Keywords

Silicon photonics; Moore’s law; sensors; disruptive technology; optical industry; LIDAR; AIM Photonics

8.1 Looking Back to See Ahead

This final chapter of the book reflects on the likely course of silicon photonics near term and offers a perspective on longer-term possibilities.

A look back 20 years demonstrates how quickly technologies can emerge. It also reveals the perils of prediction.

Back in 1995, the first version of Microsoft’s Internet Explorer web browser was launched; data and messaging services using GSM—the second-generation cellular wireless standard—began; and you could access the Internet with a download speed of several tens of kilobits a second using a dial-up modem. Amazon was a 1-year-old company, while cloud computing, smartphones, Big Data, YouTube, and Google were all to come.

Looking forward, there is a good reason to believe that the pace of change will be even quicker in the next two decades. Unlike in 1995, a mature wireless and wireline networking infrastructure is now in place, and many things including innovation are happening faster [4].

Accordingly, silicon photonics is emerging into a technological future that will be strikingly different from what we recognize today. A faster pace of change will benefit silicon photonics and inevitably dictate its future course. Hence, it is right to be cautious, but there is value in forecasting, extrapolating current trends over differing time spans to predict silicon photonics’ impact.

8.2 The Market Opportunities for Silicon Photonics: The Present to 2026

In a 2016 report [5] LightCounting Market Research analyzed integrated optical component trends and the role of silicon photonics compared to established optical component technologies such as indium phosphide and gallium arsenide.

LightCounting’s report included several notable findings. One finding is that only 1 in 40 optical components is an integrated design. Here, optical components refer to transceivers and lasers but not components such as optical amplifiers and optical switches. This means that in 2016, less than 3% of all the optical components sold in the telecom and datacom markets were integrated—a surprisingly small fraction.

Yet while integrated components account for a small fraction of total volumes, they generate one-third of the total annual optical component market revenues. Integrated devices are valuable designs due to their relatively high average selling price.

Looking ahead, LightCounting concludes the market for silicon photonics devices will grow threefold to $1 billion in 2021. By then, 1 in 10 optical components will be integrated, and will account for 60% of global revenues. Total sales of integrated devices, including indium phosphide and gallium arsenide products, are projected to reach $5 billion by 2021.

LightCounting concludes that the optical component market impact of silicon photonics will not be significant in the near term: 2016–21. Still, a tripling of market value to $1 billion represents significant growth. Such revenue growth is welcome but will be shared among many silicon photonics companies.

The 2021 $1 billion market also does not include the use of silicon photonics by companies within their own systems, players such as Huawei, Mellanox, Cisco, Ciena and Juniper using the technology in their servers, transceivers, or switches. Nor does it include silicon photonics products from emerging start-ups such as Ayar Labs, Sicoya, and Rockley Photonics.

Such examples are not counted as direct optical component sales but these developments will generate revenues for the parent companies and will advance silicon photonics. They also represent examples of the core benefit of silicon photonics, working alongside electronics as part of system designs that give companies a technology edge.

8.2.1 Near-Term Opportunities: 2016–21

Fig. 8.1 shows the likely timescales of emerging markets for silicon photonics as well as important developments. As shown, telecom and datacom are markets where silicon photonics is already playing a role, and this is the obvious main opportunity for the next 5 years.

These opportunities include 100-Gb transceivers from mid-reach links in the data center and 100-Gb and faster modules for coherent optical transport. Mid-reach optics spans 0.5–2 km and is served currently by such module designs as PSM4 and the CWDM4 and CLR4. There are also 100-, 200-, and 400-Gb coherent applications using the CFP and CFP2-ACO designs.

It is also likely that multiwavelength terabit-plus coherent photonic integrated circuits will appear, as indicated by the emerging CFP8-ACO pluggable module [6].

Also shown are Microsoft’s Madison module requirements. Microsoft has data centers made up of several buildings distributed on a campus that are separated by distances of 2 km. It also has buildings making up a data center spread as far as 70 km apart. The Madison modules are the optical components industry’s answer to Microsoft’s demand for optics outside standard multisource agreement initiatives—an example of an Internet content provider driving new optical developments.

The first Madison QSFP28 module will support 100 Gb using just two wavelengths, each 25 Gb, coupled with four-level pulse-amplitude modulation (PAM4) signaling. Microsoft is working with semiconductor supplier Inphi to develop the module. Microsoft is also talking to interested optical module makers to develop a follow-on, known as Madison 1.5, to create a 100-Gb QSFP28 design using one wavelength only based on 50-GBd signaling and PAM4. The specification is to achieve a total bandwidth of between 6.4 and 7.2 Tb down a fiber.

Lastly, Microsoft is also interested in a tailored coherent optics solution that will allow up to 38 Tb transmission. Madison 2.0 is designated to be implemented using the Consortium of On-Board Optics (COBO) form factor (see Sections 5.5.1 and 7.5.4).

One multisource agreement, the OpenOptics wavelength division multiplexing design cofounded by Ranovus and Mellanox, is being aimed at higher-than-100-Gb links within the data center [7]. New higher-capacity pluggable form factors will also be launched during this period, such as the QSFP-DD, the µQSFP, and the QSFP56.

Data center networking will also drive new Ethernet speed standards such as 25-, 50-, 200-, and 400-Gb Ethernet. The simplest way to implement 50-Gb signaling will be to use two 25-Gb lanes but that will quickly be replaced with more elegant 50-Gb single lanes. A 50-Gb lane—25-GBd optics and PAM4 modulation—will enable these new-speed Ethernet configurations. And the combination of 50-GBd optics and PAM4 will enable 100-Gb single lanes. Mellanox is one company that will use 50-Gb nonreturn-to-zero signaling and is confident that the approach can be extended to 100-Gb using its silicon photonics technology. But at 100-Gb, copackaged optics will be needed.

Midboard optics designs using COBO are also to be expected, with first hardware supporting 400 and 800 Gb/s projected toward the end of 2017. Midboard optics will become a more pressing need as switch chip companies announce 6.4-, 12.8-, and 25.6-Tb switch silicon.

New system architectures using silicon photonics will also be deployed in the near term, such as Rockley’s switch architecture for the data center, and chip-to-chip optical designs. Other innovations will be developed by start-ups that are still in stealth mode, and silicon photonics will also be adopted by systems vendors in their own equipment, developments that are not always announced.

In turn, it is possible that high-capacity sliceable transponders will emerge for long-distance optical transmission toward the end of the near term.

Thus the number and diversity of silicon photonics products will grow as we move into the 2020s.

While this book has focused on emerging silicon photonic opportunities for datacom and telecom, spectroscopic products will also benefit from photonic integration on a CMOS platform. The new Fourier transform infrared spectrometer being developed by Lumux Technology is one example [8].

The sensor is a potential high-volume, low-cost product for applications like environmental monitoring in the oil and gas industries. The company’s goal is a consumer spectrometer that works with personal mobile devices.

8.3 The Great Cultural Divide

All the chapters up till now have focused on technology. How technology takes hold in the marketplace, semiconductor technologies, photonics technologies, and technologies used in systems. Such a tech-centric focus is necessary to assess the significance of silicon photonics and identify where it will play a role.

Yet the biggest challenge facing silicon photonics does not concern the technology. That is not to downplay the technical and manufacturing obstacles that remain. It is just that with investment and time, such challenges are being overcome. Silicon photonics is gaining momentum and the challenges being addressed are engineering ones, not issues awaiting a fundamental technological breakthrough.

Silicon photonics’ biggest challenge is, in fact, people. Or, more accurately, the cultural divide between the chip and the photonics communities.

The optical component industry has a tradition of making relatively small numbers of custom devices. And when the industry has embraced standardization such as pluggable optical components, competition has been fierce, and the ability for any one company to make differentiated products has been limited. The advent of the data center market has brought shorter product cycles and the promise of volumes but for now the optical component industry is not prepared for that.

Silicon photonics brings high-volume manufacturing and the promise of integration. But the optical components industry has limited opportunity for system integration and significant volumes for such products do not exist.

In contrast, the chip industry is a master of high-volume manufacturing, it has widely embraced and benefited from standardization, and unlike the optical component industry, it can use billions of transistors to produce differentiated products. The chip vendors also have more scope to develop systems, albeit systems at the chip level.

But chip engineers are skilled in electronics design and have little knowledge of photonics and are unlikely to start changing how they design chips. This explains why optics in the data center is largely confined to pluggable optics on the edge of systems and to cabling, even though the Internet content providers want optics brought inside systems.

In August 2016 Juniper Networks announced its intention to acquire US silicon photonics start-up Aurrion [11]. Juniper joins several systems vendor that have made acquisitions to bring silicon photonics technology in-house. Such moves by the equipment makers are an acknowledgment that as photonics moves closer to the silicon and away from a system’s faceplate, silicon photonics is becoming strategically important.

So will the telecom and datacom system vendors, that design their own chips and now have in-house photonics expertise, be the ones to perform a mashup and bridge this cultural divide? The answer is most likely yes, or at least they will be trailblazers, but only for their benefit.

A consequence of these acquisitions is that the silicon photonics technology becomes the property of the systems houses and is lost to the market at large. For the wider community, Aurrion represents yet another silicon photonics start-up whose technology has been removed for use by the wider marketplace.

This is what AIM Photonics—included in Fig. 8.1—is looking to address. AIM Photonics is a US public–private partnership that is developing technology for integrated photonics. In particular, AIM is looking to advance the manufacturing of silicon photonics, making the technology available to small-to-medium-sized businesses and entrepreneurial companies.

Silicon photonics luminary, Professor Lionel Kimerling, sees AIM Photonics as bringing the manufacturing discipline of the electronics industry to photonics. AIM will also allow chip designers to start exploring the design benefits of silicon photonics. AIM, e.g., will be offering a system-in-package service by the end of 2018. AIM Photonics is a 5-year project to be completed in 2020 by when it hopes to be self-sustaining [12].

This explains why Professor Kimerling is spending much of his time putting together educational material to help attract individuals to pursue a career in silicon photonics. Much of the technology is in place, he says, what is required is to make it accessible to people. “Once we get it accessible and people start to experiment and design, good things are going to happen,” says Kimerling.

AIM Photonics has the potential to help bring new silicon photonics players and their products to market but it alone will not bridge the cultural divide. What will is when electronics and photonics replaces Moore’s law as the basis for system scaling. The data center and its requirements are already showing that this is starting to happen.

In Chapter 1, Silicon Photonics: Disruptive and Ready for Prime Time, we asked if silicon photonics is disruptive. Any emerging technology that promises to continue scaling post-Moore’s law suggests it is disruptive. This new scaling force will also cause silicon photonics to cross over to the chip industry. Silicon photonics driven by the chip industry with also disrupt the existing optical industry, as is now explained.

8.4 The Chip Industry Will Own Photonics

The need to continually scale systems will mean that the chip industry will sooner or later embrace silicon photonics. Silicon photonics will then move from a technology being developed by the optical industry to one being driven by the much larger chip industry. Or as silicon photonics luminary and Aurrion cofounder Professor John Bowers puts it, photonics will transfer to silicon [2].

This will significantly change the optical industry which, today, has a hierarchical structure (see Fig. 8.2, left). It is not clear whether silicon photonics design houses will emerge, as ASIC design houses did for companies that wanted custom chips. But ultimately the foundries and large chip companies with their own fabrication plants will come to own silicon photonics.

Already there are chip foundries making silicon photonics chips, while leading chip companies such as Intel and STMicroelectronics are actively developing silicon photonics chips in their fabs. Both Intel and STMicroelectronics offer 100-Gb PSM4 module designs. But their motivation to embracing silicon photonics was not to become optical transceiver players. Meanwhile, the systems vendors with their silicon photonics acquisitions are already moving toward this new ecosystem.

Silicon photonics will become the optical integration platform, just as CMOS became the platform for the bulk of the chip industry. And similarly, while not all electronic chips are designed in CMOS, silicon photonics will dominate volumes but indium phosphide, gallium arsenide, lithium niobate, and planar lightwave circuits will continue to be made. But they will become increasingly specialist; silicon photonics will become the platform for optics.

By 2035, people will not think in terms of CMOS and silicon photonics. Design tools will guide developers as to what is best done electrically and what is best done optically. Such tools will span systems: how chips will be copackaged, how system-in-package devices will be integrated on a board and how such boards will be combined to make systems. And of course system design will cover new applications. In-body sensor devices will have very different systems requirements to today’s telecom and datacom platforms.

Specialist functions and materials will also emerge in the next 20 years. There will be other forms of computation such as optical alongside digital, and more materials such as graphene will be added to the mix. These system elements will need to be stitched together and communicate to create more complex and varied designs. So while photonics and electronics may be coming together, the form factors and the technologies will likely become more varied.

It may be premature to talk about silicon photonics’ legacy when the industry is still openly questioning its merits and significance. But our expectation is that, as its name implies, silicon photonics will be viewed as the technology that bridged two distinct industries: photonics and semiconductors.

Indeed, silicon photonics will come to be seen as nothing less than a strategic inflection point of the Information Age.