Founded in 2018, the semiconductor startup Kepler Computing officially emerged from stealth on September 10. The company aims to address critical challenges in AI hardware, specifically focusing on memory bandwidth, energy efficiency, production capacity, and manufacturing costs.
Kepler Computing is developing a memory solution based on 3D stacking and FeRAM technology. The company claims this architecture offers bandwidth comparable to SRAM, while providing higher capacity than both SRAM and HBM. The design notably eliminates the need for EUV lithography equipment.
The startup is currently utilizing GlobalFoundries' 28nm production capacity in Singapore for trial manufacturing. Kepler Computing reports that it successfully retrofitted existing wafer fabrication facilities for its memory production in just 8 months, which is one-third of the industry standard time.
Looking ahead, the company plans to leverage material innovations to overcome the inherent limitations of CMOS technology. By implementing lower voltage operations, Kepler aims to further enhance chip energy efficiency and increase overall manufacturing output.
The company has set a clear roadmap for its product development. It expects to provide initial samples of its memory technology by the end of 2026, with plans to scale up production capacity significantly throughout 2027. Future manufacturing plans also include domestic production in the United States.
