom memory solution that has been tested in a customer’s AI accelerator system, showing great commercial potential.” said Nam Sung Kim, senior vice president of DRAM products and technology at Samsung Electronics. “With technology With the development of standardization, the application of this technology will be further expanded, extending to HBM3 for next-generation supercomputers and AI applications, even mobile storage for AI on equipment, and storage modules for data centers."

"Xilinx has been cooperating with Samsung Electronics, starting with the Virtex UltraScale+ HBM series, to provide high-performance solutions for data center, network and real-time signal processing applications. Recently, the two parties have launched the exciting Versal HBM series." Arun Varadarajan Rajagopal, senior director of Lingsi Product Planning Department, said, "I am very pleased to continue our cooperation with Samsung. We help evaluate the potential of the HBM-PIM system to achieve significant performance and energy efficiency in AI applications."

DRAM module driven by PIM

Accelerated DIMM (AXDIMM) can be "processed" in the DRAM module to minimize the large amount of data exchange between the CPU and DRAM to improve the energy efficiency of the AI ​​accelerator system. Because the AI engine is embedded in the buffer chip, AXDIMM can process multiple memory arrays (DRAM chipsets) in parallel instead of accessing only one set at a time, which greatly improves system performance and efficiency. Since the AXDIMM module can retain the dimensions of the traditional DIMM, it can be directly replaced without modifying the system. Currently, AXDIMM is being tested on the customer's server and can provide approximately 2 times the performance in AI-based recommended applications and reduce the energy consumption of the entire system by 40%.

Samsung Semiconductor AXDIMM

"SAP has been cooperating with Samsung for the application of in-memory database (IMDB) for SAP-HANA," said Oliver Rebholz, head of core research and innovation at SAP HANA. "Based on performance predictions and potential integration solutions, We expect that the performance of the in-memory database management (IMDBMS) will be significantly improved, and through the decomposition calculation on AXDIMM, higher energy efficiency will be achieved. SAP hopes to continue to cooperate with Samsung in this field."

Mobile storage brings AI from the data center to the device

Samsung LPDDR5-PIM mobile memory can provide independent AI functions without connecting to the data center. Simulation tests show that when LPDDR5-PIM is used in applications such as speech recognition, translation, and chatbots, it can increase performance by more than 1 times while reducing energy consumption by more than 60%.

Stimulate the vitality of the ecosystem

Samsung plans to expand its AI memory product portfolio by cooperating with other industry leaders to standardize the PIM platform in the first half of 2022. Samsung will continue its efforts to cultivate a highly sound PIM ecosystem to ensure the wide applicability of the entire memory market.

*Actual performance may vary depending on the device and user environment

*The product pictures and models, data, functions, performance, specifications, features, and other product information (including but not limited to product advantages, components, performance, availability, and capabilities) in this article are for reference only. Samsung may comment on the above The content is improved and can be changed without notice or restriction. For specific information, please refer to the actual product and product manual. Unless otherwise specified, the data involved in this article are the results of Samsung's internal testing, and the comparisons involved in this article are all compared with Samsung products


Previous
Next