During my 30-year career, multiple generations of networking interfaces, from the early days of copper 10 Mbit per second Ethernet, to state-of-the-art terabit per second optical coherent networks and nearly every speed in between have advanced. Over the same period, the industry has scaled the speed per networking interface by 100,000x, which equates to a 1.46x speed increase per year.
Throughout my career in developing cutting-edge networking technologies, I’ve had the honor of working with many talented colleagues, partners, and customers to develop and deploy transformative innovations in high-speed interconnects. Today’s trends in cloud and edge computing, the Internet of Things (IoT), compute-intensive artificial intelligence (AI), factory automation and robotics, autonomous vehicles, new video formats, 5G wireless networks, and the explosion of video conferencing during the COVID-19 pandemic will continue to drive demand for even more data and higher data transmission speeds over physical lines.
As we discussed in our recent blog, The Rise of Internet 3.0, I believe that bandwidth demand will continue to accelerate after the COVID-19 pandemic abates. Instead of declining, network traffic and our collective appetites for more data will only increase as people everywhere continue to work, play, interact with others and attend events virtually online. The pandemic has set in motion a cultural transformation that coincides with imperatives to leverage technology to save time, cost and energy throughout all aspects of our lives. Sending digital bits over fiber, for example, consumes much less energy (especially fossil fuels) than traveling across town or the country to attend a meeting. This paradigm shift, however, suggests we need more bandwidth and faster pipes.
As we continue to drive the next generation of interconnect technology, it’s unlikely the networking industry will keep progressing at the same rate of 100,000x per interface over the next 30 years. As we approach the physical limits of interconnect speeds, I doubt any future semiconductor engineers, network developers, and scientists will reach 100 exabyte per wavelength transmission. Luckily, we are far away from needing an exabyte per interface. What customers want most is more cost-effective, easier-to-deploy 400G, 800G, and 1.6T network interfaces for physical lines over the next five years.
Customers tell me that the cost per bit for high-speed interconnects must continue to drop every year. This is a pragmatic technology and business need that interconnect solution providers must address. I often hear from my customers that they require a doubling of bandwidth at the same power level every three years because networking transport systems are power limited while bandwidth demands continue to soar. As a result, power per bit must shrink by more than fifty (50) percent and increase front plane capacity with each new technology generation.
Customers also require greater flexibility so that networking system vendors can design line cards with the freedom to choose the proper network interface for each application. Another pressing customer concern is the need for vendor diversity as well as the need to drive standardization so that end customers have choices when optimizing costs for their applications.
Low-power, high-performance coherent pluggable optical modules can address these customer requirements, and market adoption of coherent optics is well underway. With the recent innovations in 7 nm digital signal processors (DSPs) and compact integrated electro-optical components, powerful, small-form-factor coherent interfaces can now be developed to deliver 400 Gbit/s (with a roadmap to 5nm and 3nm) over physical lines. The Optical Internetworking Forum (OIF) was the first organization to standardize 400G ZR coherent pluggable modules for data center interconnect (DCI) links. With the OIF standardization in place and multiple vendors offering 400G ZR solutions, cloud customers are investing substantial effort in qualifying 400G ZR modules. Undoubtedly, 400G ZR deployments will be successful in cloud networks in 2021 and beyond. As an early pioneer in ZR optics, Inphi stands fully behind the ZR deployment trend.
400G ZR technology will cover the majority of DCI link requirements. Small form factor, pluggable optical modules can drive network system costs down significantly since standard-based solutions facilitate vendor competition, and small form factor implementation allows line cards to maximize capacity, making the system more cost effective. To help lead the industry toward this application paradigm, in December 2019, Inphi introduced COLORZ® II, the industry’s first 400G ZR QSFP-DD pluggable coherent transceiver for cloud DCIs, enabling large cloud operators to connect metro data centers at a fraction of the cost of traditional coherent transport systems.
Moving beyond DCI links, DSP technology innovations have sparked a revolutionary array of performance boosts that have not been possible before now. Low-power DSPs with higher coding gain forward error correction (FEC) and probabilistic shaping at incremental power dissipation over the 400G ZR mode can address the majority of metro and long haul requirements for 100-400G speeds. Inclusion of probabilistic shaping technology in coherent module form factors is especially attractive in that it can maximize data rates at longer fiber distances and deliver better link transmission robustness and margin.
Multiple vendors have successfully developed CFP2 digital coherent optics (DCOs) with 7 nm DSPs for multi-rate 100-400G connectivity for metro and long haul applications. This class of pluggable CFP2 DCOs consume less than 30% power dissipation compared to the equivalent performance of coherent custom line card versions from prior generations. In addition, to reduce power, these CFP2 DCOs are pluggable, easy to deploy, easy to service and provide multiple source options, supporting a “pay as you grow” model.
CFP2 DCOs are now the preferred choice for the majority of high-speed networks. Of course, there are non-pluggable applications such as subsea links and super-long haul networks that require higher performance DSPs with substantially greater power dissipation. However, these applications represent a small fraction of the overall market, and CFP2 DCOs continue to be the mainstream solution.
In November 2019, Inphi introduced a low-power, high-performance DSP solution that covers the majority of the worldwide DCI, metro and long haul optical networking requirements. The industry’s first merchant 7 nm coherent DSP, the Inphi Canopus™ solution paves the way for an industrywide paradigm shift in deployment models by enabling low-power, high-density QSFP-DD, OSFP and CFP2-DCO coherent pluggable modules for cloud and telecom operators over physical lines. In developing Canopus, we challenged our engineers to create a single-chip solution that can address very low power 400G ZR applications while delivering the highest performance for demanding metro and long haul links at incremental power dissipation.
We believe industry demand for power efficiency and high performance in network interconnect devices will continue in the future. In response, we are developing next-generation DSP solutions in 5 nm process technology that will expand the applicability of low-power DSPs and extend coherent pluggable optics to even higher performance applications.
Continuous advances in DSP technology will enable pluggable coherent optics to cover the majority of optical networking application needs for 100G, 200G, 400G and 800G networks through nearly the end of this decade. This trend will make interconnect systems more cost-effective, power efficient and easier to service. Equipment makers and cloud operators will also gain greater flexibility in choosing the right modules for their application needs while using the same line card designs. In my view, the continuous adoption of pluggable modules, due to advanced capabilities, is forecasted to address more than 75% of Coherent ports shipped by 2024, show in graph below.
In our era of constant connectivity, unlimited demand for bandwidth and the rise of the Internet 3.0, the rapid pace of our technology treadmill will only increase, challenging engineers to deliver faster, cheaper and more power efficient interconnect devices. From the cloud to our homes, inside and between data centers, and from the IoT everywhere to the rise of 5G wireless networks, Inphi engineers are redefining what it means to move big data faster.