初创公司 Kepler 宣称用铁电存储(FeRAM)+3D堆叠在格罗方德28nm成熟产线上做出HBM替代方案，计划2026年出样、2027年在新加坡投产

[News] Kepler Claims HBM Alternative Without EUV Using GlobalFoundries’ 28nm Fabs, 2027 Production
As the AI boom strains supply and pushes up costs, startups are exploring new ways to expand memory production. According to TechPowerUp, Kepler Computing claims to have developed an alternative approach to HBM based on a new 3D stacking technique and novel materials, with production designed around mature nodes rather than cutting-edge processes.
Notably, while modern memory manufacturing typically relies on EUV lithography, Kepler plans to use mature-node capacity and ferroelectric RAM (FeRAM) technology for its memory products. OC3D notes that Kepler’s ferroelectric approach is designed to “approach the bandwidth per watt of SRAM” while offering up to 10 times the capacity of SRAM and HBM.
Kepler’s technology can reportedly be manufactured at a retrofitted GlobalFoundries 28nm fab, with the conversion taking an estimated eight months, potentially shortening the path from technology development to volume production compared with building a new advanced fab.
Kepler has so far processed around 2,000 wafers using its new technology, with its first samples expected later in 2026. The company aims to begin production runs at GlobalFoundries’ Singapore facility in 2027, followed by U.S. manufacturing in 2028, according to TechPowerUp.
Kepler says its use of FeRAM could maintain standard HBM capacity while significantly lowering manufacturing costs. OC3D notes that the company is using materials science to develop the technology and claims to have already achieved the “fundamental breakthroughs” needed to make the new memory approach possible.
Kepler Targets a Lower-Cost Path to Challenge the HBM Status Quo
The use of older manufacturing processes could also make capacity expansion faster and less expensive. As demand for HBM surges alongside generative AI and large language models, major memory makers including Samsung, SK hynix, and Micron have committed billions of dollars to expanding production. According to Commercial Times, Kepler argues that building new DRAM or HBM fabs is both costly and time-consuming, while converting existing facilities could provide a faster, lower-cost way to add memory capacity. If the technology proves viable at scale, it could also introduce a different manufacturing model into a market currently dominated by established memory suppliers.
According to TechPowerUp, Kepler has also secured around $468 million in funding over the past several years from GlobalFoundries, Intel Capital, and AMD Ventures, with Intel and AMD potentially becoming future customers. Still, the report notes that Kepler has yet to demonstrate production yields and technical viability at scale.
Read more
- [News] Hot Chips 2026: Samsung’s zHBM Claims 70% Power-Efficiency Gain; SK hynix Evaluates Intel EMIB
- [News] Intel Patent Reveals XBM Matching HBM4 Footprint Without Interposers; Commercialization Seen After 2030
(Photo credit: Kepler on X)