One of the biggest bottlenecks currently facing the AI industry is the memory wall. An American startup named Kepler Computing, after seven years of low-key R&D, has made a bold move, claiming it will mass-produce a new memory superior to HBM next year.
Kepler Computing stated they would abandon their previous low-profile approach, officially launching an X account to announce their major product: a new frontier-grade memory solution. Achieves bandwidth per watt close to SRAM without requiring EUV processes, while offering more than 10 times the capacity of SRAM and HBM memory.
They have secured $468 million in funding and built a dedicated wafer fab. Memory samples will be released this year, with mass production scheduled for 2027, and expansion in the US planned for 2028.
The new memory they mention is likely some form of 3D RAM, though specific details have not yet been released. To prevent suspicions of fundraising fraud, Kepler Computing emphasized that their team co-led one of the industry's earliest 3D chip stacking technologies, led and expanded the seventh generation of DRAM memory and multiple logic designs, produced billions of chips, and built the world's first commercial physical synthesis tools and technologies, which are core to the logic chip design industry. They also addressed long-standing challenges with ferroelectric materials and invented logic devices surpassing CMOS technology.
Seeing this mysterious, low-key, yet highly sophisticated promotional style, my first thought was whether this company was founded by someone from a major South Asian country. I checked the founder team on the company's page, and sure enough, it was.
Of course, their team includes members of other ethnicities, including White, Black, and Asian individuals. Their resumes indicate they are technical experts in the semiconductor industry, Often cited as pioneers or inventors of certain technologies, though the specific companies where these developments occurred are rarely mentioned.
Regarding Kepler Computing's chip products, it should be some form of 3D memory, Achieves ultra-high density and performance through chip stacking, allowing production without expensive EUV processes.
However, 3D memory technology is no longer new; various revolutionary products were announced under its banner over a decade ago. While promotion is easy, the real challenge lies in mass production. The company's claim that they can build a wafer fab with just $468 million in funding is somewhat baffling, as this amount might not even be enough for tape-out and production at an advanced process foundry.