Established in early 2021, Kiwimoore , based on on Chiplet and RDMA high-performance interconnection communication technologies, provide intergrated interconnection architectures and product solutions for AI datacenter server clusters.

The company's core products include UCIe Die2Die IP, NDSA series DSA network acceleration dielets, IO Die, and other interconnection chiplet solutions, helping customers build the next-generation computing platform more efficiently and at a lower cost.

Kiwimoore is one of the earliest high-tech enterprises in China involving in Chiplet & heterogeneous integration technology. Currently, our interconnection architecture solution for AI computing centers mainly addresses the pain points of interconnection challenges among network clusters , Chip to Chip and Die to Die communications in AI datacenter.

For Die to Die interconnection, we have launched the Die2Die interconnection and IO Die interconnection chiplet series. This year, based on the UCIe standard, we have launched the world's first batch of Die2Die IP supporting UCIe V1.1, with an interconnection speed of up to 32 Gbp and a latency as low as several nanoseconds, easy to plug-and-play.

In terms of IO Die interconnection, we have conducted strategic cooperation with Ventana Micro Systems, a well-known RISC-V CPU company, and provided I/O Die for its second-generation server CPU - Veyron V2. Each dielet includes 32 cores. Multiple dielets are based on the chiplet architecture and connected to the high-performance IO Die provided by Kiwimoore through the UCIe interface, thereby achieving up to 192 cores and supporting various field accelerators.

Regarding communication between GPUs, our under-research high-performance network acceleration DSA - Kiwi NDSA, based on Ethernet RoCE V2 high-performance RDMA and dozens of offloading/acceleration engines, can be used as an independent dielet in the transmission accelerator of GPU/NPUs. The GPU Link Chiplet "NDSA-G2G" for GPU interconnection builds a high-speed data exchange network between chips through RDMA and Die2Die technologies, which can achieve ultra-high-speed data transmission of Tb/s level. Its performance reaches the world's leading edge tech. The solution can be a substitute to NVIDIA NV-link, enhancing the computing power and transmission capacity of domestic AI chips.

Kiwinoore has received orders worth hundreds of millions in China currently. Our innovative technologies have been recognized by some customers in the ecosystem. At the same time, we are a domestic fabless design companie that could provide a full-range interconnection architecture solution for datacenter server clusters, which is still rare in then domestic market.
Kiwimoore is semiconductor pioneeri in China dedicated to Chiplet technology. Our co-founders and the team attach great importance to building a more open-source semiconductor industry ecosystem, and empower the vigorous development of the domestic Chiplet industry by joining international and domestic interconnection standard committees.
We joined the international interconnection D2D standard committee UCIe in 2022 and continuously input the technical concepts and interconnection requirements of related standards in the industry. Secondly, we have been jointly working with well-known domestic EDA manufacturers, advanced packaging foundries and related partners to make positive contributions to domestic AI chip design，tape out and mass production.