With the emergence of HDC and super-large AI, PIM technology has emerged as an alternative for high-speed processing of massive data. It has been argued that securing demand is important for Korea, a powerhouse in memory semiconductors, to take the lead in the market.

On the 12th, the Ministry of Science and ICT announced its goal of increasing the domestic AI semiconductor market share in the domestic data center market to 80% and investing a total of 826.2 billion won by 2030 to improve it to the world's best level. To this end, it has revealed a plan to combine DRAM-based commercial PIM and domestic NPU by 2028.

The government and industry are focusing their attention on this 'PIM' technology. PIM (Process In Memory, intelligent semiconductor) is a chip that adds the processor function that performs computational work to the memory semiconductor that performs storage work.

Even if the processor's capabilities improve, it cannot perform well if the memory is not sufficient. That is why the domestic semiconductor industry came up with the idea of merging memory semiconductors and system semiconductors into chips specialized for AI operations.

■ 'PIM Strategy Technology Symposium' held… Sharing industry status and vision

Related research by government and industry is still in its infancy.

On the 13th, the '2022 PIM AI Semiconductor Strategy Technology Symposium' hosted by the PIM AI Semiconductor Business Group was held for the first time this year at the COEX Conference Hall in Seoul. Keynote speeches were given by Samsung Electronics, SK Hynix, Seoul National University professors, and KAIST professors, sharing the current status and outlook of the PIM AI Semiconductor market.

On this day, Samsung Electronics Master Son Kyo-min gave a keynote speech on the topic of 'Next Generation DRAM Solution and PIM', and SK Hynix Vice President Jeon Jun-hyun gave a presentation on the topic of 'Prerequisites for the Emergence and Success of PIM'. Following this, Seoul National University Professor Ahn Jeong-ho and KAIST Professor Jeong Myeong-su gave presentations.

Last year, Samsung Electronics unveiled 'HBM-PIM', which integrates PIM into HBM (high bandwidth memory).

Samsung Electronics announced that when equipped with HBM-PIM, performance is increased by about twice or more compared to systems using existing HBM2, and system energy is reduced by more than 70%.

Samsung Electronics' Master Son Kyo-min said, "PIM is an existing concept, but it is a technology that has recently begun to receive attention. Samsung Electronics is also focusing on researching it and developing CIM and PNM together," and announced plans to release HBM4 next year.

He also emphasized continuous collaboration with customers to commercialize PIM. “As standardization progresses, we will be able to provide solutions that satisfy many people,” he said, revealing Samsung Electronics’ ambitions.

Last June, SK Hynix introduced 'GDDR6-AiM', which has a graphic acceleration unit in DRAM. It was revealed that it reduces energy consumption by 80% and increases specific operation speed by up to 16 times. SK Hynix Vice President Jeon Jun-hyun added, "SK Hynix's PIM solution is 2.5 times faster than the CPU based on the current FPGA board."

The former vice president cited cost and development environment as the difficulties in commercializing PIM, arguing that “mass production will reduce costs, and it will be necessary to secure demand for large-scale PIM consumption.”

He also emphasized that “our main customers, hyperscalers, are driving cloud vertical optimization and driving decision-making at each technology layer, and that it is important to lead the market through collaboration among industry players.”

On that day, Seoul National University Professor Ahn Jeong-ho also said in a presentation on the early stages of PIM technology, “Rather than unconditionally pushing forward with PIM, we should think about killer applications that require PIM while understanding, analyzing, and optimizing constraints.”