Background of 800G Optical Module Technology
For 800G optical module interface technology, there are two main organizations: the 800G Pluggable MSA working group and the QSFP-DD800 MSA. Based on the 800G technology white paper published by the 800G Pluggable MSA working group, its main application scenarios are the 8x100G solution for SR short-reach scenarios around 100 meters in data centers, the 4x200G solution for FR scenarios around 2 km, and possible 800G solutions for DR scenarios around 500 m. Today we focus on the 8x100G technology in SR scenarios.

Current data center architectures can mainly be divided into two types. One is the three-tier hyperscale data center architecture, with a certain oversubscription ratio between layers, such as 3:1. Above the Spine layer, coherent optical interconnect solutions such as ZR are used to connect with other data centers, as shown below. This structure is a traffic-converging network, usually depending on the balance between service requirements and CapEx optimization.

The other is a two-tier data center network architecture based on AI clusters, as shown below. It has fewer layers than the first architecture, with no oversubscription between layers, which means larger data traffic. In this structure, each Leaf connects directly to server interfaces, but the traffic exchange is much smaller. At the same time, its deployment speed is faster than the first architecture, with almost no traffic convergence.

So in which scenarios do the above architectures require 800G interfaces? According to the white paper: in the first typical architecture, if the rate between Server and TOR reaches 200G, then 800G (4x200G) is needed further up. For the second data center architecture, when the traffic between Server and Leaf reaches 2x200G, the Leaf layer needs 800G bandwidth interfaces upward.
In addition, in the above two architectures, the distance from service to Leaf is usually 4 m within a rack / 20 m across racks, from TOR to Leaf is about 60-100 m, from Leaf to Spine is typically 500 m to 2 km, and finally from Spine to DCI and interconnects between them is around 80-120 km. As mentioned at the beginning, one goal of the MSA is to address 800G interface technology in SR scenarios, namely the SR 8x100G technology covering around 100 m, and the other is to address the FR and DR scenarios of 500 m to 2 km.
Technical Analysis of 8x100G in SR Scenarios
For the MSA and related vendors, their goal is to develop low-cost 8x100G optical modules, and this low-cost approach is to define a transmitter technology that reduces cost linearly through high integration. Given the trend of ever-increasing switch ports and decreasing server count per rack, low-cost 800G can also be used to support serial 100G server connections. Traditional interconnects between domestic data centers rely on VCSEL multimode transmission supporting 30-50 m, while the MSA side is based on single-mode fiber interconnect using 100G PAM4 technology. Instead of VCSEL-based multimode solutions, it adopts parallel single-mode fiber (SMF) transmission using PSM8 with PAM4 modulation format (with built-in DSP chip) to meet the 800G SR scenario requirements, as shown below.

To ensure the low-cost and low-power advantages of 800G-SR8 over single-mode SMF, the MSA has established three principles for defining a reasonable PMD layer standard for 800G-SR8:
- Multiple transmitter technology options, such as DML, EML, and silicon photonics (SiPh);
- Full utilization of device capabilities to achieve the target link performance;
- Relax PMD layer parameters as much as possible as long as reliable link performance is met.
The power budget of the SMF-based 800G-SR8 solution is similar to that defined in IEEE 400G-SR8, with the difference being the added insertion loss of the PSM8 connector. That is, the SR8 power budget can be guaranteed based on current 400GE DSP ASICs and mature optical/electronic components. In addition, the biggest challenge of the 800G-SR8 PMD lies in finding the right transmitter optical modulation amplitude (OMA), extinction ratio (ER), transmitter eye diagram closure penalty (TDECQ), and receiver sensitivity. The MSA working group compares the BER performance of different transmitters to determine these parameters, as shown below.

(a) BER vs. OMA results of EML transmitters tested with commercial 400G DSP chips; (b) BER vs. OMA results of SiPh transmitters tested with commercial 400G DSP chips; (c) BER vs. OMA results of DML transmitters tested with commercial 400G DSP chips.
The above comparisons are all based on test results on 400G DSP chips. Schemes a and b have been widely discussed. In scheme c, the BER performance of DML shows that under these conditions the OMA sensitivity is comparable to that of EML and SiPh. Considering the lower transmit power of SiPh and the sufficiently high sensitivity of all three schemes, the minimum OMA requirement in 800G-SR8 can be appropriately relaxed. Meanwhile, considering the relatively low sensitivity requirement in SR scenarios and the power constraints of 800G modules, the MSA also recommends using low-complexity DSP modes. Another parameter is the extinction ratio (ER), which is also directly related to power consumption; generally, as long as the ER does not affect link reliability, the smaller the ER the better.
Based on the above analysis, in the 800G-SR solution, a low-cost and low-power solution based on single-mode SMF is also feasible. Next time we will look at the 4x200G solution.
The above article is a partial interpretation of the MSA 800G optical module technology white paper.

Related Reading
- 800G OSFP Optical Module Evolution Roadmap
- Ciena: 800G and Higher-Rate Coherent Pluggable Optical Modules
- Key Requirements of 800G MSA Pluggable Modules
Enjoy this article?
Subscribe to get new optical communication insights delivered to your inbox. Free, unsubscribe anytime.