The Route to Market for Silicon Photonics
Abstract
After excessive early enthusiasm for silicon photonics, the technology is now on a solid course with companies bringing products to market. There are already several product success stories, but further development is needed for silicon photonics to become established in the marketplace and that requires revenues. One emerging revenue-making opportunity is 100 Gb/s optical interconnect for the data center. However, this market opportunity alone will be insufficient to launch silicon photonics into the mainstream; very few suppliers will benefit significantly from the hundreds of millions of dollars in revenues that the 100-Gb interconnect market will generate. Further opportunities will follow, including high-speed coherent optical transmission, 400-Gb interfaces, and the trend of optics moving closer to silicon. Even without a single killer application, a strong silicon photonics ecosystem is taking shape, while the continuing acquisition by equipment makers of silicon photonics companies is ensuring another path for the technology’s long-term success.
Keywords
100-Gb optical transceivers; 400-Gb Ethernet; coherent transmission; technology development cycle; variable optical attenuator; start-ups; silicon photonics ecosystem
A technology does something. It executes a purpose.
W. Brian Arthur [1]
Once you can do something in silicon and do it adequately well, it tends to displace everything else from the majority of the market. We have seen that over and over again, and I don’t think it will be any different in the optics world.
Michael Hochberg [2]
4.1 The Technology Adoption Curve
How does a new technology arise and become adopted in the marketplace?
In his book on the nature of technology, W. Brian Arthur talks about how new technology builds on existing ones, identifying a process he calls internal replacement [1].
Once a technology is adopted, its performance is pushed to deliver more. If it is not operating close to its limit, it is, by definition, being used inefficiently. When competition is severe—as it is for the optical component and equipment vendors addressing the datacom and telecom markets—even a small edge can pay off handsomely, says Arthur.
But you can only push a technology so far before part of the system hits a barrier. One way to overcome this limitation is to replace the impeding component—a subtechnology—with a better one. This improved component will likely require adjustments in other parts of the system to accommodate it. Arthur cites how moving from a wooden to a metal aircraft frame in the 1920s and 1930s led to a rethink of aircraft design.
Internal replacement improves a system’s parts and subparts and impacts all levels of its hierarchical design.
Similarly, if silicon photonics is to be adopted widely, it must internally replace existing component technologies, delivering an edge that established technologies cannot match. Its impact could be as a component, or for more advanced parts of a system. Silicon photonics could even lead to new, novel systems given its compactness, interconnect reach, and integration potential.
In turn, technologies that are adopted in the marketplace go through a recognizable development cycle, and silicon photonics is proving no different. Market research and advisory firm Gartner has identified a four-stage technology development cycle, shown in Fig. 4.1.
The four-stage cycle starts with an innovation: a development or an advancement that builds on existing knowledge to create the new technology [3]. Research embraces and builds on this innovation, sparking enthusiasm and a market expectation once the technology’s potential is recognized. At this next stage, the market explores potential applications, and start-ups may enter the market with an innovative product using the technology. The enthusiasm feeds on itself and soon expectations become inflated. But eventually this crescendo of interest reaches a peak and is followed, inevitably, by deflation.
Several reasons account for this disillusion phase. Developing a technology sufficiently to enter the marketplace is challenging; setbacks and delays are inevitable. Equally, the marketplace continues to evolve such that the technology’s initial promise may lose some luster. But like the peak of expectation, this disillusionment period also passes; successful technologies are adopted after all.
What then follows is the long hard slog to bring the technology to its productivity phase, and this is the current status of silicon photonics.
This chapter details the key events that have led silicon photonics to reach the productivity phase. Four notable early silicon photonics product case studies are detailed. The chapter also addresses the challenges facing silicon photonics and the current state of the industry.
4.2 A Brief History of Silicon Photonics
Silicon photonics may first have been thought about in the mid-1980s, but the rapid strides made in its development have occurred since 2000. Fig. 4.2 highlights key silicon photonics developments during this period.
In the following sections we discuss the events in Fig. 4.2, from earliest to latest.
NTT [4] and Cornell University [5] developed an inverted taper fiber coupler design in 2002 and 2003, respectively. The taper is used to match light between the fiber core and the much smaller silicon photonics waveguide, as discussed in Chapter 3, The Long March to a Silicon Photonics Union.
Intel announced its first silicon photonics breakthrough, a 1-GHz modulator, in 2004. The world’s leading chipmaker decided that silicon photonics would be a core technology, and the company’s R&D engineers, over several years, pushed laser, modulator, and photodetector development for the technology.
Mario Paniccia, who headed Intel’s silicon photonics efforts at the time, spotlights a particularly creative period between 2002 and 2008. During that time, his Intel team had six silicon photonics papers published in science journals Nature and Nature Photonics, and it held several world records. These were for the fastest modulator, first at 1 Gb/s, then 10 and finally at 40 Gb/s; and the first pulsed and continuous-wave Raman silicon laser. Additional records followed: the first hybrid silicon laser based on work with the University of California, Santa Barbara, and the fastest silicon germanium photodetector operating at 40 Gb/s. These piece parts were required before Intel moved on to start designing optical transceivers [6]. These achievements were all in one place, in labs within 100 yards of each other, says Paniccia: “You had to pinch yourself sometimes.” [7]
Intel was not alone. IT giants IBM, HP, and Sun Microsystems (since acquired by Oracle) used their research and development labs to explore the technology also. IBM developed a 10-Gb/s Mach–Zehnder silicon photonics modulator in 2007 and the first 40-Gb/s germanium receiver in 2010. But Intel was the main cheerleader and did most to highlight the technology’s potential, heightening industry expectation toward its peak.
The innovation era also spawned three significant silicon photonics start-ups: Kotura, Lightwire, and Luxtera. Kotura was first to market with its silicon photonics–based variable optical attenuator product in 2005. Lightwire launched a 10-Gb Ethernet optical module in 2008, and Luxtera followed in 2008 with a 40-Gb/s active optical cable (AOC) for linking equipment within the data center.
Two of these three early start-ups were also notable for their high market valuations, given that there were only a few silicon photonics products and little revenue at the time. Cisco Systems paid $271 million for Lightwire in 2012, making it the most expensive silicon photonics start-up acquisition to date [8]. And in 2013 datacom equipment maker Mellanox Technology acquired Kotura for $82 million [9]. Meanwhile, Luxtera sold its AOC business to Molex in 2011 [10].
In hindsight, the first wave of start-up acquisitions marked the peak of enthusiasm for silicon photonics on the technology adoption curve. Innovation and industry developments have continued, but the spate of acquisitions, and the sums paid, put the spotlight firmly on silicon photonics during this period.
Financial analysts also alighted on the technology’s potential, warning leading optical component companies such as Finisar that not having a silicon photonics strategy would harm their businesses [11].
Leading Chinese equipment maker Huawei bought a small Belgium silicon photonics start-up Caliopa for an undisclosed fee in 2013 [12], although reports suggested it was $5 million, while in 2014 Macom acquired BinOptics for $230 million [13]. The amount paid was not solely for the company’s silicon photonics since BinOptics also makes lasers, a valuable technology asset. Macom also bought Photonic Controls, a silicon photonics design company [14].
Other notable product developments included Intel announcing its “disaggregated” server with silicon photonics earmarked to play an integral role in its design [15]. Recall that the server is the computing platform used in huge volumes in large-scale data centers, while disaggregation refers to separating the processing, memory, and storage functions as part of an industry reassessment as to how best to architect servers.
Cisco also launched its custom CPAK 100-Gb/s transceiver, its first silicon photonics product based on Lightwire’s technology [16].
Start-up Acacia Communications began selling a silicon photonics–based 100-Gb/s, low-power, pluggable coherent transceiver for metro and long-haul applications [17]. A year later, Acacia announced a follow-on coherent module that supports data rates of 200, 300, and 400 Gb/s [18].
The burst of acquisitions of 2012 and 2013 was followed by a quieter phase, dampening industry expectations. Adding to the sense of lost momentum was the announcement in early 2015 by Intel—which has done most to trumpet the technology—that its silicon photonics–based product plans would slip a year [19].
IBM, like Intel, has been developing silicon photonics technology for over a decade. It announced in 2015 a nearly complete 100-Gb/s transceiver using wavelength-division multiplexing technology. However, while the company said it would use the transceiver for its computing systems, no schedule was given. IBM added that its design would need wider adoption to become cost-competitive with existing optical component technologies [20].
Accordingly, 2014 through the first half of 2015 was silicon photonics’ era of disillusionment. Now, the industry has exited its quiet period. Ciena bought Teraxion’s silicon photonics team for $32 million in early 2016 [21], while Juniper Networks acquired start-up Aurrion for $165 million in August [22]. In between, Acacia Communications achieved a successful initial public offering, raising $105 million for the company [23].
More and more companies are adopting the technology and announcing first products. For inside the data center, Macom announced a 100-Gb/s CWDM4 (coarse wavelength-division multiplexing 4) module [24] as has the firm Kaiam [25], while Lumentum detailed its PSM4 (parallel single mode 4) module [26]. All three designs use silicon photonics in a pluggable QSFP28 form-factor module. Start-up Ranovus detailed a 200-Gb/s CFP2 direct detection optical module with a reach of up to 130 km for Layer 4 data center interconnect applications [27]. Intel too announced its first silicon photonics optical modules for PSM4 and CWDM4/CLR4 [28].
All these companies are addressing the challenges that need to be overcome to make silicon photonics a cost-competitive alternative to the established optical technologies. Companies and start-ups will fail even as the overall silicon photonics proposition continues to take root, but that is the case with any technology transition.
4.3 Four Commercial Silicon Photonics Product Case Studies
Several silicon photonics products are being sold in volume, including one sold since 2005. Studying these products helps identify silicon photonics’ merits as well as the challenges companies face when bringing the technology to market.
4.3.1 Kotura’s Variable Optical Attenuator
Kotura’s variable optical attenuator was the first silicon photonics–based product to be shipped in volume. A variable optical attenuator is used to trim a fiber’s optical signal power levels. Applications include leveling the power exiting an optical amplifier across a fiber’s spectrum, and protecting a photodetector at a receiver from being overwhelmed by too strong a signal (Fig. 4.3).
Kotura introduced its variable optical attenuator in 2005. One decade later, 10000 units were being shipped each week. Kotura was acquired by Mellanox Technologies in 2013.
The product’s success stems from its responsiveness. Its signal leveling and protection occur in under a microsecond, far faster than equivalent variable optical attenuators based on other technologies. The product is thus ideal to control transient—spurious—light. Also Mellanox’s variable optical attenuator is highly reliable as there are no moving parts, unlike other approaches. And given the volumes made, its low cost and price make it a formidable competitor.
This is silicon photonics’ first example of a product trumping the competition by having superior optical performance. For Kotura, it was also a shrewd move. The company made a simple yet performance-leading product for telecom while gaining valuable production experience as well as revenues, a winning combination for a start-up.
4.3.2 Luxtera’s 40-Gb Active Optical Cable (AOC)
Luxtera launched its 40-Gb/s AOC using silicon photonic–based transceivers in 2008. AOCs are used to connect equipment over short reaches: from 10 m to a few hundred meters, although the bulk of links are 30 m or less.
Standard form-factors are used at each end of an AOC slot into the equipment’s faceplate, but the transceiver optics are embedded in housing that is part of the cable. Since vendors control each end of the link, they can use proprietary—nonstandardized—optical designs.
Luxtera’s 40-Gb product used a single-mode fiber ribbon cable: four fibers to transmit and four to receive. For the transmitter, Luxtera cleverly used a single laser shared across the four channels, saving costs and reducing power consumption. Each of the four channels has its own silicon photonics modulator. The resulting design, claimed Luxtera, offered the lowest-power option for data center interconnects, consuming 20 mW/Gb, 30% less than competing approaches [29].
Luxtera spent several years exploring applications where it could use silicon photonics to produce superior products. “We ship product, and to ship you need to have product differentiation against a pretty competitive landscape,” explains Luxtera executive Brian Welch [30].
The start-up studied short-reach links but was put off by the strong competition from vertical-cavity surface-emitting lasers that operate at 850 nm. Luxtera also examined wavelength-division multiplexing but concluded that while silicon photonics faced no technical hurdles in addressing this application, it offered no advantage compared to indium phosphide–based products.
This led Luxtera to alight on a parallel single-mode fiber design for the data center—in effect a 40-Gb PSM4 design before the PSM4 multisource agreement was conceived (see Appendix 1: Optical Communications Primer).
It is important to stress that the AOC’s optical performance is not specific to silicon photonics; a similar design could be made using indium phosphide. But Luxtera used its silicon photonics expertise and achieved first-mover advantage.
Luxtera has since developed a 56-Gb/s four-channel PSM4 cable, having discontinued its 40-Gb AOC. Luxtera sold its AOC business to Molex in 2011 [10].
The company also makes a 100-Gb/s (4 lanes, each at 25 Gb/s) PSM4 transceiver as well as selling the internal chips, dubbed optical engines (Fig. 4.4). In September 2016 it announced that it had shipped its one millionth PSM4 transceiver—the sum of all its 40- and 100-Gb/s shipments, including AOCs and standalone optical transceivers [31].
4.3.3 Cisco Systems’ CPAK
Cisco Systems acquired Lightwire, the silicon photonics start-up, in 2013 after realizing its next-generation switching and routing equipment would be delayed waiting for the industry to start selling 100-Gb pluggable modules in the CFP2 form factor.
Having a 100-Gb CFP2-sized module was important for Cisco to provide the necessary input–output traffic to feed its 400-Gb network processor. A network processor is a specialized chip that processes and classifies Internet Protocol packets at line rates up to 400 Gb/s. Cisco designs its own network processor ASICs (application-specific integrated circuits) to gain a competitive edge. We will meet network processors again in Chapter 7, Data Center Architectures and Opportunities for Silicon Photonics.
Because the industry was late with the 100-Gb CFP2 optical module specification, Cisco decided to buy Lightwire to make its own 100-Gb custom pluggable module, dubbed the CPAK (Fig. 4.5). This sounds like a risky strategy because making its own module would also take time, but Cisco had already been working closely with Lightwire before the acquisition.
“By the time someone could provide us with the CFP2 optics that could achieve the input–output density, that network processor silicon investment would have gone stale,” says Russ Esmacher, a Cisco Systems executive [30].
The CPAK transceiver is not solely silicon photonics–based. The four channels, each operating at 25 Gb/s, use indium phosphide lasers and photodetectors, but the modulators and optical waveguides are based on silicon photonics.
By making the CPAK, Cisco quadrupled the packet-processing performance of its line card while keeping power consumption fixed. This is a common equipment requirement: doubling or quadrupling the performance of a card and system without increasing power consumption, an expectation set by Moore’s law and largely fulfilled.
The equipment maker upgraded its 100-Gb line card to one with a 400-Gb network processor and four CPAKs, each implementing the IEEE 100GBase-LR4 100-Gb Ethernet standard that has a 10 km reach.
Cisco gained an enabling technology by buying Lightwire, resulting in a year’s lead on its switch-vendor competition. Such platforms sell for far more than the cost of the individual modules. Cisco also gained valuable experience with an optical technology that it now has in-house, a photonic complement to its ASIC expertise.
4.3.4 Acacia Communications’ Coherent Transceivers
Acacia Communications is one company targeting the challenging long-distance optical transmission market.
Founded in 2009, Acacia has made the most highly integrated silicon photonics product to date: a 100-Gb coherent transceiver in a CFP module for metro applications [17]. Acacia started selling its AC-100 CFP module in late 2014 (Fig. 4.6).
The company has integrated a variety of optical functions on its silicon photonics transceiver coherent transmission chip.
These include four modulators to implement the modulation scheme for long-haul optical transport, polarization beam splitters to separate the two polarizations of light, a 90-degree optical hybrid whose function is to mix the received signal with the reference signal to implement coherent detection, variable optical attenuators—a component discussed in the Kotura case study—to avoid receiver saturation, and photodetectors to recover the transmission and for monitoring purposes.
In 2015 Acacia introduced its AC-400, the first dual-carrier transceiver [18]. The two-wavelength “superchannel” design supports 200-, 300-, and 400-Gb/s line rates. To achieve the higher data rates, more complex quadrature amplitude modulation schemes are used. Coherent transmission and the associated modulation schemes are discussed in Chapter 5, Metro and Long-Haul Network Growth Demands Exponential Progress.
Acacia sees CMOS-based optics as key to lowering design costs. The company highlights testing the chips while still on the silicon wafer—wafer-level testing—as one important cost-saving measure. And photonic integration reduces the number of discrete components. The laser is the only optical function not integrated on-chip.
Using integration, Acacia avoids the use of lenses, for example. The chip does not need to be hermetically sealed, a common requirement to protect the laser and other optical components from performance degradation due to moisture and dust. The silicon photonics waveguides also confine the light, resulting in a compact design, while integration reduces the touchpoints needed when making the module, reducing manufacturing costs.
Acacia’s expertise is not confined to silicon photonics. The company makes its own digital signal processing ASIC for coherent transmission (see Appendix 2: Optical Transmission Techniques for Layer 4 Networks). Combining the two—a photonic integrated circuit (PIC) and the coherent DSP-ASIC—allows Acacia to trade-off the performance of the optics with the compensating electronics to deliver a cost-, size-, and power-optimized coherent CFP optical module.
Acacia faces performance challenges, however. Silicon photonics waveguides have a relatively high insertion loss, and for coherent transmission optical loss is paramount since it dictates reach. In turn, individual components cannot be optimized independently as overall trade-offs must be made. But certain optical performance issues can be countered in the electronic domain.
The coherent transceiver is a complex, highly integrated design. However, the long-distance market uses a relatively low number of such modules: tens to hundreds of thousands of units each year, less than the number of transceivers in a single large data center. This is a drawback, as volumes benefit manufacturing while overall revenues are needed to fund next-generation product development.
4.3.5 Delivering a Performance Edge
A performance edge is a common thread for the four early silicon photonics product case studies. Yet while these are trailblazing products, their performance edge is not decisive and certainly not permanent.
The submicrosecond response time of Mellanox’s variable optical attenuator is among the fastest in the industry. Mellanox also claims that competing devices cannot match the cost of its variable optical attenuator; the performance edge has led to volumes that in turn have lowered unit costs. These factors have preserved the company’s lead.
Luxtera’s PSM4 transceiver could also be implemented using indium phosphide, as could the sharing of the laser across four channels. But Luxtera’s 40-Gb PSM4 was the first such device to market, giving it first-mover advantage. Luxtera was also first to ship a 100-Gb PSM4 transceiver, in part because of the expertise it gained at 40 Gb. But now it faces stiff competition from other optical module players.
Cisco gained a year’s competitive lead with its 100-Gb CPAK transceiver. But now the market offers 100GBase-LR4 modules using indium phosphide in a CFP2 pluggable module as well as the smaller CFP4 and QSFP28 pluggable modules.
Meanwhile, Acacia’s coherent CFP transceiver is a more complex design, both serving and competing with the system vendors. So far the company continues to show solid growth, which was further confirmed with its successful initial public offering in May 2016 [23].
Acacia’s coherent modules use an unprecedented degree of integration for a silicon photonics design, yet the design does not match the integration levels achieved using indium phosphide technology. As will be discussed in Chapter 5, Metro and Long-Haul Network Growth Demands Exponential Progress, system vendor company Infinera offers commercially available products based on a PIC implemented in indium phosphide that delivers up to 2.4 Tb of capacity [32].
Longer term, however, we argue Acacia’s solution is better suited to deliver compact low-cost modules. Silicon photonics promises higher density as more functions become integrated on silicon, and it offers copackaging opportunities with the coherent DSP-ASIC as well as other electrical parts such as the transimpedance amplifier and the optical transmitter driver circuitry.
The performances advantages of the case study products are summarized in Table 4.1.
Table 4.1
Performance Differentiators of the Four Product Case Studies
| Supplier | Product | Performance Differentiator | Comments |
| Mellanox (Kotura) | Variable optical attenuator | Fastest response time | Gained and secured a market lead |
| Luxtera | A PSM4 transceiver at 40 and 100 Gb | Low cost, low power | Only one laser, an expensive part of the bill of materials, is used rather than four |
| Cisco | The 100GBase-LR4 CPAK transceiver in a small form factor | First to market with the highest density front plate switch | The use of silicon photonics advanced system performance |
| Acacia | 100-Gb transceiver chip for coherent long-distance transmission | First to market with single-chip coherent CFP transceiver | Compact, low-power design lowers the cost of coherent modules |
In summary, a performance edge is key for silicon photonics to gain a market foothold. But for further successes, product development must continue and that requires revenues. Silicon photonics must also overcome additional issues to keep developing, as is now discussed.
4.4 What Silicon Photonics Needs to Go Mainstream
The four early product case studies detailed in Section 4.3 are important on many levels. They encourage other silicon photonics players and help build momentum. Product sales also raise revenues, vital for companies not just to balance the books but to fund future product development. Without this, silicon photonics stalls.
However, much work remains to take silicon photonics from early products to the stage where a design and manufacturing infrastructure is in place to serve a global industry.
A much-heard argument for silicon photonics is its potential to reduce component cost by piggybacking on the huge investment already made by the chip industry. There is some truth to this, but as usual things are more complex than they seem.
Silicon photonics may share the chip industry’s equipment and manufacturing processes, but it has its own processes and testing requirements.
Companies are also pursuing different designs and techniques for laser attachment, photodetectors, silicon waveguides, modulators, and other optical functions. This diversity, or lack of market convergence, is troublesome because developing manufacturing rigor, uniform tools, and the aggregation of volumes are all impeded. This runs counter to what brought success in the chip industry. That said, the chip industry also started with a diversity of approaches and solutions before settling down.
Silicon photonics is still not mature enough to direct developers to the best solutions covering all aspects of manufacturing. For example:
• New optical circuit probes and testing techniques must be developed. This is photonics after all; electronic integrated circuits have never needed such things.
• Packaging is also a key component of a photonic device’s cost, and coupling light to the silicon photonics chip presents challenges, as discussed in Chapter 3, The Long March to a Silicon Photonics Union.
• Eliminating hermetically sealed packages to protect the optical components from dust and moisture is a benefit of silicon photonics since the optical functions are buried in the chip. Nonhermetic packages are needed for all optical functions, yet some vendors still use hermetically packaged lasers, for example.
Device testing and yields are also issues. Wafer testing is being developed by many of the silicon photonics players. Correlating the device performance to the wafer measurements is a key factor in reducing cost. But such correlation comes with experience, once millions of devices have been fabricated.
Getting established foundries to back and offer silicon photonics services is also a critical issue. Freescale, acquired by NXP (itself subject to a bid from chip company, Qualcomm), and STMicroelectronics are active industry participants. Intel has its captive foundry. And Cisco is big enough to dedicate resources to silicon photonics. Yet we worry that there is not enough volume to interest the big foundries while supporting the research and development needed to enable commercialization of silicon photonics. That will likely come, but more likely only when the chip industry needs such technology.
Other elements of optical include eliminating lenses used to focus and collimate light, and the use of isolators to protect the waveguide from spurious optical reflections that affect overall operation. But there is no one uniform approach to packaging even though the industry well understands that it is one of the most important paths to lowering cost.
These processes will evolve just as they have for the semiconductor industry. Production time will decrease and product yields will go up, but it takes time. If silicon photonics is to deliver on its low-cost promise, large volumes will be needed and that requires markets.
The good news, and the subject of Section 4.5, is that 100-Gb links for the data center constitute one such high-volume market. But for now silicon photonics revenues are a fraction of the total sums invested to develop the technology, as is now discussed.
4.4.1 Hefty Investment, Little Return
Table 4.2 shows the investment in silicon photonics start-ups by venture capital firms in the last decade, while Table 4.3 highlights silicon photonics company acquisitions. Summing the venture capital funding and the total amount spent on mergers and acquisitions, the investment comes to $1081 million. And this ignores the technology development work done over the last decade by IT giants IBM, Intel, and Hewlett-Packard, among others.
Table 4.2
Venture Capital Funding for Silicon Photonics Start-Ups Based on Publicly Available Data
| Company | Investments (millions) | Comments |
| Kotura | $38.7 | 4 rounds |
| Rockley Photonics | $13 | Estimate and assuming series B |
| Lightwire | $18 | 1 round |
| Ranovus | $35 | 2 rounds |
| Skorpios Technologies | $67.9 | 5 rounds |
| Luxtera | $91.1 | 4 rounds |
| Aurrion | $22.5 | 4 round |
| Caliopa | $3.6 | 2 rounds |
| Sicoya | $3.8 | 1 round |
| Ayar Labs | $2.5 | 1 round |
| Total | $296.1 | – |
Source: Data from Crunchbase, company reports.
Table 4.3
Silicon Photonics Company Acquisitions
| Acquirer/target | Date | Amount (millions) |
| Cisco/Lightwire | March 2012 | $271 |
| Mellanox/Kotura | August 2013 | $82 |
| Huawei/Caliopa | September 2013 | $5a |
| Macom/BinOptics | December 2014 | $230 |
| Ciena/Teraxion | January 2016 | $32 |
| Juniper/Aurrion | August 2016 | $165 |
| Total | – | $785 |
aEstimate.
Source: Data from company reports.
Meanwhile, the total telecom and datacom markets that silicon photonics can address was worth almost $1 billion in 2015. The investment appears large compared to the optical component market size.
Silicon photonics as a new technology must take market share from established photonic technologies in a fiercely competitive marketplace. We estimate that silicon photonics has captured less than 20% market share in two of the most promising early markets for the technology: the variable optical attenuator market and the market for 40- and 100-Gb/s optical interfaces.
Market research firm ElectroniCast estimates the variable optical attenuator market was worth $200 million in 2015 [33]. We estimate Mellanox’s variable optical attenuator, with its superior optical performance, has captured half the market, raising $100 million in revenues.
Ovum, another market research firm, valued the 40- and 100-Gb single-mode transceiver market at $800 million in 2015 [34]. These among the most competitive optical component and module markets, making it hugely challenging for an entrant to gain share without a compelling competitive advantage.
Moreover, only recently have silicon photonics 100-Gb/s commercial products become available. We estimate that the silicon photonics market share was 10%, or $80 million, in 2015. Summing the two markets, total silicon photonics revenue in 2015 was $180 million (Fig. 4.7).
For 2016, market research firm LightCounting forecasts that transceiver sales using silicon photonics will reach $300 million. It shows the market is growing, but recall that $1081 million has been spent developing the technology. And that ignores the decade plus of R&D funding by the large IT players.
4.5 100-Gb Market Revenues Are Insufficient for Silicon Photonics
Martin Schell, professor for Optoelectronic Integration at Technical University Berlin, and director of the Fraunhofer Heinrich Hertz Institute, in a 2013 presentation at the European Conference on Optical Communications (ECOC), estimated that $20 million is needed to develop a silicon photonics device using the mature 130-nm CMOS process [35]. This requires a vendor to generate $200 million in sales over a 5-year period to fund its next-generation product development.
The figure is arrived at as follows. The typical operating margin—effectively net profit—of an optical component module vendor is 10%. Given that a product’s lifecycle is 5 years commonly, a company needs $200 million in sales to raise the $20 million quoted by Schell. Silicon photonics companies must strive for such self-sufficiency if they are to get beyond this phase of steep investment with little return.
Market research firm Ovum forecasts aggregate revenues of nearly $6.0 billion in 2016–20 for 100-Gb single-mode transceivers with up to 10 km reach in the data center. A second market, the 100-Gb coherent transceivers used in Layer 4, is valued at $3.5 billion through to 2020.
Any silicon photonics vendor that gains a 20% market share approaches self-sufficiency in either of these markets. But only one or two companies in each market can expect to achieve that.
4.5.1 The Near-Term Data Center Opportunity
The silicon photonics data center opportunity is both significant and taxing.
When the large Internet content providers turn up a new data center, as many as 100,000 optical transceivers may be used. The market for 100-Gb/s interfaces in the data center started to ramp in the second half of 2016, and because this opportunity is well recognized by the optical component and module players, competition is intense.
Luxtera, Molex, Cisco, Kaiam, Lumentum, Intel, and Mellanox are all companies that have 100-Gb silicon photonics transceiver products, while IBM and Skorpios are developing silicon photonics transceivers. All these players also compete with transceiver market leaders Finisar, Avago, Fujitsu Optical Components, Sumitomo, Oclaro, and Lumentum, which have indium phosphide–based products. And firms like InnoLight, Applied Optoelectronics Inc., Colorchip, and Source Photonics are other players attacking the 100-Gb opportunity.
There are also multiple 100-Gb interconnect interface types, which fragments market revenue. The main 100-Gb interfaces are the PSM4, 100GBase-LR4, 100GBase-LR4-Lite, CWDM4 and its more stringent specification counterpart, the CLR4, and OpenOptics [36].
Revenues for transceivers inside the data center are attractive. However, no one solution addresses all requirements due to different reaches, multisource agreements, and standards. This means multiple products need to be developed, increasing the overall investment companies must make. And with many suppliers, some with deep pockets and others with formidable technology, the competition limits the market share any one company can expect.
Fig. 4.8 illustrates the challenges. It shows estimated revenues by technology for single-mode 100-Gb transceivers up to 10 km. As the incumbent technology for 100GBase-LR4, the de facto standard at this reach, indium phosphide is best positioned to win the lion’s share of the revenue. The first single-mode 100-Gb transceiver commercialized for reaches to 10 km is based on indium phosphide, and while it is not expected to be the highest volume product, its tight specifications drive higher prices, resulting in the biggest revenue among its interconnect peers.
The PSM4 and OpenOptics are expected to be supported by silicon photonics, but the CWDM4 and CLR4 will use either silicon photonics or indium phosphide technologies.
A market leader usually emerges when it becomes the first to introduce a new product, gaining over half the market share. Finisar’s 100GBase-LR4 transceiver in the CFP form factor is one such example. A company needs to develop products that win market share, yet choosing which 100-Gb products to back is not straightforward.
What does all this mean?
The 100-Gb data center opportunity is timely for silicon photonics, but this market is insufficient to deliver self-sustaining revenues for transceiver suppliers. Venture capital, public markets, and even government resources will need to be tapped to help take suppliers to the next level. There is also another development: system vendors are acquiring silicon photonics players, as we discuss in Section 4.6.
4.5.2 The Coherent Transceiver Market Has Its Own Challenges
Acacia is the only vendor currently shipping a silicon photonics–based coherent transceiver. Fujitsu Optical Components and Oclaro are shipping competing products that deliver 100- and 200-Gb coherent links in a CFP2 Analog Coherent Optics module, and they are being joined by Finisar and Lumentum.
But the competitive landscape here is different to that of the 100-Gb data center market because of the digital signal processor ASIC and the electronics needed to sample and convert the analog signal to a digital one. These are critical technologies for product differentiation.
Optical equipment vendors such as Huawei, Infinera, Ciena, Nokia, and Cisco all have proprietary digital signal processor ASICs owing to the device’s importance in determining overall optical link performance. The DSP-ASICs are becoming highly sophisticated in the features they enable and give companies an important time-to-market advantage [37,38]. Meanwhile, merchant DSP-ASICs are available from suppliers such as ClariPhy (set to be acquired by Inphi) and NEL, but the optical equipment vendors’ designs are ahead of the game.
Acacia’s ambitious design highlights the merits of silicon photonics. Its AC-400 transceiver supports 200-, 300-, and 400-Gb rates, making it the market’s first flexible-rate compact module that also has low power consumption. Silicon photonics is important in realizing this performance. But other companies are developing silicon photonics products to compete with Acacia. NTT, e.g., has detailed development results for such a design.
4.5.3 Emerging Opportunities for Silicon Photonics
The focus of this chapter has been the 100-Gb/s opportunity, but higher data rates are emerging in the data center, such as 400-Gb Ethernet. Four formats are being standardized by the IEEE for 400-Gb Ethernet which uses more sophisticated modulation than standard 100-Gb technologies. The four styles of 400-Gb Ethernet also require more lanes to achieve the higher data rate [39]. Increasing the number of lanes favors photonic integration and hence silicon photonics. Rather than coming relatively late as happened at 100 Gb, silicon photonics will be competing with indium phosphide technology as the 400-Gb standard emerges. Silicon photonics will therefore be an even more formidable competitor at this rate.
There are also new developments for optical transport suited to silicon photonics: higher-speed coherent optics, and new direct detection technologies for shorter-reach (up to 130 km) data center interconnect. These are discussed in Chapter 5, Metro and Long-Haul Network Growth Demands Exponential Progress.
Another trend is to move optics away from the faceplate and closer to the silicon—first on-board optics and then copackaged optics; developments explored in Chapter 7, Data Center Architectures and Opportunities for Silicon Photonics. These applications are the most significant opportunities for silicon photonics in the datacom and telecom domains.
4.6 The Silicon Photonics Ecosystem: A State-of-the-Industry Report
This chapter has highlighted two key findings:
• The successful silicon photonics products to date have all offered a particular performance advantage compared to the established products. This may be an optical performance edge, a first-mover advantage, or an advantage at the system level.
• The emerging 100-Gb interface market for the data center offers a much needed high-volume opportunity, yet this market will be insufficient to deliver self-sustaining revenues for the transceiver suppliers.
In the tech sector, acquisitions offer one exit strategy for investors and start-ups, and usually help ensure the technology’s long-term survival. The trend of system vendors acquiring silicon photonics suppliers that was first evident in 2012–14 has reemerged in 2016. Ciena, Cisco Systems, Huawei, Juniper Networks, and Mellanox have all made such a move.
This shows that system vendors recognize that silicon photonics is a technology they need for their future product plans and, more importantly, they cannot just work with and depend on silicon photonics or chip players; it is a technology they must own.
Another development that bodes well for silicon photonics is the emergence of a strong ecosystem.
Fig. 4.9 shows a 2015 snapshot of market participants and how it has evolved in 1 year. Represented are diverse elements of the market: start-ups and established vendors; system, module, and component vendors; incumbent transceiver vendors; and the equipment makers. Also included are end customers as well as suppliers making products for various market segments, including the Layer 4 telecom network core and Layer 3 data centers.
Also featured are multiple top-ten global chip companies. This chapter has focused on the current telecom and datacom market opportunities and rightly so, as this is the current market route for silicon photonics. But we expect electronic designers will also require silicon photonics for their chip designs, and when that happens the large semiconductor foundries will start to lead the market. This will be a disruptive development, a topic explored in Chapter 8, The Likely Course of Silicon Photonics.