Kepler Aims to Launch Energy-Saving Replacement for HBM in 2027
Kepler targets 2027 production for 3D ferroelectric memory promising 5–10× better bandwidth per watt than HBM. The post Kepler Aims to Launch Energy-Saving Replacement for HBM in 2027 appeared first on EE Times.
Kepler Computing is developing a 3D ferroelectric memory technology that aims to replace HBM and SRAM in AI chips, with production expected in 2027. The technology promises significant improvements in bandwidth per watt and could address current memory bottlenecks in AI applications.
What happened
Kepler Computing is developing a 3D ferroelectric memory technology that could serve as an energy-efficient replacement for HBM and SRAM in AI chips. The company plans to start production in 2027 and has partnered with GlobalFoundries for manufacturing. The technology uses monolithic stacking to avoid thermal problems and reduce the need for EUV lithography, potentially addressing current memory bottlenecks in AI applications.
Why it matters
The development of energy-efficient memory solutions is critical for AI applications, which require high bandwidth and low power consumption. Kepler's technology could alleviate the strain on data centers by reducing energy use and improving memory scalability, which is essential for the continued growth of AI computing.
- Kepler's 3D ferroelectric memory offers significant improvements in bandwidth per watt over HBM.
- The technology uses monolithic stacking to avoid thermal issues and reduce EUV dependency.
- Partnerships with industry leaders like GlobalFoundries are crucial for large-scale manufacturing.
The bigger picture
Kepler's 3D ferroelectric memory could lead to more energy-efficient AI chips, reducing the power demands of data centers and improving memory scalability. This could enable the development of more powerful AI systems without the need for extensive infrastructure upgrades.
Kepler Computing is advancing memory technology with its 3D ferroelectric memory, aiming to replace HBM and SRAM in AI chips by 2027. The technology's energy efficiency and scalability could address critical challenges in AI data centers, making it a promising development for the industry.
Technical details
- 3D ferroelectric memory technology
- monolithic stacking of devices within a chip
- avoiding thermal problems through 3D architecture
- compatibility with CMOS
- reduction in reliance on EUV lithography
- backend of line process for higher memory density
FAQ
What is Kepler's 3D ferroelectric memory technology?
Kepler's 3D ferroelectric memory is a new type of memory designed to replace HBM and SRAM in AI chips. It uses monolithic stacking to improve bandwidth per watt and reduce thermal issues, offering significant energy efficiency gains.
How does Kepler's technology differ from traditional HBM and SRAM?
Kepler's technology uses 3D packaging and monolithic stacking to achieve higher memory density and bandwidth per watt compared to HBM. It also avoids the need for EUV lithography, reducing manufacturing costs and complexity.
What are the potential benefits of Kepler's memory for AI applications?
Kepler's memory could reduce the energy consumption of AI data centers, improve memory scalability, and alleviate the strain on power grids. It also offers higher capacity and bandwidth than traditional SRAM and HBM.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by EE Times on Oct 5, 2026.
Comments
Sign in to join the discussion.