Understanding the Key Technologies and Benefits behind 800G Coherent

Coherent optical technology has evolved rapidly, going from 40 Gbps (2008~2009) to 100 Gbps (2010~2011), from hard-decision forward error correction to soft-decision forward error correction (2012~2014), and from 200-Gbps flexi-rate interfaces (2015-2017) to 400 Gbps (2016-2018), and then to 600 Gbps (2018~2019). More recently coherent optical technology has evolved to 800 Gbps in the high-performance embedded segment and 400 Gbps in the form of compact pluggables. In this article I will focus on the high-performance embedded segment, describing the key enabling technologies, the key features, and the key benefits.

800G coherent enabling technologies

To understand the key enabling technologies we need to look inside a typical coherent optical engine. As shown in Figure 1, key components include a digital ASIC/DSP, analog electronics, and photonics, together with radio frequency (RF) electrical interconnects. Amplification, in the form of micro-EDFAs and/or semiconductor optical amplifiers (SOAs), and tunable optical filters (TOF) are also included in some coherent transceivers for enhanced performance.

7-nm CMOS digital ASICs/DSPs

The digital ASIC/DSP plays a key role in terms of both the performance and functionality of the transceiver. It also has a large influence on power consumption and footprint. As the digital ASIC/DSP is made with silicon, it is subject to the same CMOS process node improvement cycle, made famous by Moore’s Law, that drives the entire chip industry. For example, 7-nm CMOS, which is used for 800-Gbps embedded and 400-Gbps pluggables, gives a 30% performance improvement or a 60% power reduction, and a 70% area reduction, relative to 16-nm CMOS, which is used for the previous 600-Gbps embedded generation. This extra performance is a key enabler of higher baud rates and advanced processor-intensive features.

Indium Phosphide modulators and photonic integration

In addition to the digital ASIC/DSP, high-performance 800 Gbps also requires advanced photonics including the modulator. The modulator provides a critical function on the transmit side of the transceiver. It takes light of the required frequency from the laser and, by changing the phase and amplitude, adds the data that is being transmitted. It does this by using an electric field to change the refractive index of the material the light is passing through. Doing this at the high baud rates required for 800 Gbps is very challenging. Indium phosphide with its inherently superior modulation effect has been the modulator material of choice for all the 800-Gbps optical engines on the market.

Another key enabling technology in this area is photonic integration, integrating multiple photonic functions including lasers, modulators, and photodetectors on a single integrated circuit with resulting benefits related to cost, power consumption, footprint, performance, and reliability, as shown in Figure 2.

Ultra-high baud rates

7-nm CMOS DSPs and advanced photonics including indium phosphide modulators have enabled the baud rate to increase from 60~70 Gbaud with the previous 600G generation to 90~100 Gbaud with 800G. Increasing the baud rate enables a proportional increase in the data rate of a wavelength with minimal impact on its reach.

Higher baud rates enable the use of lower-order modulation to achieve the same data rate. Lower-order modulations benefit from greater Euclidean distance between constellation points, making them easier to distinguish in the presence of noise.

At the same time, as the spectrum of the wavelength is proportional to the baud rate, a higher-baud-rate wavelength can leverage higher power for the same power spectral density and therefore the same level of nonlinearities. Together, lower-order modulation and higher power more than offset the increased sensitivity to noise and nonlinearities of the higher baud rate itself, resulting in significant capacity-reach improvements. Furthermore, at 800 Gbps even small differences in the baud rate can have significant benefits in terms of the reach. For example, going from 96 Gbaud to 100 Gbaud can deliver an extra 30% reach.

Advanced features including probabilistic constellation shaping

The increased processing power provided by 7-nm CMOS has enabled 800G DSPs to deliver multiple advanced features. And while different vendors may differentiate based on the performance of their implementation, one common feature of 800G embedded coherent is probabilistic constellation shaping (PCS).

Unlike conventional modulation, where each constellation point has the same probability of being used, PCS uses the lower-energy/power inner constellation points more frequently and the higher-energy/power outer constellation points less frequently, as shown in Figure 3. By adjusting the probabilities, PCS provides the option of fine bits-per-symbol granularity, enabling a far smoother capacity-reach curve than conventional QAM modulation. Furthermore, for the same average wavelength power and spectral efficiency, there is greater Euclidean distance between the constellation points relative to conventional QAM, which increases tolerance to linear noise such as the amplified spontaneous emission (ASE) noise from amplifiers. Alternatively, for the same linear noise tolerance as conventional QAM, less wavelength power is required, and therefore nonlinearities are lower.

Another feature supported by some 800G engines is subcarriers, referred to as Nyquist subcarriers or frequency-division multiplexing (FDM), depending on the vendor. This technology takes a single high-baud-rate carrier and digitally divides it into multiple lower-baud-rate subcarriers. This approach significantly reduces chromatic dispersion and the noise created inside the optical engine when compensating for chromatic dispersion.

Increased wavelength capacity-reach equals lower cost, space, and power

Baud rates in the 90- to 100-Gbaud range and advanced features including PCS have enabled dramatic increases in wavelength capacity-reach. For example, 400G-generation embedded coherent could deliver 400-Gbps waves to 100~300 km while 600G-generation embedded coherent could deliver 600 Gbps to 100~200 km. And while 800G reach will vary by vendor, there are engines that can deliver 800 Gbps to 1,000+ km, 600 Gbps to 3,000+ km, and 400G to 7,500+ km.

Increased wavelength capacity-reach provides significant benefits in terms of reducing the cost per bit, power consumption (which can be as low as 0.2W/G), and footprint with 6.4 Tbps of line interface capacity in 2 or 3 rack units. Such wavelength capacity-reach also makes the optical network easier to manage, with fewer wavelengths to install, provision, and monitor.

Better spectral efficiency and more fiber capacity

Deploying and lighting new fiber is an expensive and slow endeavor. Network operators therefore naturally want to maximize the capacity of their existing fiber. One way in which 800G-generation coherent can help with this relates to improved spectral efficiency. 800-Gbps optical engines leveraging features including 64QAM modulation, PCS, spectral shaping and super-channels can deliver spectral efficiency in excess of 8 bits per symbol with up to 40 Tbps in the C-Band. Several vendors now provide L-Band variants of their 800G engines, enabling up to 80 Tbps per fiber when combing the C- and L-Band together.

Deliver 400GbE services

Compact pluggable form factors such as QSFP-DD are making 400 Gigabit Ethernet (GbE) an attractive option for routers, with network operators seeing increasing demand for 400GbE transport both as an internal service interconnecting their own routers and as a service for wholesale and enterprise customers. Delivering these services cost-effectively is key. 800G-generation coherent enables 400GbE on a single wavelength to be delivered over very long distances. Over shorter distances, 800G enables two 400GbE services to be delivered on a single wavelength, thus driving down the cost per service.

Some 800G-generation engines also support two wavelengths and bandwidth virtualization. This feature can enable three 400GbE client services to be delivered over two 600-Gbps wavelengths in scenarios where 800 Gbps cannot be supported but where 400 Gbps would leave excess margin on the table.

Bandwidth virtualization can also be useful under challenging conditions (poor-quality fiber, long spans, G.655 fiber with low chromatic dispersion, etc.) where a single 400-Gbps wavelength cannot be supported. An example is enabling a 400GbE service to be delivered over two 200-Gbps wavelengths.

High-performance coherent beyond 800 Gbps

While coherent pluggables have a strong role to play in applications including metro data center interconnect (DCI), metro edge, metro-regional network, and in the future even inside the data center, high-performance embedded optical engines will continue to play a leading role in long-haul and subsea applications where performance and in particular spectral efficiency are top priorities. The next generations of coherent optical engines will leverage advanced CMOS process nodes for more powerful DSPs, together with higher-performance analog electronics, modulators, and other photonic components. Expect both 140+ Gbaud, 1.2 Tbps per wavelength, and 200-Gbaud, 1.6-Tbps optical engines to emerge in the coming years.

Paul Momtahan is director, solutions marketing, at Infinera.