Silicon photonics extends its reach
The optical communications market is undergoing a seismic shift. This is driven in part by the emerging role of internet content providers who have established themselves not only as leading users of interconnect technology, but also as a disruptive force aggressively transitioning the market toward a fast-paced cloud, software-driven, and data centre-optimised design model. The widespread adoption of cloud-based services is leading to a tremendous increase in deployed capacity and a fast ramp-up of 100 Gigabit Ethernet, small form factor, short-reach interconnects. Those who have suffered through the slow introduction of first-generation 100G short-reach pluggable modules in the CFP form factor are astonished by the volumes and steep price decreases for the latest QSFP28 designs, with pricing reaching an estimated $3 per Gb/s shortly after the market introduction, and potential to drop further before the end of the year.
Each disruption in the end-user market requires an underlying technology which enables the transition toward a scalable, high-volume market appliance. What could be the equivalent of the Arm processor that broke Intel’s grip on the computing market and enabled the dominance of current smart phones? In this context, the optical market looks to silicon photonics as one of the potential enablers for scalable interconnect technology.
Initial research on silicon photonics started about 20 years ago, with the likes of Intel and IBM setting the ground for the industry’s first commercial applications. Despite being a technology with years of research behind it and wide application in other fields, silicon photonics is only now finding its way into commercial applications in the telecom industry.
So far, the application range for silicon photonics has been rather short distances, with focus mainly on datacom applications. One commercial component supplier, for example, is delivering its first pluggable transceiver, the 100G QSFP28, to support the CWDM4 MSA and CLR4 Alliance industry standard specifications.
However, silicon photonics is being proposed for, and already being used in, longer distance applications such as 100G data centre interconnect (DCI) and metro transport. Several component and module suppliers have recently announced products based on a silicon photonics engine, with some already using the technology in the 100G coherent CFP and CFP2-ACO form factors. Moreover, silicon photonics has been an enabling technology addressing specialised applications such as a DWDM direct detect 100G QSFP28, which can be plugged directly into networking switches, making a stand-alone data centre muxponder unnecessary in this scenario.
Recipe for success
What is silicon photonics about, and what are the key ingredients and success factors? In general, looking at integrated photonics – and silicon photonics in particular – the benefit lies in building complex chip-scale systems rather than discrete devices, and this is what you see within the commercial applications that exist today. An integration of most building blocks on-chip eliminates free-space optics, isolators, and all other discrete optical components. Therefore, the appeal of silicon photonics is not in its use in simple applications such as PON – where it would have to compete head-to-head with the already very cost-effective directly modulated laser or vertical-cavity surface-emitting laser (VCSEL) devices – but to focus the technology on network that traditionally required more complex interconnect designs.
The appeal of silicon as a material goes well beyond the integration of complex chip-scale systems, however. With the industry looking to comply to $1 per Gb/s pricing and below (a target originally voiced by Facebook), a holistic look at silicon reveals the material’s benefit in the big picture. In this perspective, silicon photonics enables:
- the leverage of existing billion-dollar scale manufacturing facilities to enable rapid scaling into production and low chip cost;
- the use of photonic development kits (PDK) in order to innovate on design in an established process, rather than innovating on process improvements;
- the use of photonic integrated circuit design tools in order to have device level simulation, wafer-scale process, schematic, layout and simulation unified in a single environment;
- Taking advantage of a mature electronics chip packaging technology, going away from expensive gold box butterfly housings toward non-hermetic ball-grid array (BGA) electrical circuit types of package; and
- the utilisation of a wafer-scale assembly, test, and characterisation.
Therefore, using silicon photonics is about more than just using silicon as a material; it’s about leveraging the material and all the capabilities of an ecosystem established by a large-scale CMOS industry.
Looking into the past, silicon always succeeded not because of the material itself, but because of the established ecosystem, which is the main difference in comparison with other material systems used within the optics industry so far. In the past, researchers tried to find the best respective material for all the dedicated functions used in the optical transmission system, ending up with ‘gold box’ types of package, including a large number of individual components in different materials that had to be manually and actively aligned with lens systems and rather complex optical paths. This might result in optimised performance but, from a design and manufacturing standpoint, this approach does not scale well to high volumes.
As all functions besides the gain medium could be integrated in silicon photonics itself – and the III-V gain medium (such as indium phosphide) can be bonded or butt coupled to the silicon – the solution is a true integrated photonic circuit, where a large scale wafer process and automation techniques could be used, instead of multiple tedious processing steps. Thanks to its integrated functionality, silicon photonics circuits can have a large test coverage on the wafer already, whereas classical optical components first need to be packaged up completely before running the majority of tests. As a result, impacts on yield can be eliminated at a much earlier manufacturing stage, contributing largely to the lower overall cost profile of this technology.
Will emerging technology advances in silicon photonics platforms be able to support the required bandwidth increases at lower cost per bit? With experience as an early adopter of silicon photonics in metro and long-haul solutions from the outset, the answer is clearly yes.
Today’s commercial silicon photonics applications support symbol rates up to 34Gbaud carrying 100, 150, or 200Gb/s per lambda, which enables next-generation transmission technology, as seen on the ultra-dense DCI coherent muxponders with up to 3.2Tb/s per rack unit. Going forward, the evolution of silicon photonics in coherent networks could address two distinct trends – the miniaturisation of a 100G coherent pluggable form factor, scaling down to CFP4-ACO footprint, or an increase of the bandwidth of a physical port to more than 64Gbaud, enabling scalable modulation rates between 100Gb/s and 600Gb/s per lambda.
Here, one of the enablers will be a co-design of optics (modulator, receiver) and electronics (driver amplifiers, transimpedance amplifiers) in a co-packaged assembly in order to achieve the highest bandwidth, without the need for several discrete packages, as currently seen for standard lithium niobate modulators. Although some others have speculated about a further integration of photonic and digital electronic circuitry down the road, in our view the growing difference in the process nodes used for optical and electronic circuitry could eventually allow for a co-packaging on an interposer only, without the ability to realise a fully integrated design of optics and digital signal processor chip in a single die long-term.
The principle of good enough
The optical performance of coherent silicon photonics has been good enough to open up a door in metro and long haul network applications. Although limitations existed for first-generation products, performance has continually improved. Going forward, there is no fundamental obstacle that would prevent silicon photonics achieving performance on-par with competing material technologies, with first published demonstrations of 64QAM transmission backing up this claim. Even assuming a slight difference on a technological level, it is unlikely to translate into a perceivable differentiation for the end user when compared to III-V semiconductor or lithium niobate-based products.
So what is the long term role of silicon photonics in datacom and telecom markets? As today’s fast-paced and demanding cloud and data centre-centric communications world raise the networking performance bar as it strives for higher density, lower power consumption, and lowest cost-per-bit, silicon photonics is without question a potential game changer. Silicon photonics has clear application value in both metro and long haul core networking environments, and is also a fundamental technology enabler of disruptive architecture models such as disaggregation and open line systems. In the end, silicon photonics will give the industry a scalable tool set to address a wide breadth of current and future challenges. Such a shift in the industry requires a disruptive enabler, and silicon photonics is certainly looking like a prime candidate to fill this role.
Dr. Marc Bohn is head of research and technology at Coriant GmbH
Gazettabyte (May 2016) Mario Paniccia: We are just at the beginning