Views: 399 Author: Anna Publish Time: 2025-10-15 Origin: Site
The key metrics for evaluating data center optical modules are density, power consumption, and cost. Thermal capacity is one of the indicators used to measure optical module power consumption; the greater the thermal capacity, the greater the power consumption the module can withstand. 400G optical modules are primarily categorized by package type: CDFP, CFP8, QSFP-DD, and OSFP. CDFP and CFP8 are larger in size and have higher thermal capacity, primarily used in the telecommunications market. QSFP-DD, backwards compatible with the previous QSFP-28, offers the smallest size and higher density. It is more suitable for short-distance data center use and has numerous supporters, including companies such as Facebook, Alibaba, and Tencent. Supporters of the OSFP MSA include Google and Arista. OSFP is slightly larger than QSFP-DD. QSFP-28 optical modules require an adapter to be compatible with OSFP sockets. OSFP is backwards compatible with 800G and includes a built-in heat sink, capable of supporting 12W-15W thermal capacity. OSFP is particularly well-suited for the telecommunications market.
Data centers primarily encompass three application scenarios: intra-cabinet ToR switch and server interconnection, inter-cabinet switch interconnection, and data center interconnection (DCI).
ToR switch and server interconnection offers two solutions: direct-attach copper cables (DACs) and active optical cables (AOCs). DAC transmission distances decrease as network speeds increase. DACs are categorized into active copper cables (ACCs) and passive copper cables (PCCs). 400G PCC passive copper cables support a maximum reach of 2.5 meters, while 400G ACC active copper cables offer a maximum reach of 5 meters.
The advantage of DACs is their low price, but their disadvantages are also significant: short reach, bulky cables, and difficulty managing them. With increasing data center server network speeds and server density, DACs pose a significant challenge to server cabinet heat dissipation.
AOCs offer advantages such as light weight, long reach, electromagnetic interference (EMI) resistance, and easy cable management. Using multimode fiber, AOCs have a theoretical maximum transmission distance of 150 meters. However, due to the use of double-sideband modules, AOCs are not suitable for cross-row deployment. AOCs are generally used for distances less than 30 meters.
There are currently four 400G solutions for spine-leaf or leaf-to-ToR switch interconnection in data centers. 400G SR16, due to its high fiber count, is not adopted by most hyperscale data center users and will not be discussed here.
First, from the perspective of optical module cost, the 400G SR8 solution uses mature 25G baud rate VCSEL laser chips on the market and employs PAM4 pulse amplitude modulation. 25G baud rate VCSEL laser chips are already very mature in the market, so 400G SR8 optical modules have the lowest cost.
The 400G SR4.2 optical module uses dual-wavelength 25G baud rate VCSEL laser chips and requires a 2:1 mux combiner and a 1:2 demux, which increases the cost of the optical module. Furthermore, the 400G SR4.2 optical module ecosystem is not very complete, with only one optical chip supplier. Therefore, the cost of 400G SR4.2 optical modules is significantly higher than that of 400G SR8 optical modules.
400G DR4 optical modules use relatively expensive DML lasers or SiPh silicon photonics technology. Silicon photonics can integrate traditional optical components such as modulators and detectors onto a silicon substrate using a complementary metal oxide semiconductor (CMOS) process, significantly reducing the power consumption, size, and packaging costs of optical modules. However, silicon photonics is still in its early stages of development, with low shipment volumes, preventing it from achieving economies of scale. According to a recent LightCounting research report, overall shipments of modules based on silicon photonics are significantly lower than those of traditional optical modules based on indium phosphide (InP) or gallium arsenide (GaAs).
Currently, mainstream switch ASIC chips use 25G PAM-4 signals, while 400G DR4 optical modules use 50G PAM-4 signals, which translates to 100Gbps per wavelength. To ensure consistent rates between switch electrical signals and optical module signals, 400G DR4 optical modules require a gearbox to convert 8x50Gbps to 4x100Gbps. This gearbox increases cost and power consumption, making 400G DR4 optical modules the most expensive of the three solutions. 400G DR4 is suitable for switch interconnection distances of 150-500 meters.
From a cabling perspective, 400G DR4 utilizes 8-core single-mode fiber for parallel transmission, and uses APC (angled 8-degree) MPO/MTP-12 or MPO/MTP-8 connectors. The angled 8-degree connector reduces return loss, lowering overall fiber link loss and ensuring bit error rates.
400G SR4.2 uses 8-core multimode fiber for parallel transmission, and uses MPO/MTP-12 or MPO/MTP-8 connectors. 400G SR8 uses 16-core multimode fiber for parallel transmission. To minimize return loss, the connectors are MPO/MTP-16 connectors with an 8-degree APC endface. The MPO-16 connectors feature an offset keying design to prevent confusion with MPO/MTP-12 or MPO/MTP-8.
For new data centers, to support 400G SR8 optical modules, it is recommended to use an MPO/MTP-16 fiber cabling system, which directly supports 400G SR8 Ethernet.
For data centers currently using MPO/MTP-12 or MPO/MTP-24 cabling, a smooth upgrade to 400G SR-8 can be achieved by replacing the 4x3 MPO/MTP converter box.
From an application perspective, 400G SR-8 supports the widest range of fan-out options, including 4x100G, 8x50G, and 2x200G fan-out. Both 400G SR4.2 and 400G DR4 support 4x100G fan-out.
The total cost of ownership (TCO) for data center networks includes both optical module and cabling costs. While 400G SR-8 requires a relatively high fiber count, data centers generally adopt modular designs, and switch interconnection distances are typically around 50 meters. Therefore, cabling cost differences are much lower than those for optical modules. Furthermore, the 400G SR-8 ecosystem is the most comprehensive, with the largest number of optical module vendors and support for the most fan-out applications. Therefore, for newly deployed data centers, 400G SR-8 offers the lowest TCO within distances of 115m or less, making it the most cost-effective solution for 400G switch interconnections.
For data center interconnects within the 2km-10km range, there are four options. 400G FR8 requires eight DML lasers. To reduce the number and complexity of optical module lasers, 400G FR4 optical modules only require four EML lasers, lowering both device and assembly costs. 400G FR8 utilizes LAN-WDM wavelength division multiplexing technology, supporting eight wavelengths, each transmitting 50Gbps. 400G FR4 utilizes CWDM technology, supporting four wavelengths, each capable of 100Gbps. In the current market, achieving 100Gbps per wavelength requires the use of expensive 50Gb/s optical and electrical chips.
There are three options for achieving 100GDR transmission. Option 1 uses a 4:1 transmission to convert the switch's 4x25bps electrical signal into a 1x100Gbps optical module. The optical module consumes 3.5W of power. Option 2 uses a 2:1 transmission to convert the switch's 4x25bps electrical signal into a 1x100Gbps optical module. The optical module consumes 2.5W of power. Option 3 uses a single-wavelength 100Gbps transmission. This optical module does not require an expensive transmission, and its power consumption is reduced to 1.5W.
The advantage of single-wavelength 100Gbps transmission is that it avoids the use of expensive transmissions, reducing component costs and also lowering optical module power consumption. Therefore, with improvements in chip manufacturing technology and increased shipments of 50G baud rate chips, 400G FR4 will eventually replace 400G FR8 in the market.
400G FR4 uses a 20nm wavelength spacing. A larger wavelength spacing relaxes the requirements for multiplexers and demultiplexers, eliminates the need for laser cooling, and reduces optical module costs. 400G LR4, however, has a wavelength spacing of only 5nm, requiring a TEC (electrical cooling system) to control temperature. Consequently, 400G LR4 optical modules are more expensive than 400G FR4 modules. Therefore, for distances within 2km, 400G FR4 is a more economical solution.
For data center interconnects within 40km, 400G ER8 or 400G ER4 can be used. 400G ER8 uses expensive EML lasers and APD detectors, as well as multiplexers and demultiplexers. Furthermore, EML lasers consume a lot of power, and to ensure stable modulation signals, a TEC (electrical cooling system) is required. These factors contribute to the high costs of both 400G ER8 and 400G ER4 optical modules. The difference between 400G ER4 and 400G ER8 is that 400G ER4 uses a single 100Gbps wavelength, which translates to expensive 50Gbaud EML laser chips, while 400G ER8 uses the currently mature 25Gbaud EML laser chips. Furthermore, 400G ER4 optical modules need to convert 8x50Gbps electrical signals on the switch side into 4x100Gbps signals. Therefore, 400G ER4 requires an additional transmission unit compared to 400G ER8, resulting in a significantly higher price.
With the continuous increase in network speeds, traditional direct detection methods based on simple on-off keying amplitude modulation will broaden the optical signal spectrum, leading to crosstalk and thus reducing transmission distance.
Coherent detection communications are infiltrating data center interconnect scenarios, moving beyond long-distance communications. Coherent detection is a more complex, multi-dimensional signal modulation method that combines amplitude modulation, phase modulation, and polarization modulation to carry more information per bit.
The Optical Internetworking Forum (OIF) has developed specifications for 400G ZR, which combines coherent detection with dense wavelength division multiplexing (DWDM). 400G ZR utilizes the more complex polarization-multiplexed 16-level quadrature amplitude modulation (DP-16QAM) scheme, enabling transmission distances of up to 80km. 400G ZR optical modules require expensive ITLA integrated tunable laser components, IQM integrated IQ phase modulators, high-performance DSP digital signal processors (to compensate for chromatic and polarization dispersion), and ICR integrated coherent receivers (to detect complex modulated optical signals). The price of a 400G ZR optical module is approximately twice that of a 400G ER4 module.