Yuanxu Semiconductor’s Micro-LED optical interconnect chip achieves a measured 3-dB bandwidth of 3.5 GHz.

Recently, Yuanxu Semiconductor, in collaboration with Tsinghua University and Beijing Xinghuo Chenguang Technology Co., Ltd. (hereinafter referred to as “Xinghuo Chenguang”), has achieved a milestone in the research and development of Micro-LED optical interconnect chips.

According to joint tests conducted by Tsinghua University and Xinghuo Chenguang, a Micro-LED chip sample achieved a −3 dB bandwidth of 3.5 GHz at a current density of 500 A/cm²—a metric that quantifies the device’s high-speed response capability.

The present results demonstrate that, under the test conditions employed, the jointly developed Micro-LED chip samples exhibit response characteristics on the order of >3 GHz, providing empirical evidence to support the subsequent co‑design of driver circuits, receiver modules, and optical coupling schemes.

The development of AI computing power is driving the evolution of short-range interconnect technologies.

As artificial intelligence training, inference, and high-performance computing continue to advance, data exchange between computing chips—and between computation and storage—is becoming increasingly frequent. System efficiency depends not only on the computational capabilities of the chips but is also constrained by data‑transfer bandwidth, latency, and energy consumption. In scenarios such as within server cabinets, at the board level, and in near‑package integration, the demands for short‑reach, high‑density interconnects are growing ever more stringent.

Under high-speed, high-density conditions, copper interconnects face constraints such as transmission loss, reach limitations, and the power consumption of signal compensation, driving the industry to explore optical interconnect solutions tailored for short-reach scenarios. Micro-LED optical interconnects leverage a large number of relatively low‑speed optical channels operating in parallel to carry data, offering a technical pathway distinct from the few ultra‑high‑speed channels and holding the potential to optimize bandwidth density and energy efficiency. This approach is primarily geared toward short‑reach applications, complementing other light‑source‑based solutions to address diverse distance requirements and system needs.

Tripartite collaboration: bridging the extended ecosystem, chips, and integrated packaging.

In this project, Yuanxu Semiconductor is focusing on the R&D of Micro-LED chip fabrication processes, mass transfer technologies, and glass‑substrate‑integrated optoelectronic packaging. Tsinghua University and Xinghuo Chenguang are concentrating on Micro-LED epitaxial growth and high‑speed chip design, among other technologies. By leveraging each partner’s strengths and integrating epitaxial optimization, chip fabrication, and optoelectronic integration with packaging and testing feedback, the project ensures that materials and device research can iteratively evolve in tandem with downstream integration requirements.

MOG platform technology: accumulated expertise underpinning the R&D of optical interconnect integration.

Yuanxu Semiconductor’s R&D in optical interconnects is built on Micro-LED chips and optoelectronic integrated packaging. By leveraging the technologies accumulated through its MOG (Micro-LED on Glass) glass‑substrate integration platform, the company is pioneering new AI‑driven applications and establishing a foundational cornerstone for critical AI components. The significance extends beyond merely achieving high‑speed light‑emitting chips; it also lies in progressively developing engineering capabilities that span from individual devices to arrays and ultimately to optoelectronic integration.

In the next phase, Yuanxu Semiconductor will continue R&D focused on device consistency, array integration, and package‑level interconnects, while collaborating with partners to advance driver, receiver, and optical‑coupling adaptation. Through link‑level testing, it will continuously validate communication performance, thereby extending this chip‑level achievement into application‑oriented product development.