The value of optics and ASIC co-packaging for data centers could reach US$2B in 2032 if the technical challenges are solved.
Over the past 50 years, mobile technology innovations have been rolled out in waves each decade. Mobile bandwidth requirements have evolved from voice calls and texting to ultra-high–definition (UHD) video and a variety of augmented-reality/virtual-reality (AR/VR) applications, and online services have driven massive increases in data traffic. Despite the profound implications of the Covid-19 outbreak for the telecom infrastructure supply chain, consumers and business users worldwide continue to create new demand for networking and cloud services.
Social networking, business meetings, video streaming in UHD, e-commerce, and gaming applications will continue to drive growth. With the advent of new digital devices packing increased capabilities and intelligence, Yole Intelligence, part of Yole Group, has observed higher device adoption rates every year. The bandwidth for connected cars, automated logistics, and manufacturing facilities, including AR/VR, is much higher than for current UHD streaming applications.
Current form factors of pluggable optics are expected to be limited in their ability to support capacities of 1.6 Tb/s, 3.2 Tb/s, and higher in terms of the required electrical and optical densities, thermal issues, and power consumption. As a result of discrete electrical device implementation, power dissipation and thermal management are becoming limiting factors for future pluggable optics.
Co-packaged optics (CPO) is a new approach that aims to overcome these challenges by bringing the optics closer to the switch ASIC. CPO technology is considered a new deployment model for the whole ecosystem and an alternative to the well-established pluggable optics. Once it achieves commercial status, CPO may dominate in specific applications, though this doesn’t necessarily mean that pluggable optics will disappear.
Revenue generated by the CPO market reached about US$6 million in 2020 and is expected to reach US$2.2 billion in 2032, for a 65% compound annual growth rate in the 2020–2032 period.1 This growth will be driven by CPO’s substantial energy savings (>30%) and capital expenditure savings, measured in dollars/Gbps, over pluggable optics.
Advantages of silicon photonics’ integration capabilities
CPO provides optical I/O for Ethernet switches or disaggregated computing in a single packaged assembly and addresses some of the challenges created by pluggable optics, including port density, power consumption, thermal management, and bandwidth. The objective is to move the electro-optic conversion process as close as possible to the compute, switch, or ASIC die to achieve higher bandwidth and energy efficiency.
Silicon photonics (SiPh), while still considered an emerging technology platform, has been well-established in pluggable optics for intermediate-reach applications. The commercial goal of CPO is to achieve a price that undercuts the current 400G (DR or FR) pluggable optics, and a SiPh-enabled higher level of photonic-electronic integration could accomplish that objective. SiPh has been found to provide advantages in integration and reliability at a lower cost. In 2021, more than 8 million SiPh-transceivers were shipped. We expect that SiPh transceiver shipments will grow significantly in the next five years, with a 25% CAGR in unit terms.
Externally mounted versus integrated lasers
Highly integrated optics require mature, high-yield fab processes and device photonic integrated circuit (PIC) manufacturing technology. These requirements force the industry to collaborate closely on a new basis, by defining new roles. This shift will not be possible to achieve overnight. In the meantime, a number of CPO technical challenges remain to be addressed.
Tentatively, instead of the deep integration of lasers with waveguides and modulators, the use of externally mounted laser sources is preferred to address concerns about laser reliability and thermal management at the expense of introducing other complexities. The external laser sources will require higher laser power to overcome additional optical losses in the path from the laser to the modulator, additional faceplate space, higher laser packaging costs, and additional fibers and connectors.
To simplify CPO assembly architecture and reduce the number of laser modules and fibers, it is advantageous to use high-power lasers and split the power among optical chiplets. SiPh PICs employing a monolithic integration approach enable higher manufacturing yield and low-loss integration of III–V devices. SiPh-based PICs can integrate lasers sharing more channels, waveguides, modulators, detectors, multiplexers, and V-grooves for fiber attachment. This would allow the elimination of packaging steps and thus smooth the path to volume manufacturing processes.
Difficult transformation ahead for optics industry
Evidence is growing that the industry is taking this transition seriously. In 2020, the optical interconnect and switch equipment industry started intensive and broad discussions on the further development of CPO. Several strategic collaborations have been announced, and the first proofs of concept have recently appeared. On the standards front, the Optical Internetworking Forum (OIF) and the Consortium for On-Board Optics (COBO) have established internal projects to make progress on CPO, and Multi-Source Agreement (MSA) activities are poised to follow suit. And two of the big four hyperscale cloud operators, Facebook and Microsoft, are actively working to support the penetration of CPO into their network architectures.
Today’s CPO industrials have formed around the merchant silicon and switch-equipment vendors, which have acquired or partnered with innovative SiPh designers. They seek new ways to supply the data center operators with the CPO ecosystem. The vast majority of CPO-involved players were identified as U.S. companies; traditional Chinese pluggable-optics vendors have not yet introduced CPO technology into their roadmaps. While the CPO market will be shaped to enable a multivendor business model, its formation may lead to consolidation in the pluggable-optics industry, although CPO extrusion will not come from those players. Rather, CPO technology will require new collaborations and the alignment of strategic partnering to address user requirements, technological feasibility, and economic viability.
The first full-scale deployment of CPO is anticipated in 2028 in a 200-terabit switching capacity. Despite the technological advantages of CPO, it will be hard to compete with pluggable modules, which will remain preferable for a long time yet. Only a few players (Broadcom, Intel, Ranovus, and some others) will bring proprietary solutions to the market. To satisfy the market demands and convince end users of CPO viability, multivendor business models and manufacturing yields must be acceptable. The coming years should bring clarity to the outlook for CPO technical solutions, especially the commercial and supply chain models.
What’s next for CPO?
Today, the optical-modules market is well established along the vertical: component suppliers, optics suppliers, and assembly/test integrators. The multivendor model, by definition, involves many suppliers, and the interoperability of multiple, different pluggable modules at one switch box contributes to the industry’s flexibility. These are principal advantages of the pluggables approach over CPO today.
When CPO becomes mainstream, the traditional industrial optics landscape could shrink significantly. CPO technology will rely heavily on SiPh. With highly integrated optics and silicon chips, new engineering capabilities and foundries will be highly desired. This will not be acceptable for traditional mid-sized players, as only the billion-dollar club of optics suppliers can afford to switch from pluggables to CPO.
Many smaller enterprise data centers don’t adopt the latest interconnect technologies until those technologies have established a track record, so technology exchange occurs at a much slower pace. That means that even if CPO becomes a mainstream technology, pluggable modules will remain in high demand for several applications in which CPO is not technically or economically viable, such as long-haul applications or edge data centers.
At Yole Intelligence, we do not believe that pluggable technology will be phased out altogether in the next 10 years. However, the pluggable optics industry may consolidate as the CPO market evolves to enable a multivendor business model.
This article was originally published on EE Times Europe.
Martin Vallo is senior analyst for Photonics at Yole Intelligence, part of Yole Group.