News
TSMC introduces C-HBM4E using N3P process node for enhanced power efficiency in AI memory
Sunday, January 25, 2026 at 06:14 AM
TSMC has introduced C-HBM4E, a custom High Bandwidth Memory technology featuring a logic die manufactured on the N3P process node. This transition from the N12 process used in standard HBM4 base dies is expected to double power efficiency, shifting AI memory from pure storage to integrated compute capabilities.
Context
TSMC has unveiled its next-generation C-HBM4E technology, a custom high-bandwidth memory solution that integrates advanced N3P (3nm) logic dies directly into the memory stack. This shift marks a pivotal evolution in AI hardware, moving memory from simple storage to an active "computation" domain. By utilizing the N3P node for the base die instead of the N12 (12nm) process used in standard HBM4, the company aims to deliver a 2x increase in power efficiency.
This technical leap is achieved through a voltage reduction from 0.8V to 0.75V and the direct integration of memory controllers, which are traditionally housed on the host processor. These enhancements allow for higher bandwidth and better thermal management, critical for the massive power demands of next-gen AI clusters. TSMC and its partners, including SK Hynix and Micron, target a market rollout between 2026 and 2027, positioning the foundry to dominate the high-end supply chain for future flagship GPUs and TPUs.
Sources (2)
Related Companies
TSMC
2330