Next-Generation Memory Technologies Preparing for the Post-HBM Era: The Rise of Magnonics and Vertical FeRAM

The memory industry has entered a supercycle due to the surge in HBM demand, but new technological breakthroughs are required due to the miniaturization limits of existing DRAM and the physical constraints of HBM. Analysis suggests that simple performance improvements are insufficient to secure a competitive edge, considering Nvidia's one-year product release cycle for the H100 and its software ecosystem dominance.

Limitations of HBM and New Demand

Currently, the yield of the HBM3e 12-layer stack is low at approximately 75%, and major players such as Samsung, SK Hynix, and Micron have completed supply contracts for the coming years. Demand is explosive, with the Stargate project alone requiring 40% of global DRAM production, yet existing DRAM miniaturization has already reached its limit. Consequently, incumbent companies are focusing on vertical stacking packaging, but this is evaluated as insufficient to resolve the fundamental speed and energy gap known as the 'memory wall'.

Candidate Next-Generation Memory Technologies

Two major technological directions are emerging as alternatives to existing HBM.

• Magnonics (Magnet Wave-Based Memory): Utilizes antiferromagnets to achieve ultra-fast dynamic characteristics at the 1-2 picosecond level and low power consumption. Recent research has improved the signal readout voltage from 100 nanovolts to over microvolts, but reaching 100 millivolts remains a challenge for commercial application.
• Vertical FeRAM (Vertically Stacked Ferroelectric Memory): A 2T-nC vertical strings structure has been proposed that can deliver HBM-level bandwidth while providing density similar to NAND. Micron invested $1 billion in related demonstrations but halted them due to degradation and cost issues, making it essential for startups to secure solid IP and manufacturing technology.

AI Acceleration and Commercialization Outlook

The advancement of AI tools such as Neural Network Potentials (NNPs) is narrowing the gap from material discovery to manufacturing scale-up. With the introduction of AI-based design tools, as seen in OpenAI's Jalapeño chip design case, development processes that previously required thousands of engineers, like those at Cerebras or Groq, are becoming feasible with small teams of around 50. This technological acceleration and the $1 trillion market opportunity increase the likelihood that heterogeneous approaches like Magnonics and Vertical FeRAM will be commercialized faster than expected.