A team from Yokohama National University and others achieves 200Gb/s and 26 femtojoules per bit using a slow-light silicon photonic modulator

The slower light travels, the faster computing power is consumed. Source: Semiconductor Horizons | Published: 2026-10-23 14:00 | Technical Article | Universities | Generate Poster. Nanoscale holes in a silicon wafer slow light down, and a compact silicon photonic modulator achieves 200 Gb/s transmission while greatly reducing energy consumption.

A team composed of Yokohama National University, Institute of Science Tokyo, the National Institute of Advanced Industrial Science and Technology (AIST), and the Photonics Electronics Technology Research Association (PETRA) published a paper in Nature Communications demonstrating a silicon photonic modulator that uses the "slow light" effect: by etching nanoscale micro-holes in a silicon photonic crystal waveguide to slow the propagation of light, and integrating a built-in electro-optic equalizer to compensate for bandwidth loss, it pushes the single-channel data transmission rate to 200 Gb/s and reduces energy consumption per bit to 26 femtojoules (1 femtojoule equals 10 to the power of negative 15 joules). The paper is titled "Reduced-DSP 200G/λ low-voltage slow-light silicon modulator enabled by built-in EO equalizer," where "Reduced-DSP" means that the equalizer reduces reliance on external digital signal processing chips, further saving power.

The research background is the interconnect pressure in AI data centers. When chips are in close proximity, traditional copper cables are still adequate for short-distance connections; once the distance exceeds 1 meter, optical fiber must be used instead. High-speed electrical signals attenuate rapidly with distance in copper cables and generate a large amount of heat, whereas optical signals in optical fiber hardly attenuate and do not generate heat. One of the key links currently constraining data center transmission rates is the Mach-Zehnder modulator (MZM), a silicon-based microchip responsible for converting electrical signals into optical signals, and the related technology is approaching its physical limits. The paper's authors judge that further increasing the rate requires modulators with higher conversion efficiency and smaller size and energy consumption.

The team's core approach is to use "slow light" to enhance the interaction between light and the device. Silicon photonic crystal waveguides are made of nanostructures filled with tiny holes and can produce light that propagates more slowly, thereby significantly improving modulation efficiency. Previous slow-light modulators were limited by relatively large resistance, which restricted high-frequency bandwidth; the research team integrated an electro-optic (EO) equalizer into the device, compensating for the frequency response and raising the bandwidth to above 80 GHz, comparable to conventional silicon modulators.

The key results are: a modulator only 320 micrometers long achieves 140 Gb/s transmission in on-off keying (OOK) format with energy consumption of 26 femtojoules per bit; after switching to four-level pulse amplitude modulation (PAM-4), the maximum rate reaches 200 Gb/s. By comparison, conventional MZMs are typically 2 to 4 millimeters long and consume thousands of femtojoules per bit to achieve the same modulation rate. The new modulator is about an order of magnitude smaller in size and about two orders of magnitude higher in modulation efficiency.

Of greater industrial interest is process compatibility. The slow-light structure can be manufactured using CMOS-compatible processes, meaning large-scale mass production is possible without building expensive new production lines; by contrast, electro-optic modulators using next-generation materials, such as thin-film lithium niobate (TFLN), often require special bonding steps for materials other than silicon. This characteristic arrives exactly at an industrial generational window: AI data center optical modules are iterating from 800G to 1.6T, with the new-generation 800G and 1.6T products typically architected as 4×200G and 8×200G respectively, and 200G per channel is their common technological foundation. 1.6T products have now entered the stage of large-scale deployment, and domestic manufacturers Eoptolink and Innolight have both confirmed continued volume shipments of related products.

From the perspective of the industry landscape, silicon photonics remains the most mature photonic material solution in terms of industrialization and holds the highest market share, but the industry generally judges that silicon photonic modulators will approach bandwidth and power consumption boundaries at 400G per channel (corresponding to the 3.2T era), and thin-film lithium niobate is seen as the successor candidate; according to industry statistics, only three companies worldwide currently mass-produce high-end TFLN modulators, with Advanced Fiber Resources representing China. The value of this research lies in extending the silicon photonics route toward 200G/λ and even higher rates using mature CMOS-compatible processes, buying a window for silicon photonics to defend its generational share. At the same time, co-packaged optics (CPO) is becoming an important direction for high-density data centers: NVIDIA's Quantum-X and Spectrum-X series CPO switches have entered mass production and delivery, and TSMC's COUPE packaging platform has also started mass production. The paper's corresponding author, Toshihiko Baba, a professor in the Faculty of Engineering at Yokohama National University, said the architecture is expected to become a building block for future co-packaged optics.

It should be noted that the new device currently has relatively large optical insertion loss, which weakens the optical signal strength after insertion into a transmission link; the research team says this problem is expected to be minimized through design optimization. This achievement, with its combination of "slow light + built-in equalizer," strikes a balance among speed, power consumption, and size; whether it can move toward mass production still depends on subsequent loss optimization and yield performance.

This article is reproduced from media reports or online platforms and represents the author's personal stance or views. Our reproduction is solely for sharing and does not mean we endorse or agree with it. If the source attribution is incorrect or your lawful rights and interests are infringed, please contact customer service promptly, and we, as a neutral platform service provider, will promptly correct, delete, or handle it in accordance with the law. Comments: No user comments yet. Latest News More Related Recommendations

Samsung orders 2-nanometer equipment, Taylor plant adds 6,000 wafers per month of capacity 2026-10-23 16:33

The global LED display panel market will still achieve high value-added growth 2026-10-23 16:32

Northern Rare Earth's Q3 net profit attributable to parent company was 1.593 billion yuan, up 161.13% year-on-year 2026-10-23 16:31