On February 16, 2022, SK Hynix (or 'Company', www.skhynix.com) announced today that it has developed the next generation of memory semiconductor technology with computing capabilities, "PIM (processing-in-memory, processing in memory)" 1.
So far, it is widely recognized that storage semiconductors are responsible for storing data, while non-storage semiconductors (such as CPU or GPU) are responsible for processing data. In spite of this, SK Hynix is still actively exploring innovation in the field of new-generation smart memory, and then publicizing the company's research and development achievements in related fields for the first time.
SK Hynix plans to announce the development of PIM chip technology at the most prestigious international academic conference in semiconductor field-ISSCC (International Solid-State Circuits Conference, International Conference on solid State Circuits 2022 held in San Francisco at the end of February. With the continuous development of PIM technology, SK Hynix expects storage semiconductors to play a more central role in ICT products such as smart phones, and even successfully realize "memory centric computing (Memory Centric Computing)" in the future.
SK Hynix also developed a sample of the company's first PIM-based product, GDDR6-AiM (Accelerator-in-Memory, memory Accelerator 3). GDDR6-AiM is a product that adds computing capabilities to GDDR6 memory with a data transfer speed of 16Gbps. Compared with traditional DRAM, the system that combines GDDR6-AiM with CPU and GPU can increase the computing speed up to 16 times in a specific computing environment. GDDR6-AiM is expected to be widely used in machine learning, high-performance computing, big data computing, storage and other fields.
The operating voltage of the GDDR6-AiM is 1.25V, which is lower than the standard operating voltage of GDDR64 memory (1.35V). Not only that, the application of PIM also reduces the data transmission with CPU and GPU, thus reducing the energy consumption of CPU and GPU, thus GDDR6-AiM successfully reduces power consumption by 80%. SK Hynix believes that with the low energy consumption characteristics of the product, the company will be able to reduce carbon emissions from equipment and improve the company's ESG operating results.
In addition, SK Hynix also plans to work with SAPEON Inc., an artificial intelligence semiconductor company recently spun off from SK Telecom. Work together to launch a technology that combines GDDR6-AiM with artificial intelligence semiconductors. SAPEON Inc. CEO Liu Xiujing said: "in recent years, with the rapid increase in the use of artificial neural network data, there is a growing demand for calculus techniques that optimize related computing characteristics. We will combine the technologies of the two companies to maximize efficiency in terms of data calculation, cost and energy consumption. "
An Hyun, vice president of SK Hynix solution Development, said: "based on PIM technology with independent computing capabilities, SK Hynix will build a new ecosystem of memory solutions through GDDR6-AiM. The company will strive to constantly adjust its business model and advocate a new direction of technological innovation. "
