Recently, the world of semiconductors is undergoing a crucial technological transition. An internationally renowned Korean expert has predicted a significant shift from exclusive use of HBM to the combined adoption of HBF, driven by the explosive growth in computational demands of artificial intelligence.
How Voice AI Amplifies Memory Demand
Professor Kim Jeong-ho, globally recognized as the founder of HBM technology at the Korea Advanced Institute of Science and Technology, recently reiterated the increasing importance of AI’s cognitive processing capabilities. During a briefing on HBF technological development strategy, the Korean researcher emphasized how the evolution from text-based interfaces to voice-based ones will necessarily lead to a substantial increase in data volumes to be processed.
This transition is not accidental. Voice interaction in AI requires processing significantly larger amounts of data compared to text, creating unprecedented pressure on current memory systems. According to the Korean professor, current HBM will not be capable of handling this explosion of requests, making the adoption of HBF not only desirable but inevitable.
Technological Evolution: From Current Systems to MCC Architecture
Today’s architecture involves the vertical connection of up to two GPUs with adjacent HBM to handle workloads. However, the Korean expert argues that this configuration will quickly reach its physical limits.
The proposed solution involves the simultaneous use of HBM and HBF on the same systems, a strategy that could eliminate current bottlenecks. But the true revolution will come with MCC (Memory-Centric Computing) architecture, which organically integrates CPU, GPU, and memory on a single base chip. This innovation represents a paradigm shift: memory will no longer be a separate component but the central core around which all other computational functions are organized.
With the implementation of MCC, HBF capacity demands will further increase, exacerbating the need for this new memory standard.
2038: The Year of HBF Supremacy According to Korean Experts
The Korean researcher provided an intriguing timeline: starting from 2038, the demand for HBF is expected to surpass that of HBM. This is not a random figure but an estimate based on the development trajectory of AI, the scaling of memory systems, and the availability of manufacturing technologies.
The Korean expert’s analysis suggests that the coming years will be a transitional period, where HBM and HBF will coexist. But looking beyond 2038, the semiconductor memory market could decisively tilt toward HBF, marking the end of an era and the beginning of a new phase in high-bandwidth memory systems.
This prediction by Korean researchers offers a clear roadmap for the industry: plan not only for current needs but for the inevitable evolution of computational requirements in the next decade.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
The Korean Scientist's Vision: From HBM to HBF Under the Pressure of Artificial Intelligence
Recently, the world of semiconductors is undergoing a crucial technological transition. An internationally renowned Korean expert has predicted a significant shift from exclusive use of HBM to the combined adoption of HBF, driven by the explosive growth in computational demands of artificial intelligence.
How Voice AI Amplifies Memory Demand
Professor Kim Jeong-ho, globally recognized as the founder of HBM technology at the Korea Advanced Institute of Science and Technology, recently reiterated the increasing importance of AI’s cognitive processing capabilities. During a briefing on HBF technological development strategy, the Korean researcher emphasized how the evolution from text-based interfaces to voice-based ones will necessarily lead to a substantial increase in data volumes to be processed.
This transition is not accidental. Voice interaction in AI requires processing significantly larger amounts of data compared to text, creating unprecedented pressure on current memory systems. According to the Korean professor, current HBM will not be capable of handling this explosion of requests, making the adoption of HBF not only desirable but inevitable.
Technological Evolution: From Current Systems to MCC Architecture
Today’s architecture involves the vertical connection of up to two GPUs with adjacent HBM to handle workloads. However, the Korean expert argues that this configuration will quickly reach its physical limits.
The proposed solution involves the simultaneous use of HBM and HBF on the same systems, a strategy that could eliminate current bottlenecks. But the true revolution will come with MCC (Memory-Centric Computing) architecture, which organically integrates CPU, GPU, and memory on a single base chip. This innovation represents a paradigm shift: memory will no longer be a separate component but the central core around which all other computational functions are organized.
With the implementation of MCC, HBF capacity demands will further increase, exacerbating the need for this new memory standard.
2038: The Year of HBF Supremacy According to Korean Experts
The Korean researcher provided an intriguing timeline: starting from 2038, the demand for HBF is expected to surpass that of HBM. This is not a random figure but an estimate based on the development trajectory of AI, the scaling of memory systems, and the availability of manufacturing technologies.
The Korean expert’s analysis suggests that the coming years will be a transitional period, where HBM and HBF will coexist. But looking beyond 2038, the semiconductor memory market could decisively tilt toward HBF, marking the end of an era and the beginning of a new phase in high-bandwidth memory systems.
This prediction by Korean researchers offers a clear roadmap for the industry: plan not only for current needs but for the inevitable evolution of computational requirements in the next decade.