'Save+Compute' at Once... SK Hynix Develops Next-Generation Intelligent PIM (Comprehensive)

Maximizing Data Processing Speed... Reducing Energy Consumption by 80%
Prototype Applying PIM to Be Unveiled at World's Leading Conference

Next-generation memory semiconductor PIM developed by SK Hynix applied to 'GDDR6-AiM' (Photo by SK Hynix)

[Asia Economy Reporter Lee Hye-young] SK Hynix has developed a next-generation memory semiconductor (PIM) equipped with computing capabilities. This is the first result of SK Hynix's attempt to expand into the field of converged memory. SK Hynix also introduced a prototype of the ‘GDDR6-AiM,’ which applies PIM and reduces power consumption by about 80%, marking its expansion into new areas.

On the 16th, SK Hynix announced that it will reveal specific PIM development achievements at the ‘2022 International Solid-State Circuits Conference (ISSCC),’ the world's most prestigious semiconductor conference, held in San Francisco, USA, at the end of this month.

PIM is a semiconductor with next-generation technology that adds computing functions to memory semiconductors, solving the bottleneck caused by data movement in AI and big data processing fields.

Traditionally, memory semiconductors have been responsible for data storage, while computing functions, analogous to the human brain, have been handled by non-memory semiconductors such as central processing units (CPU) or graphics processing units (GPU). However, recently, global semiconductor companies have begun to break down these boundaries, intensifying competition in developing and launching converged semiconductors that combine memory and systems.

SK Hynix has also been dedicated to developing PIM semiconductors, recognizing this trend. This is because the share of system semiconductors such as AI in the global semiconductor market has exceeded 55%, making expansion in the memory sector inevitable. Currently ranking second in the global memory semiconductor market after Samsung Electronics, SK Hynix plans to further strengthen its growth engine through technological innovation led by PIM development.

As this technology evolves, it is expected to enable ‘memory-centric computing,’ where memory semiconductors play a central role in ICT devices such as smartphones.

Ahn Hyun, Vice President of Solution Development at SK Hynix, said, "SK Hynix will build a new memory solution ecosystem using PIM-based GDDR6-AiM with built-in computing functions," adding, "We will continue to evolve our business model and technology development direction."

Development of 'GDDR6-AiM' Reducing Power Consumption by About 80%

SK Hynix has developed a sample of ‘GDDR6-AiM’ as the first product applying PIM. This product adds computing functions to GDDR6 memory, which processes data at 16 gigabits per second (Gbps). When this product is mounted together with CPU and GPU instead of conventional DRAM, the speed of specific computations can be up to 16 times faster. In the future, GDDR6-AiM is expected to be used in machine learning, high-performance computing, big data computation, and storage.

Next-generation memory semiconductor PIM developed by SK Hynix applied to 'GDDR6-AiM' (Photo by SK Hynix)

Notably, this product operates at 1.25V, which is lower than the existing operating voltage of GDDR6 at 1.35V. Additionally, PIM, which performs its own computations, reduces data movement to CPU and GPU, thereby lowering power consumption. SK Hynix explained that its internal simulations confirmed a revolutionary reduction in energy consumption by about 80% compared to existing products. The company also expects this product to contribute to reducing carbon emissions from devices that incorporate it.

SK Hynix plans to accelerate technology development to expand the memory sector, including intelligent semiconductors. To this end, it recently announced plans to collaborate with Sapeon, an AI semiconductor company spun off from SK Telecom, to showcase technology combining GDDR6-AiM and AI semiconductors.

Ryu Su-jeong, CEO of Sapeon, said, "The use of artificial neural network data has rapidly increased recently, requiring computing technology optimized for such computational characteristics," adding, "By integrating the technologies of both companies, we will maximize efficiency in terms of data computation, cost, and energy usage."

This content was produced with the assistance of AI translation services.