A chip startup founded in 2018, Kepler Computing, has officially stepped out of the shadows with a public debut on September 10th. The company's core mission is to dismantle the persistent challenges in the AI semiconductor sector, which include memory energy consumption, bandwidth limitations, and the high costs tied to manufacturing capacity.

Kepler claims its novel memory solution offers bandwidth on par with SRAM while providing superior storage capacity compared to both SRAM and HBM. This breakthrough is achieved without the need for costly EUV lithography equipment, a significant departure from traditional manufacturing processes.

The technology is built on a 3D stacked integration architecture, which points to the use of FeRAM (ferroelectric RAM). The company's roadmap indicates that samples will be available by the end of 2026, with production scaling slated for 2027. Looking further ahead, Kepler plans to leverage material science innovations to push past the inherent constraints of current CMOS technology, aiming to improve energy efficiency and boost production yield through lower operating voltages.

To bring this vision to life, Kepler is currently utilizing GLOBALFOUNDRIES Inc.'s 28nm production lines in Singapore for pilot runs, with future plans to manufacture on American soil. The company asserts that its process of retrofitting existing fabs for this novel memory production took only 8 months, a timeline that is roughly one-third of the industry standard.