Skip to main content

原粒半导体

Category: AI Chips / Semiconductors

An AI Chiplet (chiplet) company developing modular, high-efficiency AI inference chips for edge and end-side deployment of large multimodal models. 原粒半导体 was founded in 2023. The company is led by 方绍峡 (Fang Shaoxia). Based in Beijing, China. Team size: 101-500. Total funding raised: $75.5M. Key investors include IDG Capital, Wuyue Capital (武岳峰科创), China State-owned Capital Fund (国新基金), Shangshi Capital (尚势资本), Hongniao Qihang Fund (红鸟启航基金), Lingyi Investment (领屹投资), Shoufazhan Chuangtou (首发展创投), Hong Kong Mianhua Fund (香港木棉花基金), Innoangel Fund (英诺天使基金), CAS Star (中科创星), Yiwei Chuangtou (一维创投), Tsinghua Alumni Seed Fund (水木清华校友种子基金).

Founded
2023
Headquarters
Beijing, China
Team size
101-500
Total funding
$75.5M

Value proposition

Enables server-grade AI inference (up to 671B parameter models) to run locally on edge/desktop devices using modular chiplet architecture, eliminating cloud API costs, latency, and privacy concerns while providing unlimited local tokens.

Products and solutions

CCS-1 series (first end-side productivity AI chip, successfully taped out and lit up May 2026), AI Chiplet components (NPU chiplets, SoC chiplets), Zeus Gen2 AI accelerator chiplets, Modular "Lego-like" chiplet architecture for flexible edge AI deployment, Toolchain for multimodal large model deployment on edge devices

Unique value

"Lego-like" modular chiplet architecture that allows flexible, cost-effective configuration of AI inference chips for edge deployment, supporting native OpenClaw ecosystem and running 671B-parameter models on a desktop-sized device.

Target customer

SoC and system OEMs; edge computing device manufacturers; enterprise AI deployers needing local inference; AI agent hardware makers; PC/edge server vendors

Industries served

Edge AI computing, AI PC/desktop AI supercomputing, Industrial AI, Enterprise AI inference, AI Agent hardware

Technology advantage

Chiplet-based modular architecture (vs. monolithic die); self-developed high-speed die-to-die interconnect compatible with UCIe standard; multi-modal AI processor design; adaptive compute fusion technology; team with 10+ generations of AI chip architecture experience from AMD/Xilinx; first tape-out success (CCS-1) achieved on first attempt with full spec compliance

How they differentiate

Modular "Lego-like" chiplet architecture enabling flexible configuration; focus on edge/end-side deployment of large models (up to 671B params); team with deep AMD/Xilinx heritage; native support for Agent ecosystems; "100 real employees + 3000 digital employees" AI-native company model

Main competitors

寒武纪 (Cambricon Technologies), 地平线 (Horizon Robotics), 瑞芯微 (Rockchip) in edge AI inference, 芯原股份 (VeriSilicon) in chiplet/IP space

Key partnerships

超摩科技 (Chaomo Technology) - strategic partnership for NPU+CPU heterogeneous AI solutions, OpenClaw ecosystem support

Major milestones

2023-04: Company founded, 2023-06: Seed round (tens of millions RMB), 2024-04: Angel round, 2025-12: A round, 2026-04: Pre-A round of >500M RMB led by IDG Capital, 2026-04: A+ round, 2026-05: First chip CCS-1 successfully taped out and lit up (first-pass silicon success)

Market positioning

Early-stage AI chiplet startup focused on edge-side inference, differentiated by modular chiplet architecture vs. monolithic chip designs of competitors. Positioned as the "private jet" alternative to the "high-speed rail" approach of cloud AI chips.

Geographic focus

China (primary), with potential global expansion for edge AI chiplet solutions

Patents and IP

50+ AI chip patents (US & China) held by CEO Fang Shaoxia; multiple patents on AI processor architecture and chiplet design methodology

About 方绍峡 (Fang Shaoxia)

Ex-AMD Global Chip R&D Director; Ex-Xilinx AI Processor R&D Director; Ex-Deephi Tech (深鉴科技) Chip R&D Director & Chief Architect. PhD from Tsinghua University EE Department (first Innovation Leadership Engineering PhD graduate). 50+ AI chip patents (US & China).

Official website: