Keywords

Silicon photonics; Moore’s law; indium phosphide; vertical-cavity surface-emitting lasers; disruptive technology; systems; disaggregated servers

1.1 Introduction

In the 1950s the world welcomed the rise of the electronic transistor, which ultimately led to the popularization of the computer. In the 1990s optical technology enabled the exponential growth of data transmission, connecting computers globally, which gave rise to the democratization of the Internet and the World Wide Web. The next step in the journey of the digital economy—the application of optics to electronic processes and vice versa—is silicon photonics.

Optics—the use of light to send signals through a transparent path—is playing an increasingly important role in communications across a vast scale of distances. For two decades or more, it has allowed the networks of the telecommunications operators—the telcos—to cope with huge annual growth in data traffic. Now photonics is also playing a central role in the data center, where Internet data is received, processed, and distributed.

Silicon photonics can be viewed in several ways. From an optical component industry perspective, it is the most recent technology to join several established technologies used to make optical devices. This is a valid but narrow viewpoint, because silicon photonics is much more than that.

Silicon photonics enables optical devices to be made on a silicon substrate and fabricated in a chip facility. The resulting devices are starting to be adopted by the optical industry, but the technology’s commonalities with the much larger semiconductor industry is raising its profile among the chip giants. It is a technology the chip industry recognizes it will need, to tackle input–output bottlenecks in its more complex chips. The adoption of silicon photonics by the semiconductor industry will have far-reaching consequences, as is explained in the book.

This chapter introduces silicon photonics and addresses its significance. The status of silicon photonics and whether it has reached its tipping point are also discussed. Two other points are tackled briefly: its market opportunities, and whether silicon photonics is disruptive. This chapter highlights key issues and themes that are expanded upon in the book.

1.2 Silicon Photonics: An Introduction

Silicon photonics luminary Professor John Bowers of the University of California, Santa Barbara describes silicon photonics as bringing CMOS processing to optics.

CMOS—short for complementary metal-oxide semiconductor—has been the bedrock technology of integrated circuits for decades, and chip-making is one of the most advanced mass volume manufacturing processes ever developed.

Hundreds or thousands of chips, millimeters in size, are processed in parallel on a single silicon wafer measuring 300 mm (12 in.) in diameter. The bigger the wafer, the more devices can be made on it and the better the economics of chip-making. Silicon wafers are processed in chip fabrication plants, known as fabs, that run 24 hours a day, 365 days a year. Modern chip fabrication plants are hugely expensive factories, costing billions of dollars.

CMOS transistors made on wafers in these plants now have feature sizes as small as 14 nm, a fraction of the width of a human hair. Feature size refers to a key dimension of a transistor. The continual reduction in feature size by the chip industry has enabled ever more transistors to be crammed onto a chip, the consequences of which are described by Moore’s law. Based on an observation by Gordon E. Moore, the law states that the complexity of integrated circuits doubles every 18 months (later amended to every 24 months).

Silicon photonics aims to piggyback on the huge semiconductor industry—its know-how and the vast investments it has made over decades. Silicon photonics is not the same as the CMOS process used to make chips. Silicon photonics involves creating, processing, and detecting light and, not surprisingly, has its own manufacturing requirements that differ from those used to make electronic chips. These processes include not just the wafer processing to make the photonic circuits but also custom circuit testing equipment and device packaging.

But the benefits the semiconductor industry can bring to photonics are unquestionable. The chip-making process can be used to make efficient light pipes—waveguides—that direct the light between optical functions. The precision manufacturing of chip-making improves photonic device optical performance and device yields, and large silicon wafers benefit the economics of component making. The chip industry also brings packaging and testing benefits, as well as a sophisticated design tool environment.

Silicon has a key shortfall, however: it does not lase because it does not give off light—photons—when driven with electrons, a consequence of its electronic structure. Here, optical materials such as indium phosphide and gallium arsenide—known as III-V compounds based on the columns in the periodic table—are needed to provide a silicon photonic circuit’s light source, a topic discussed in Chapter 3, The Long March to a Silicon-Photonics Union.

The same applies to detecting light—converting photons to electrical current using a photodetector circuit. Silicon needs help, and here the element germanium is used. The chipmakers have already been down this path of adding materials to advance CMOS, so it is not a deal-breaker for silicon photonics. But as will be explained in Chapter 3, The Long March to a Silicon-Photonics Union, it is not trivial either, and several approaches are being pursued by the silicon photonics players.

Another distinction of silicon photonics is that, unlike chip-making, it uses much larger feature sizes. The minimum size of silicon photonics waveguides—the optical equivalent of wires—is governed by the light’s transmission wavelength. The light is in the infrared part of the electromagnetic spectrum; its wavelength is several orders of magnitude larger than an electron, which means that much larger feature sizes and hence older CMOS processing nodes—at 130, 90, and 65 nm—are sufficient to make the optical waveguides.

Compared to today’s 14-nm CMOS processes, these larger CMOS sizes are archaic. CMOS manufacturing equipment already mothballed has been given a new lease of life, thanks to silicon photonics. And such processes, no longer state of the art, are far cheaper to operate. In turn, the optical masks used to pattern and construct the chips are a lot cheaper to make for these older processes.

In summary, while silicon photonics is driving its own requirements, being able to use a chip fabrication plant brings huge advantages such as precision manufacturing, device yield, and volume manufacturing, which ultimately promises cheaper chips.

Silicon photonics is not about using silicon for everything. That misses the point, says Professor Bowers. The key element is using silicon as a substrate—on 12-in. wafers rather than the smaller 2-,3- and 6-in. wafers used for indium phosphide or gallium arsenide optical devices—and having all the process capability of a modern silicon CMOS facility.

1.2.1 The Application of Silicon Photonics

There are several battle lines between the different technologies when it comes to communicating data. And these battle lines are shifting as data rates continue to grow.

One competitive battle is between optics and copper wire. Optical technology has long secured the role of sending core network traffic over long distances. The amounts of data are too vast and distances too great—hundreds and even thousands of kilometers—for this to be done using copper wire. That is because copper’s capacity-reach product is orders of magnitude lower than that of optical fiber.

The current battleground between copper and optics is over distances of several meters to a few kilometers. Copper wire is still used to deliver data to the home in the form of telephone wires and broadband services, but when it comes to data centers and the higher gigabits-per-second links used to connect equipment, copper starts to run out of steam after a few meters. The underlying trend is that optics continues to advance, slowly pushing copper’s use to ever shorter distances.

There is also competition within optics, between the three main technologies used to implement communications. Indium phosphide has secured the long reach while vertical-cavity surface-emitting lasers (VCSELs), made using gallium arsenide, are used for tens of meters to a few hundred meters. And like copper, the reach of VCSELs diminishes as data rates continue to rise.

Silicon photonics is the newest of the three optical technologies. Being silicon-based, the technology suits being used ever closer to electronic chips. Silicon photonics has also been shown to work within an electronic chip, sending and receiving data on and off the chip. Such an application of silicon photonics is still some way off commercially, but it is coming. Fig. 1.1 shows the breadth of silicon photonics’ reach.

The issues of bandwidth and reach for the different optical technologies and for copper are discussed in Chapter 2, Layers and the Evolution of Communications Networks. Note that Fig. 1.1 segments distances into layers, a classification we introduce and expand upon in Chapter 2, Layers and the Evolution of Communications Networks.

1.3 The Significance of Silicon Photonics

The central tenet of silicon photonics is that it has an edge based on its ability to exploit the huge investment made over decades in the mass production of semiconductor chips. This is not something that established optical technologies such as indium phosphide and gallium arsenide can benefit from to the same degree. But while the basic premise is sound, the early reality of silicon photonics has proved more complex.

Silicon photonics uses silicon-on-insulator wafers. These 12-in. wafers are not mainstream silicon wafers but ones developed specifically for high-speed electronic circuits. However, silicon photonics benefits from the wafer’s insulating layer for optical waveguide construction. Optics uses three styles of paths to guide light. One is optical fiber, another is a planar waveguide, and the third is free-space optics. Optical waveguides play a central role in all the main silicon photonic circuit building blocks.

Much investment has been made—and more is needed—for silicon photonics to exploit the manufacturing capabilities of the semiconductor industry. So far, companies have developed their own particular silicon photonic devices that require different manufacturing processes, integration approaches, and schemes to attach the laser to the chip. As mentioned, silicon does not lase.

Another issue is component volumes. The success of the chip industry stems from its huge volumes. In 2015 some 1.4 billion smartphones [3] and 276 million laptops [4] were sold. In contrast, Acacia Communications—an early success story of the silicon photonics industry—announced in 2016 that it had sold 13,000 of its silicon photonics–based coherent optical transceivers, and that was over 2 years [5]. This equates to a few hundred 12-in. wafers a year—a trifling volume for today’s state-of-the-art semiconductor fabs. Such volumes hinder the full weight of the semiconductor industry coming into play and limit the interest of the big chip-making foundries—firms such as TSMC and GlobalFoundries that make the chips for “fabless” semiconductor firms.

So the notion of riding to prominence on the back of the chip industry has caveats, but silicon photonics has already benefited from the chip industry, and it will benefit the chip industry overall as is now explained.

To understand the significance of silicon photonics, several perspectives are used: what the semiconductor industry brings to silicon photonics, how silicon photonics benefits the chip industry, and how silicon photonics benefits systems design.

1.3.1 Silicon Photonics From a Photonics Perspective

From a photonics perspective, silicon photonics gains several inherent advantages by piggybacking on the chip industry.

• Larger wafers: Silicon photonics can exploit the chip industry’s much larger 12-in. silicon wafers. Given that some photonic circuit designs can occupy a relatively large die area, the cost of silicon photonic chip-making is lower than competing technologies due to more devices being made simultaneously on a wafer—a basic premise of the chip industry.

• Manufacturing precision and device yields: The chip industry has invested trillions of dollars in the chip-making process using masks, photoresist materials, and etching to shrink continually the feature sizes of transistors. Using such advanced manufacturing results in device yields that are higher compared to that of traditional photonic integrated circuits made using indium phosphide. Indeed, higher yields are what got Acacia Communications’ Chris Doerr, a veteran in indium phosphide photonic integrated circuit design, hooked on silicon photonics [6].
Doerr was at the famed US research institute, Bell Labs, for 17 years before joining Acacia in 2011. He spent the majority of his time at Bell Labs making indium phosphide–based optical devices and also planar lightwave circuits. Doerr had an opportunity to design a photonic integrated circuit using silicon photonics and chose to make a coherent receiver used for optical transport. What hooked Doerr was the high yield of the silicon photonic devices he made. He could assume each device worked, whereas when making complex indium phosphide designs, he would have to test half a dozen devices before finding a working one. And since yields were high, he could focus on making complex designs—devices with many photonic elements.
Silicon photonic circuits also match closely their simulation results. Indium phosphide is complex, says Doerr, now Acacia’s associate vice president of integrated photonics, and designers have to worry about composition effects and etching not being that precise. With silicon, the dimensions and optical performance are known with precision, meaning a design can also be simulated precisely, enhancing the design process.

• Silicon as an integration platform: Silicon not only implements optical waveguides and modulators (see Fig. 1.2)—used to encode digital data onto light before transmission—but also makes use of other materials such as germanium for photodetection. But it doesn’t end there. It is also possible to bond III-V materials such as indium phosphide and lithium niobate, a mainstay material for optical modulation, onto silicon and process it to become part of the working photonics circuit. This is the best of both worlds: using III-V and other materials when needed and the larger wafers and the processing of silicon [7]. This topic is discussed further in Chapter 3, The Long March to a Silicon-Photonics Union.

1.3.2 Silicon Photonics From a Semiconductor Perspective

The semiconductor industry has benefited from Moore’s law for over half a century. This has resulted in the continual advancement in the processing power of computer chips and ever denser memory chips, delivered like clockwork with each new generation of integrated circuit while costing about the same. Now the chip industry faces its own challenges as Moore’s law finally comes to an end.

The industry is looking at novel ways to squeeze more life out of Moore’s law but realizes that other strategies are needed if progress is to continue at the rate that Moore’s law has consistently delivered.

One issue is getting data on and off the chip. For more advanced chips such as switch chips that are used to connect the computing resources in a data center, this input–output issue is becoming a pinch point. Optics can be used to tackle chip input–output bottlenecks as well as extend the reach while lowering the power consumption compared to today’s high-speed electronic signaling (Fig. 1.3).

Electronics already plays an important role for optics. It can be used to improve the performance of optics, enabling what some call smart photonics. Examples include the digital signal processing techniques used to compensate optical transmission impairments in fiber over long distances, detailed in Chapter 5, Metro and Long-Haul Network Growth Demands Exponential Progress, as well as circuitry to control the stability of certain optical devices for tasks such as modulation.

Indeed, optics and electronics have even been combined on a single silicon photonics chip. Two companies, Luxtera and IBM, have developed single-chip monolithic designs combining the drive electronics with optics. Luxtera and the industry in general have since chosen to separate the two domains, with each on a separate chip fabricated using an appropriate CMOS process node: an advanced feature node for the electronics and an older CMOS node suited for the larger optical features. The two chips are then co-packaged.

Creating compact single-chip photonic devices and using electronics to improve the performance of optical designs highlight the beneficial relationship between photonics and electronics. But the most significant development is how silicon photonics promises to benefit chip design in terms of greatly enhancing a chip’s input–output capacity. And by doing so, it also benefits systems design, as discussed in Chapter 7, Data Center Architectures and Opportunities for Silicon Photonics.

1.3.3 Silicon Photonics From a System Perspective

When we refer to a system, we mean equipment that combines several technologies—or functional blocks—to perform a task. This task typically involves taking an input and processing it to produce an output. One example is a server, a rack of cards on which sit advanced microprocessors, memory, storage, and networking chips. The systems encountered in this book comprise various technologies to perform such tasks as computing, networking, data storage, and sensing, or combinations of these tasks.

One trend in the data center is the development of disaggregating systems, where the basic elements that make up a system are being rearranged and separated in a nontraditional way to deliver a performance or an operational benefit. But for this to work, what is needed is a high-bandwidth, cheap interconnect to link the disparate parts of the system.

One example is the disaggregated server. Here, the components making up the server—the processors, storage, and memory—are pooled separately but interconnected using high-speed links (Fig. 1.4).

The benefits of a disaggregated architecture are that different units can be upgraded as required—e.g., a processor node—without having to upgrade the complete server, thereby saving expense. Equally, the cooling needed for the equipment can be customized to the particular units generating the most heat.

At present such links in disaggregated systems can be made with copper or with VCSELs, but as the system elements continue to improve in performance and speed, this is an obvious application for silicon photonics. We discuss this in Chapter 7, Data Center Architectures and Opportunities for Silicon Photonics.

Another example of how silicon photonics can benefit systems is a microprocessor design that uses optics for its off-chip communications. A group of academics from several US universities have used a standard 45-nm CMOS process from IBM to create an advanced processor having optical input–output on a single chip [8]. Here the bulk of the chip area is digital logic, with the silicon photonics accounting for a small fraction of the die area. Such a design is not yet commercially viable, but it is a key milestone because it already demonstrates a capability that will be needed in future.

These systems examples highlight the beneficial relationship between photonics and electronics, with photonics delivering bandwidth and performance benefits.

Indeed, many argue that for the chip industry to progress, greater thought must be given to system design innovation, especially as Moore’s law, with its guaranteed chip performance benefits delivered on cue every two years, comes to an end. System innovation has always been a design challenge for engineers but promises to be even more so in future because of the demise of Moore’s law.

In recent years it has been the systems vendors—companies such as Cisco Systems, Juniper Networks, Mellanox, Huawei, and Ciena—that have been acquiring silicon photonics startups. They realize they will need silicon photonics expertise for the design of their own systems and for their custom chips.

Silicon photonics can therefore be seen as an important and timely technology adjunct for the chip industry. Andrew Rickman, tackling switch system innovation for the data center with his startup Rockley Photonics, is a firm believer that system design is where silicon photonics promises unique benefits [9].

For Rickman, using silicon photonics to make an optical component is not playing to its strengths. You can look at nanoelectronics in isolation and use silicon photonics for chip-to-chip communications, a good thing to do, he stresses. Or you can address, as Rockley aims to do, Moore’s law and the input–output limitations within a complete system the size of a data center, where hundreds of thousands of computers are interconnected. Clearly the scale of the problem being tackled—data center scaling versus solving an individual problem around a chip—results in a different approach, says Rickman.

Professor Richard Soref, described as the founding father of silicon photonics, holds a similar view. He is already thinking about new emerging applications for silicon photonics such as sensors and microwave photonics [10].

Soref talks about the near-infrared and part of the mid-infrared spectral range: 1.5–5 µm. This is above the spectral range of light used for datacom and telecom applications. VCSELs operate at 0.85 µm typically while long-range optical transport is around 1.5 µm.

The applications in this higher spectral range include system-on-a-chip, lab-on-a-chip, sensor-on-a-chip, and sensor-fusion-on-a-chip for such applications as chemical, biological, medical, and environmental sensing. Such sensor chips could find their way into your future smartphone and play an important role in the emerging Internet of Things, says Soref.

These are different systems from the ones associated with telecom and datacom but systems nonetheless.

1.4 The Status of Silicon Photonics

Silicon photonics has still to reach its tipping point despite being used in commercial products. By tipping point we mean the point when the technology transitions from early specialist applications to become pervasive.

The concept of a tipping point, popularized in the bestseller by Malcolm Gladwell, describes the moment when an idea, trend, or social behavior crosses a threshold and spreads like wildfire [11].

Clearly silicon photonics is still in its infancy and cannot be described as pervasive. And technology always advances at a more sedate pace than wildfire. But while it took close to 20 years (1985–2005) for silicon photonics to advance from a core idea to first products, the last decade (2005–15) has seen significant progress in its development. Now, various elements are being put into place such that silicon photonics is entering the final straight, with the tipping point finishing line within sight (see Fig. 1.5).

Luminaries that have done much to develop the field of silicon photonics have conflicting views as to whether the technology has reached its tipping point.

Mario Paniccia, who headed Intel’s silicon photonics program until 2015, defines silicon photonics’ tipping point as when people start believing the technology is viable and are willing to invest [2]. AIM Photonics, the $610 million publicly and privately funded initiative set up in 2015, is one indicator of that willingness [12]; Acacia’s successful initial public offering in May 2016 is another. Not only is silicon photonics being seen as viable, but players are investing significant funds to commercialize the technology. Paniccia also points to the many companies now selling commercialized silicon photonics. In his view, silicon photonics’ tipping point has passed.

Another luminary, Graham Reed, Professor of silicon photonics at the University of Southampton’s Optoelectronics Research Centre, gives an academic perspective. In the 1990s it was a challenge to get funding to research silicon photonics, he says. Now, his group is regularly approached by companies from all over the world that are either active in silicon photonics or plan to enter the market.

Given how some of the largest companies have been investing in silicon photonics for over a decade, Reed is surprised that more products are not available commercially. But he believes it is inconceivable that the firms that have made the investments will not launch products. So, in that sense, the tipping point has already been and gone, argues Reed. All these events are of a technology pointing to mass market, he says.

Reed describes this as an industry “quiet period,” with photonics companies working to commercialize their technologies, and systems vendors developing next-generation products and evaluating various technological options. It is just a question of time before a vendor “jumps” and the market takes off—and he is confident somebody will jump. Once that happens, he expects a period of ferocious activity to follow [13].

Other luminaries do not yet see the tipping point, Richard Soref being one. For silicon photonics to be ubiquitous, it will take much investment and many commercial results, he says, and we are not at that stage.

There are hurdles for silicon photonics to overcome, and some of them are chicken-and-egg ones. Foundries will be interested in offering silicon photonics fabrication services once there are large unit volumes, but that requires a variety of companies—fabless chip players, sensor companies, and systems houses—to have their designs made in volume.

At present companies are developing custom solutions for manufacturing and packaging, as mentioned. This presents a barrier to entry to those interested in using the technology but lacking the resources to develop their own solutions. And while some of the existing silicon photonics companies may consider making their technology available to the industry, they will only make additional investment if there is enough interest—another chicken-and-egg situation.

A key industry focus is to improve the manufacturing of silicon photonics circuits on a commercial scale, a hurdle that must be overcome for widespread use to occur. Silicon photonics players such as Acacia Communications, IBM, Intel, Luxtera, Mellanox Technologies, STMicroelectronics, Macom, Kaiam, Cisco Systems, Oracle, and Juniper Networks are investing and developing silicon photonics circuit design and packaging, but they are using custom approaches based on their own intellectual property. Until there are higher volumes, foundries may not make the investment to take their intellectual property to the next level to offer it commercially to interested parties.

There is a complex interaction between technology advancement, volume demand, and manufacturing progress, as shown in Fig. 1.6. The technical and manufacturing challenges facing the industry are engineering issues rather than show-stoppers. Engineering effort and investment will thus resolve these challenges rather than requiring some fundamental innovation that cannot be scheduled or predicted. This is why we are confident silicon photonics will reach its tipping point.

Many of the conditions that must be met for silicon photonics to reach its tipping point are now in place. Investment in the technology is healthy and increasing, more companies are embracing it, and a wider base of products is coming to market.

But the technology is not widely adopted, and applications that will deliver significant unit volumes have yet to start.

We expect silicon photonics to break out in the coming years. At some stage around 2021, it will become evident that silicon photonics has reached its tipping point. What is happening now can best be described as entering the era of silicon photonics.

1.5 Silicon Photonics: Market Opportunities and Industry Disruption

For completeness, we end this chapter by touching briefly on two other core issues regarding silicon photonics aside from its significance and status. The first is the market opportunities going forward for the technology. There is also the question of whether silicon photonics is disruptive. These two issues are discussed in Chapter 8, The Likely Course of Silicon Photonics, which brings the book to a close.