Integrated photonics platforms compared: Silicon Nitride, Indium Phosphide & Silicon Photonics

08 March 2023

No technology platform can do it all in integrated photonics. In this final installment of a four-part series, we’ll review strengths and weaknesses of the major photonic platforms and take a closer look at how combinations of them unlock the full potential of integrated photonics.

Whenever an emerging technology comes in multiple flavors, you’re inclined to assume one of them will eventually win. It’s not a law written in stone – two operating systems share the smartphone market, for example – but it’s common a dynamic that whatever platform finds the most traction in the market first, starts to snowball and eventually pushed out competitors. One famous example is VHS beating Betamax and Video 2000 in the video format wars of the 1980s.

This will not happen in integrated photonics. Having reviewed the three major integrated photonics platforms Silicon Nitride (SiN), Indium Phosphide (InP), Silicon Photonics (SiPh), we can say with near certainty that we won’t see one of them prevail over the others. Yes, in some applications they may be competitors, but it’s much more useful to think of them as complementary. For many applications, any one platform can’t shine without the other: they need to be combined to unlock the desired functionality.

In this concluding part of the series, we’ll compare the characteristics of the platforms, explore how they can work together optimally.

A note on emerging platforms

Although InP, SiN and SiPh are currently the most mature integrated photonics platforms, other materials offer attractive properties and new functionalities. For example, barium titanate, lithium niobate and polymer-based photonics offer a great approach to ultrahigh-speed modulators and gallium compounds (GaN and/or GaAs) for visible-light applications. These nascent technologies will be great additions to the integrated photonics toolkit.

Comparison of the platforms Indium Phosphide, Silicon Nitride and Silicon Photonics

If you are going to make an integrated circuit based on light, you need materials that can produce light and materials that can ‘conduct’ it. InP is the only base material for integrated photonics that can do both. This means InP is the only platform with the inherent ability to integrate active components such as amplifiers and detectors. No wonder InP has the longest commercial history of all three platforms, finding use in optical communications technologies decades ago already. This is still InP’s main market, but it’s in the running for (automotive) sensing applications as well.

Wavelength ranges for a number of applications and the window of transparency for three major integrated photonics platforms. Apart from InP, other common light sources include GaAs (datacom) and GaN or GaAs (biosensing).
Credit: Roland Berger/PhotonDelta.

Still, there are very good reasons to choose SiN or SiPh over InP. The former features a spectral range that roughly spans from the UV to the edge of the near-infrared (see figure 1), rendering SiN the only platform suitable for visible-light applications such as biomedical sensing and augmented and/or virtual reality. SiN is also unique for its low light losses, which enables more complex circuitry.

Finally, SiPh can’t top InP or SiN on any of these characteristics, but it has one important redeeming quality: it can be manufactured in the same fabs already used to process electronic circuits. Having a vast manufacturing infrastructure at the ready is a boon for high-volume applications. Integrated photonics hasn’t quite reached that stage yet, but that may change once optical sensors become commonplace in consumer products, for example. Or when electrical interconnects in computing gear run out of steam and optical ones need to replace them. Currently, SiPh is mostly used to manufacture high-bandwidth transceivers.

It should be noted that SiN is, in principle, also CMOS compatible. With that in mind, there’s something to be said for considering SiN and SiPh to be part of a single silicon family. Depending on the desired material characteristics, however, SiN photonics processing may require conditions that aren’t compatible with certain functionality, so in this respect SiN may not always be a universal alternative for SiPh.

The core platforms and their main application areas. Credit: Roland Berger/PhotonDelta.

How to combine PIC platforms by integration?

It’s clear that SiN and SiPh will always need InP for it’s ability to generate light. InP, on the other hand, will never be able to process visible light, or match the vast manufacturing infrastructure of electronics. To unlock the full potential of integrated photonics, the platforms need to work together.

This is most easily achieved by manufacturing required components separately and integrating them into higher-level assemblies afterwards. This is called hybrid integration. Alternatively, a process can be developed in which two or more materials are combined. For example, a sliver of InP can be fused with silicon to create a laser directly on silicon. This is called heterogeneous integration.

Hybrid integration

Roughly, hybrid integration is what’s readily available now and what’s most appropriate for low-to-medium volumes, says Jeroen Duis, chief commercial offer of photonics assembly foundry PHIX. Hybrid integration is an everyday routine at his company, stacking photonic chips on top of each other or connecting them side-by-side, making sure optical connections are made with as less little loss as possible.

“Losing light is one potential challenges of hybrid integration. This can be mitigated through precise alignment of waveguides and ensuring high-quality connections. We’re talking 0,1 micrometer precision here. Edge-to-edge coupling of chips yields the best results, because the light stays in-plane. Emitting light out-of-plane requires gratings or mirrors, which are large and hard to produce on-chip,” Duis explains.

“Hybrid integration is highly flexible. You can chose whatever foundries work best for you to manufacture the components you need and assemble them afterwards. Or perhaps you buy an off-the-shelf InP laser to power your SiPh design,” Duis adds. “At this stage in the development of the industry, this mix-and-match method makes the most sense.”

Heterogeneous integration

For many applications, that might never change. Once an application evolves to high volumes or complexity, however, it may become more cost-effective to manufacture multiple platforms on a single substrate, called heterogeneous integration. Developing such a process is expensive, but pays off for high-volume manufacturing. Intel, for example, has been working on integrated InP light sources for SiPh. “Heterogeneous integration is still far off. But there’s no doubt in my mind that it will take off once integrated photonics becomes a high-volume business.”

Duis: “You have to take into account that a typical assembly step in integrated photonics takes a minute. In electronics, thousands of steps per minute is commonplace. This poses limitations on what can be assembled cost-effectively. Technological and yield issues aside, it’s not feasible at the moment to integrate, say, twenty light sources. I know of at least company that discovered this the hard way. It shows there’s still some way to go for certain applications.”

Hybrid integration, however, will remain a staple in the industry, not only because some applications won’t attain the volumes needed for heterogeneous integration, but also because even the most basic integrated photonic device requires control electronics. The production of every single optical module therefore involves at least one hybrid integration step: to connect and package an optical and electronic chip into a functional module.

Integration of photonic chip platforms starts in the design phase

Hybrid integration isn’t a mere afterthought. Great care must be taken that different components in the assembly play nice with each other: optimal alignment of photonic platforms starts in the design phase.

Efforts to ‘connect’ InP, SiN and Si at that level have recently taken off in the Dutch PhotonDelta ecosystem. The goal is create design libraries of building blocks, which are geared for the best results in hybrid integration. There’s probably no better place in the world to do that, as it unites the industry’s key players every platform: SMART Photonics for InP, LioniX International for SiN and Imec for SiPh.

To develop a disruptive generation of terahertz transceivers, the European Teraway consortium leverages three different photonic platforms. Credit: PHIX/Teraway.

Partly by facilitating hybrid integration, partly on its own merits, developing a world-class design library will facilitate worldwide photonics adoption by decreasing time to market. It will also serve to further strengthen PhotonDelta’s grasp on the industry, adding control over the design building blocks in addition to having manufacturing and assembly capabilities.

How does designing with building blocks work?

The term building blocks may conjure up images of engineers dragging and dropping functional pieces to create a design. But things aren’t that simple, says Ronald Broeke, CEO of technology-independent design house BRIGHT Photonics. “On a high level, perhaps you can combine building blocks using standardized software, but I wouldn’t bet on it that this results in the device or component you were hoping for. In practice, you need still to consider the physics of the manufacturing process. Tiny details and deviations can drastically affect performance, for example because of scattering or reflections of light. At BRIGHT, for example, we sometimes even need to consider which method a foundry employs to manufacture its lithography masks.”

How to increase predictability of make chip design outcomes

For a predictable outcome, you need detailed information on manufacturing variations and all the applicable interfaces, eg in hybrid integration, connecting to fibers, or connecting to drive and read-out electronics. Part of the problem is a lack of such data, Broeke explains. “You need statistics that relate design to real-world performance of the photonic integrated circuit. You obtain these through a large number of wafer runs and assemblies, but the volumes in photonics are still relatively low. Every new run, we add to our knowledge.”

“At the same time, there are so many parameters to consider, which differ between applications. It will take quite sometime before photonics will be at the same level as electronics in this respect.” Broeke notes that photonics design is more like RF and PCB design than digital electronics. It’s mostly analogue, it needs careful selection and qualification of components, and dimensions ultimately remain at the level of a wavelength.

“If, for instance, a simple beam splitter design has been put to the test for performance at 1550 nanometers, we still don’t know if works equally well for differently polarized light. And what about a broader wavelength distribution, say 1500 to 1600 nanometers? There are so many facets to a building block that are relevant for the application. But we learn more with every design that makes it to the market successfully.”

“Eventually, we want to get to a point where every design engineer or researcher can be confident that every reasonable design he comes up with, works as expected when it’s been manufactured. That kind of predictability and reliability is what the Photondelta building-block project is all about. We have three to five years to make that happen. And given the maturity of the ecosystem, if we don’t succeed by then, I’ll personally turn off the lights at BRIGHT and start a new career,” Broeke smiles.

This article was originally published in Bits & Chips. Author: Paul van Gerven.

Read more: