SK Hynix has developed PIM technology for memory

About a year ago, Samsung launched the HBM2 memory, the integrated AI processor that can provide up to 1.2 TFLOPS of embedded computing power, Enabling the memory chip itself to perform the operations of a CPU, GPU, ASIC, or FPGA. Samsung said at Hot Chips 33 that it will extend PIM (processing-in-memory) technology to DDR4, DDR5, LPDDR5X, GDDR6, and HBM3 memory in the future.

SK Hynix, another major DRAM manufacturer in South Korea, announced today that the next-generation memory semiconductor technology with computing functions has been developed, and the first product based on this technology is GDDR6-AiM (Accelerator-in-Memory/Memory Accelerator), add compute power to GDDR6 at 16 Gbps. At ISSCC 2022 (IEEE International Solid-State Circuits Conference) held on February 24, SK Hynix will disclose the development results of PIM chip technology.

Compared with the general GDDR6 standard working voltage (1.35V), the working voltage of GDDR6-AiM is reduced to 1.25V, plus the reduction of data transmission with CPU and GPU, the power consumption is reduced by 80%. This reduces the carbon footprint of the equipment and improves ESG performance. SK Hynix said that compared with traditional DRAM, a system combining GDDR6-AiM with CPU and GPU can increase the calculation speed up to 16 times in a specific computing environment. The new products are expected to have a wide range of applications in machine learning, high-performance computing, big data computing and storage.

SK Hynix believes that in the future, memory semiconductors can play a more central role in ICT products such as smartphones, and even realize “Memory Centric Computing”. Next, SK Hynix plans to cooperate with SAPEON Inc. to introduce new technologies combined with artificial intelligence semiconductors.

Both Samsung and SK Hynix seem to see PIM technology as the key to future memory semiconductor development. At present, some manufacturers have used the AXDIMM memory with PIM technology to test on the server, and the latter is more concerned about the combination with AI technology and ITC products.