Recently, at electronica China, the SmartSens booth attracted a large crowd. Visitors gathered around a prototype demonstration device transmitting data in real time—MicroLED single-channel transmission rate exceeded 3Gbps, with typical power consumption as low as 0.8pJ/bit. Wang Wenxuan, Co-General Manager of SmartSens High-Speed Optical Interconnect Business Group, revealed to eefocus reporters on-site, "The MicroLED high-speed optical interconnect solution is expected to be commercially deployed in 2027."
Wang Wenxuan, Co-General Manager of SmartSens HOC Business Group, introduces the progress of high-speed optical interconnect solutions to eefocus reporters
This chip design company, renowned for its CMOS image sensor products, is betting on a brand-new track. Behind this track lies an interconnect technology revolution currently unfolding within data centers.
Computing Power Surges, Interconnects Face Urgent Challenges
AI large model training is advancing from thousand-card clusters to ten-thousand, hundred-thousand, and even million-card scales, with computing chip performance doubling every two years. However, data transmission has become the core bottleneck determining the efficiency of computing systems.
The current AI industry faces the triple bottlenecks of the "bandwidth wall, power wall, and density wall," compounded by link reliability challenges. Specifically, HBM3/HBM4 memory bandwidth has exceeded 1.6TB/s, and single-card computing power has surpassed the PFLOPS level. Under 800Gbps/1.6Tbps system rates, traditional copper interconnects are limited by the skin effect, crosstalk, and impedance mismatch, making their bandwidth density unable to match the growth in computing power. In high-density AI clusters, the power consumption of 800G optical modules has reached 15-20W per module, and the proportion of interconnect device power consumption to the total system power consumption continues to rise.
Data sufficiently illustrates the urgency of the market. Citigroup research estimates that the global optical interconnect market size will grow from $22 billion in 2025 to $92 billion in 2028, with a three-year compound annual growth rate of 65%. TrendForce estimates that the CPO/NPO market size will increase from about $100 million in 2025 to over $39 billion in 2030. LightCounting predicts that 2026 will be a crucial year for the large-scale commercialization of silicon photonics and the batch deployment of CPO, with the industry iterating along the path of "NPO transition—CPO mainstream—OIO ultimate."
A trillion-dollar market transformation is unfolding, and the competitive landscape of the three technical routes is becoming increasingly clear.
Three Routes, Each with Its Own Strengths
1. Copper Interconnect: The "Veteran" Approaching Physical Limits
Relying on the advantages of low cost and mature processes, copper cables have long served as the standard solution for data center interconnects. However, as transmission rates advance toward 800G and 1.6T, the physical limits of copper interconnects are rapidly emerging.
At a rate of 112Gbps, the transmission distance of copper backplanes is about 2.5 meters; at 224Gbps, it drops to only about 1 meter. NVIDIA CEO Jensen Huang once pointed out at Computex Taipei 2025 that without optical interconnects, the scale ceiling for GPU clusters is 72 units—not because they don't want to scale up, but because copper cables won't allow it. The driving current required for copper cables to transmit high-speed signals grows exponentially with the rate, and power consumption increases as I²R; doubling the rate quadruples the power consumption. The skin effect and dielectric loss cause severe attenuation of high-frequency signals and serious crosstalk.
Of course, copper cables are not exiting the stage immediately. Technologies such as Active Copper Cables (ACC/AEC) are delaying their phase-out, and the high-speed copper cable market is expected to still reach shipments of 20 million cables by 2027. However, the long-term trend of "optics advancing, copper retreating" is irreversible.
2. Laser Silicon Photonics Interconnect: Current Mainstream, Accelerating Evolution
The silicon photonics solution is the absolute mainstay of current optical interconnects. In 2026, the market share of silicon photonics modules exceeded 50% for the first time. In June of the same year, NVIDIA's Vera Rubin platform entered full-scale mass production, and the supporting Spectrum-X Ethernet silicon photonics CPO switches were delivered at scale simultaneously, marking the commercial tipping point where AI computing clusters officially transition from "electrical interconnects" to "optical interconnects."
The advantages of the silicon photonics solution lie in its long transmission distance (up to the kilometer level), high bandwidth (single channels have achieved 200G), and mature supply chain. However, its disadvantages are equally significant: it requires external discrete devices such as lasers, DSP chips, and TIAs, resulting in high costs; lasers are sensitive to temperature and require TEC temperature control devices; and core components such as high-end silicon photonics chips and EML lasers remain monopolized by overseas companies.
Wang Wenxuan stated frankly in the interview, "Looking at the currently mainstream silicon photonics solutions in China, even if silicon photonics are developed, some core lasers and DSP chips still need to be purchased externally. There are very few domestic companies that can truly achieve full-link autonomous control and produce a complete commercial silicon photonics system. Many high-end components are still subject to foreign control."
3. MicroLED Optical Interconnect: Emerging Force, Poised for Takeoff
It is precisely against the backdrop of copper interconnects approaching their limits and silicon photonics solutions being constrained by others that MicroLED optical interconnect has emerged as a third route.
Unlike laser solutions, MicroLED adopts the principle of spontaneous emission rather than stimulated emission. Its threshold current is far lower than that of lasers, eliminating the need for external signal modulation devices, making the entire system simpler and lower in power consumption. More importantly, MicroLED is insensitive to temperature and does not require TEC temperature control devices—this characteristic gives it unique advantages in short-distance optical transmission scenarios between high-heat-generating chips such as GPUs and HBMs.
In terms of power consumption, the advantages of the MicroLED CPO solution are even more prominent. Taking 1.6Tbps transmission products as an example, the power consumption of current optical transceiver modules is about 30W. With this solution, it can be reduced to 1.6W, a drop of nearly 20 times, with unit transmission energy consumption of only 1-2pJ/bit. Compared to copper cable solutions, energy consumption can be reduced to 5% of that level.
Of course, MicroLED optical interconnect is currently still in the transition stage from laboratory verification to commercialization, with transmission distances mainly targeting short-distance scenarios within 50 meters. However, the ecosystem is not yet fully developed, and optical coupling efficiency and chip yield remain core technical bottlenecks.
And as Wang Wenxuan said, "The biggest challenge is that the maturity of the entire ecosystem chain is insufficient. The domestic industry has only started one or two years ago, but we adhere to open cooperation and deep collaboration with upstream and downstream partners. Improving maturity is just a matter of time."
The Logic Behind SmartSens' "Second Curve"
Why would a CIS chip design company choose to enter the MicroLED optical interconnect track?
From the perspective of technological homology, the answer is not hard to understand. The receiving end of MicroLED optical interconnect is a photodetector (PD) in the visible light band, which is highly homologous to the CIS technology that SmartSens has deeply cultivated for many years. Wang Wenxuan explained in the interview, "SmartSens is a leading CIS chip design company in the industry with rich accumulation in the field of light sensing, so we can maximize the performance of the receiving end.
The MicroLED optical interconnect system is a heterogeneous integration form of multi-component coupling, which has continuity with the heterogeneous integration technology of CIS."
From the perspective of the industrial ecosystem, the MicroLED supply chain is already relatively mature domestically—from substrate materials to chip manufacturing, China possesses globally leading production capacity and technological accumulation. This forms a sharp contrast with the situation in silicon photonics solutions where high-end chips are constrained by others.
From the perspective of market prospects, the optical module market continues to surge given the growth of AI, representing a rare incremental track. Wang Wenxuan stated, "The optical module sector will undoubtedly see continuously surging demand given the growth of AI, thus offering considerable long-term development dividends."
Meanwhile, SmartSens has not been developing in isolation. According to Wang Wenxuan, the company has engaged in deep connection with all links in the upstream and downstream supply chains at home and abroad, and joint R&D and technological collaboration have achieved substantial progress. "The MicroLED optical interconnect technology has a long chain and high R&D thresholds, covering many links such as substrate materials, CPO packaging, and system integration. It is difficult for a single enterprise to independently achieve full-process technological breakthroughs." SmartSens has adhered to an attitude of open cooperation since the very beginning of its layout.
Regarding the product timeline, SmartSens plans to launch MicroLED optical interconnect prototype products in the second half of 2026, with commercialization expected in 2027. The prototype sample making its debut at this exhibition can already transmit data in real time with high stability. Wang Wenxuan stated in this regard, "We expect to release some products close to commercial form in the second half of 2027. Subsequent iterations will be like CIS; going from 0 to 1 is the hardest part, and after that, it's just business."
Filling the Gap in the "Middle Ground"
The strategic value of MicroLED optical interconnect lies not only in the breakthrough of the technology itself but also in its precise entry into an overlooked market gap.
In the landscape of data center interconnects, copper cables dominate ultra-short-distance scenarios within 1 meter, while silicon photonics solutions rule long-distance transmission over 100 meters. However, in the "middle ground" of 1 to 50 meters—interconnects within racks, between racks, and between boards—copper cables fall short, and silicon photonics solutions are overkill.
MicroLED optical interconnect perfectly fills this gap. It adopts a "wide and slow" parallel architecture—replacing traditional few high-speed electrical channels with hundreds of parallel low-speed optical channels, fundamentally solving multiple pain points of electrical interconnects such as insufficient bandwidth, high power consumption, signal crosstalk, and limited transmission distance. Overall, it can support ultra-large parallel transmission bandwidths of 1Tbps+.
This technological positioning has also gained widespread recognition in the industry. In July 2026, BOE disclosed that it has established a project team for tackling key technologies in MicroLED optical interconnect systems and glass substrate CPO, and agreed that MicroLED optical interconnect technology is expected to break the inherent trade-offs between copper cables and optical links in data center networks.
From a broader perspective, the rise of MicroLED optical interconnect highly aligns with China's policy orientation toward the domestic substitution of computing infrastructure. The relevant guiding opinions on the domestic substitution of computing infrastructure issued by the MIIT in 2026 clearly proposed that the domestic substitution rate of core optical components in newly built intelligent computing centers must not be lower than 70%. Against the backdrop where core components of silicon photonics solutions still rely on imports, MicroLED optical interconnect, leveraging the mature domestic supply chain foundation, is expected to become a crucial breakthrough for domestic substitution.
Challenges and Prospects
Of course, from prototype to large-scale commercialization, SmartSens and the entire MicroLED optical interconnect industry still face numerous challenges.
At the technical level, single-channel rates still need to be further enhanced, optical coupling efficiency needs to be optimized, and chip yield needs to climb. At the supply chain level, from substrate materials and CPO packaging to system integration, the maturity and collaborative efficiency of each link still have room for improvement. At the market level, customers remain wait-and-see regarding the stability and reliability of the new solution, requiring time for verification.
Wang Wenxuan remains clear-headed about this: "We are still in the prototype verification stage. The overall system architecture is complete. The next step is to condense it into miniaturized products and solve specific problems in IC design and system integration." He also emphasized, "There are no insurmountable obstacles."
From a longer time scale, MicroLED optical interconnect is expected to become the mainstream solution for ultra-short-distance high-speed interconnects in the next 5-10 years. Whether SmartSens can successfully open up a second growth curve with MicroLED optical interconnect remains to be tested by the market. But what is certain is that in an era where AI computing power is surging, whoever seizes the first-mover advantage in the generational shift of interconnect technologies will have the opportunity to define the rules for the next round of infrastructure.
This race concerning "light" has just begun.