Ricore, a chip developer from Suzhou, was named a computational ecosystem partner for consumer robot manufacturer PrimeBot during PrimeBot's product presentation in Shanghai on September 20. Both companies announced their intention to integrate silicon solutions with a data streaming architecture for inference in humanoid devices at the edge.
This chip, named Radiant D35, is positioned as an edge AI inference device. It combines a fully developed RISC-V processor core with a data streaming compute structure designed for physical AI tasks such as recognition, motion control, and multimodal reasoning, while adhering to strict power consumption and latency constraints.
Unlike traditional flow control designs that move tensors between memory and ALU upon instruction issuance, the data streaming architecture activates operation as soon as operands arrive. This reduces energy movement and latency, which are dominant in many inference graphs. The chip uses a Chiplet layout, allowing computation levels to be configured for specific scenarios, and comes with a full software stack for major deep learning frameworks.
The device is promoted due to its high-capacity local memory—according to company materials, this can reach about 128 GB of high-speed memory for on-device models that do not require cloud access. Ricore also reported that its team holds dozens of patents on data streaming and received a nomination for best scientific paper at ISCA 2026. Furthermore, its RISC-V core has been included in the global procurement list of an international semiconductor company after passing technical certification.
Sources such as People.cn and Shanghai Securities News interpreted the collaboration with PrimeBot as a deepening of internal supply chains in embodied technologies, affecting not only the assembly of complete robots but also the silicon level, aligning with the general political support for RISC-V industrialization in AI and embedded systems.
However, the official partnership announcements did not disclose detailed specifications regarding sustainable TOPS, power consumption parameters, or PrimeBot SKU compliance. Nevertheless, this move signals that a major humanoid robot manufacturer has chosen a RISC-V based data streaming inference chip as a partner for edge computing, rather than simply acquiring a ready-made test board.
