Silicon photonics expands from its datacoms roots into new markets

Over the past 20 years silicon photonics has made a successful transition from academic research field to industrial ecosystem. Transceiver products are thriving in the market. Industrial foundries offer mature process flows and process design kits (PDKs). EDA-companies (electronic design automation) offer photonic IC design tools. Nevertheless, in many ways, silicon photonics remains a small niche field in the semiconductor industry.

Meanwhile, both the research community and a multitude of start-up companies are preparing the next wave in silicon photonics, with scientific achievements and innovative products in a dozen new applications and markets, some of which may become large volume. The value proposition is clear. But the diversity of applications requires new functionalities and new — more heterogeneous — process flows. Will the business proposition follow and become sustainable?

These developments and questions were considered in the OPTO Plenary on January 29, by silicon photonics expert Roel Baets, who is an emeritus full professor at Gent University and imec, both located in Belgium. For many years Prof. Baets has made contributions to research on integrated photonics (silicon, silicon nitride, III-V) and its applications in datacom/ telecom as well as in medical and environmental sensing. He has founded and has chaired ePIXfab, the European Silicon Photonics Alliance. He is a Fellow of IEEE, EOS and Optica. He has been recipient of amongst others the 2020 John Tyndall Award and the 2023 IEEE Photonics Award.

He told Show Daily, “The overall aims of the plenary talk are to discuss the enormous value proposition of silicon photonics for applications and markets beyond transceivers, but also to discuss the techno-economic challenges involved including my thoughts about the best way forward. My aim is to make the community more aware of the strengths and the opportunities, but also about the threats. I believe it makes sense to try to put noses in the same direction.

“Attendees can expect to learn how we should deal with new SiPh applications of high value that require substantial process flow investment; and how we should deal with new SiPh applications of high value that are not geared towards high volume anywhere soon.”

Roel Baets, professor at Gent University and imec, Belgium. He is pictured by the clean room facility where his group develops micro-transfer printing. Credit: U Gent / imec.

Diverse new market opportunities for the previously datacoms-focused SiPh include the likes of optical gyroscopes, mid-infrared spectroscopy, AI and neuromorphic applications, quantum technologies, chemical analysis, lidar, and diagnostic sensing systems. What is important to enable these diverse applications is the integration with silicon of different materials (semiconductor and other) that enable new photonics applications. “That huge diversity is part of the promise and part of the challenge,” said Prof. Baets.

In relation to SiPh, Prof. Baets wears two hats: he is involved in the Photonics Research Group at Gent University, which is also associated with imec. In Gent the research group consists of over 100 people working in the field of silicon photonics. A notable capability of his research group at Gent and at imec is heterogeneous integration by micro-transfer printing. Baets’ second role is with ePIX fab, which is the European silicon photonics alliance, which he chaired for many years.

He commented on the growing diversification of SiPh application areas: “The simple story is that if there could be one type of silicon photonics for all the application cases that people are now considering then matters would be relatively simple both for large-volume and small-volume customers. If one looks at the mainstream commercial uptake of this field, high bit rate, high capacity transceivers are obviously the main product of silicon photonics in the market today and they rely on similar forms of manufacturing, enabling passive waveguides and high-speed modulators and detectors in germanium.”

It is because of this diversification, argues Prof. Baets, that the sector needs additions to existing processes. “Some people even dream of entirely different process flows in silicon photonics,” he said, “but the big challenge is to bring up new process flows in an industrial foundry environment.”

He added, “Many of those new applications are being explored by the research community but also by new start-ups that live under venture funding. So I think I recognise the mismatch between what it takes to bring on stable, mature process flows for these new variations on a theme and the market potential. The value proposition is clear but of course to make it work in a techno-economical context is a big challenge. In my plenary talk, I want to first of all spell out this challenge. I think that is being done too little.”

Commercial position and expectations

Considering the commercial potential of the non-telecoms SiPh market, Prof Baets is cautiously optimistic: “In the long run, it will be easier when there will be sufficient volume to sustain the whole operation of manufacturing along the complete supply chain,” he said. “That means not only the chip-making but also from design and TDA and through to packaging and testing.”

He added, “The long-term picture is sort of rosy, but it’s not obvious that there is sufficient financial capability to get through the gap. For a while the volumes will be small meaning that revenue for the fabs will be small. But somebody will have to invest heavily in developing stable process flows with all its associated machinery.”

What triggered the market to diversify beyond telecoms? Prof Baets said there is a clear explanation for this, which started a few years ago: “It’s quite simple. The first thing is the existing market — in the transceiver case, people want to move onto higher speeds and lower power consumption. Already, the mainstream SiPh platforms have run out of steam. Rates up to 50 gigabaud are now mainstream for SiPh. Of course, there is always demand for more throughput, so users and system developers want to move up to 100 and 200 gigabaud. Those rates are not yet mainstream, so developers are looking in different directions,” he said.

Prof. Baets continued, “Over the past two years or so there have been several independent demonstrations of 100 gigabaud and more. Intel, for example, demonstrated such at OFC 2023 and there are a few academic groups who have also achieved that rate. In parallel with that, other groups are already thinking beyond 100 gigabaud. I think most people will agree that using the normal carrier depletion-type plasma dispersion modulators in silicon photonics will indeed run out of steam.

“Then you will really want to move to a proper electro-optic Pockels modulator and that’s where lithium niobate, barium titanate, and other materials come in. We are already seeing quite a dynamic scene around the world of various groups using different routes; there is the BTO route, in which IBM has been successful; it has a spinoff developing that approach. With lithium niobate, there are several companies developing platforms based on that material.”

Wavelengths beyond communications

When the field of silicon photonics revolved around fiber optics, most development and commercial attention focused on singlemode fiber optics, particularly that supporting the C-band (1530- 1565nm) and O-band (1260-1360nm) ranges of wavelengths. As the focus moves from communications into sensing, for example, the need arises to support other wavelength ranges.

Prof. Baets explained, “You really want to move either in the direction of the visible, for example, in the fields of lidar and biosensors, and for other applications you want to move towards the mid-IR, such as applications in vibrational spectroscopy, meaning chemical sensing.”

Besides the needs for higher speeds (and lower power demand), and new wavelength bands, there is a third objective that SiPh innovators like Roel Baets are seeking: to integrate the light source on the chip. “With classical SiPh you don’t have an integrated light source,” he said. “But now there are already a few examples where integrated light sources have achieved industrial maturity; the best known case is, of course, Intel’s transceivers, in which they have the integration of bonded indium phosphide multilayer stacks and the processing of lasers, SLEDs or SOAs onto the same platform.”

Micro-transfer printing

A new approach, which is one of the focus areas of the UGent research team — and part of his plenary presentation — is micro-transfer printing. He explained, “A prominent technology gaining a lot of traction today is micro-transfer printing; a well-known technique already employed for micro LED integration, is now also being used in enhancing photonic integrated circuits, with, for example, a finished DFB laser or a semiconductor optical amplifier (SOA).”

Heterogeneous integration. Microscopy image of a C-band InP semiconductor optical amplifier (fabricated by III-V Lab) micro-transfer printed on a Si-on-SiN waveguide circuit (fabricated by imec) to make narrow-linewidth tunable lasers. Credit: U Gent / imec.

“In my presentation, I will be advocating heterogeneous integration techniques, and micro-transfer printing is one example, that minimize the complexity and the cost of developing mature manufacturing process flows. It helps when the integration falls back as much as possible on already existing process flows and when the integration happens late in the backend.

“I will be illustrating what I’m talking about with various examples. What we see today is that while micro transfer printing up to not so long ago was mostly an R&D tool, we are now seeing considerable interest from industry and we see traction from the industrial side to really see it as a manufacturing tool and to start developing the supply chain for such a method.”

So who is likely be doing this in a possible commercialized future? Prof. Baets said this remains an open question: “Because we would be bringing together wafers from a silicon foundry on the one hand, with wafers from a different III-V foundry, so there could be two different actors,” he said. “Who will be the industrial actor taking care of integration? Will that be in the silicon fab, perhaps? “That’s one model and I know of at least one case where it seems to be evolving in that direction; or could it be by distinct actors or boutique fabs that are specializing in exactly that.”

Relatively small volumes

When considering the volume of today’s silicon photonics transceiver business, Prof. Baets estimates that there are probably around 5 to 10 million SiPh-based transceivers being manufactured and sold each year. He commented, “From a photonics and a telecoms perspective that is considered to be quite a high volume, but from a semiconductor industry perspective that is small volume. Because if you calculate how that translates to the number of wafers then you come to the conclusion that a major manufacturer, whether GlobalFoundries or TSMC or whoever, would have to run a very small fraction of their capacity to serve all the needs for silicon photonics transceivers in the world today.”

He said therefore that, whatever the likely evolution of silicon photonics beyond communications, it will require significant investment and beyond that a great deal of patience by investors and manufacturers before the application volumes become commercially significant — and therefore valuable.

He said, “We can list more than 100 companies today that are developing products based on silicon photonics that are not for transceivers. None of them is yet in the market with any level of scale; some have their products for sale at the moment but they’re selling the order of hundreds of them into the market. So the volume today of non-transceiver silicon photonics products going into the market is the peanuts of peanuts! Many of them, unless they can piggyback on the mainstream existing platforms, have not secured a reliable scale-up route for mass manufacture.”

Prof. Baets concluded, “What are the positives in this? If somehow we can solve the problem of establishing this scaled-up manufacturing route then the value that we can create eventually remains huge.”

This article was originally published in the Photonics West Show Daily in January 2024.