Kepler Computing launches FeRAM-based AI memory solution
芯片新创 Kepler 亮相：利用 FeRAM 与 3D 堆叠突破 AI 内存瓶颈
Kepler's FeRAM-based memory could solve the bandwidth and power consumption bottlenecks that limit the performance of AI models, offering a more efficient alternative to current solutions.
Kepler Computing, a chip startup founded in 2018, announced its FeRAM-based AI memory solution on September 10. The company claims its solution offers bandwidth close to SRAM, capacity better than SRAM and HBM, and does not require EUV lithography. The technology uses 3D stacking integration and is expected to be in a prototype by the end of 2026, with mass production in 2027. The company is currently using GigaDevice's 28nm fab in Singapore for pilot production and plans to expand to the US.
芯片新创 Kepler 亮相：利用 FeRAM 与 3D 堆叠突破 AI 内存瓶颈
IT之家 9 月 14 日消息，成立于 2018 年的芯片初创企业 Kepler Computing 当地时间本月 10 日正式宣布亮相。该企业的目标是化解 AI 半导体领域当前存在的内存能耗与带宽、产能与成本瓶颈。 Kepler Computing 宣称其可提供带宽接近 SRAM、容量优于 SRAM 乃至 HBM 的内存方案，同时还无需 EUV 光刻设备。根据介绍， 这一方案采用 3D 堆叠集成技术 ， 应该属于 FeRAM（铁电存储器） ，将于 2026 年底出样，2027 年扩大产能。 而在未来，该企业计划利用材料创新突破现有 CMOS（互补金属氧化物半导体）的固有限制，通过更低电压改进芯片的能效表现并提升生产能力。 Kepler Computing 正利用格罗方德的新加坡 28nm 产能试产，未来也将在美国制造。企业宣称其改造既有晶圆厂用于创新内存生产的过程仅用时 8 个月，是一般情况下的 1/3。
