Silicon photonics has been an area of much research and investment for at least two decades, but the market has yet to really take off. A company called OpenLight, which is offering an integrated and, yes, open approach to designing and producing silicon photonics-based devices, is looking to finally unlock greater potential, and its approach could also help the technology be applied to more products across more industries, including automotive lidar sensor systems, supercomputing, AI, and more.
OpenLight itself has been on a journey that mirrors the meandering path silicon photonics has been making toward relevance. At the beginning of the previous decade, a start-up called Aurrion was spun out of important silicon photonics research being done at the University of California-Santa Barbara. That happened just as a handful of other silicon photonics start-ups were emerging and optical communications system vendors were becoming enamored with the potential for photonic integrated circuits, rather than electronic ICs, to be the basis for new products supporting data transmission at hundreds of gigabits per second.
Then a wave of consolidation hit as larger companies started looking for ways to more easily integrate the technology into their own products, resulting in Aurrion being acquired by Juniper Networks in 2016. But, more broadly, silicon photonics’ rise was limited by challenges, including concerns around the complexities of accurately integrating lasers into silicon, the slow and challenging production process for the technology, and the difficulties of finding and working with CMOS foundries to produce photonic integrated circuits.
More recently, interest has been rising again. Intel, Nvidia, and others invested in silicon photonics start-up Ayar Labs in early 2022, and by June of last year, Aurrion became into OpenLight, as Juniper spun off the company and welcomed Synopsys as a majority investor in the firm. A little over one year later, OpenLight, now led by optical communications veteran Adam Carter as CEO, believes it has figured out how to spur production of silicon photonics-based devices by making things easier for product designers through an “open foundry” model.
OpenLight’s different take
The company offers a process design kit (PDK) that packages integrated lasers and other components and processes that foundries need to manufacture silicon photonics devices, which makes it much easier than before for a product designer to get their own design from the blueprint phase into production with a foundry partner. This addresses one of the major hurdles the sector has faced in recent years. If customers do not have a foundry partner, OpenLight can pass the production-ready design on to its own foundry partner, Tower Semiconductors.
“If our customers have their own silicon photonics design team, they can go in and access all components. We don't hold anything back,” said Carter, who has tracked silicon photonics for many years, since he was involved in Cisco Systems’ earliest efforts to invest in the technology. “In the olden days, when silicon photonics first came out, most companies kept their PDKs to themselves. They made the product and sold that product to the end customer. Why this model is very different is it basically offers customers the ability to do their own designs, and they have full access to all the componentry, and as part of that we do heterogeneous integration of the laser in the silicon. We don’t make and sell the products.”
Carter added that offering a PDK with integrated lasers will be increasingly valuable as silicon photonics designs scale and it becomes even harder to accurately align lasers with optical waveguides or fibers. “Where the heterogeneous integration really plays is as you scale the number of channels that you need,” he explained. “It's getting harder and harder from an alignment point of view of single mode lasers to either an optical waveguide or an optical fiber. There's a lot of capital equipment that needs to be bought to do that because you have to actively align these. What integrating lasers does for you is it takes out a lot of that cost in the backend manufacturing of whatever the module is that you're designing for whatever application.”
OpenLight’s PDKs work with photonic IC design platforms from Synopsys, which owns 75% of OpenLight, and Luceda, and OpenLight is intent on expanding its roster of both IC design partners and founder partners, Carter said.
The Synopsys photonic IC design solution that OpenLight’s PDK works with includes indium phosphide active optical elements on-chip that can be directly used by Synopsys OptoCompiler and simulated with the Synopsys OptoSim photonic simulator, all of which allows customers to create PICs with optical amplifiers, on-chip lasers, and high-speed, low-loss modulators tailored to their design requirements.
OpenLight recently started offering a PDK Sampler to help customer better understand what the PDK can provide and how it can benefit them.
But, OpenLight also offers its own design services to create reference architectures for customers that do not have their own design teams. Carter said that in this case it can give the customer a file with “everything on it, a full design that they can take to a foundry,” such as Tower or another foundry that is willing to work with that customer but prefers the customer come in with their own full design.
An expanding market opportunity
Though demand for silicon photonics has been fairly limited to optical communications, Carter said he sees opportunities expanding into lidar systems used in automotive, as well as supercomputing and other high-performance compute and memory applications, and AI and machine learning.
Carter explained at length how photonic integrated circuits could vastly improve the economics of manufacturing long-range lidar systems: “To get refresh rates from a pixel level for long range lidar, you have to increase the number of channels, as [the system] has to be looking at things are 200 to 300 meters away. So, you need to amplify the signal, and today a lot of companies are using discrete optical amplifiers, which is pretty costly. For example, if you need 16 channels to get the minimum pixel refresh rate you need, then you need 16 amplifiers. There's also a lot of alignment to be done. If you misaligned amplifiers to the waveguides, you basically have to scrap the whole whole system. So what we do is and we're actively designing for one or two customers now are photonic integrated circuits that have 16 channels and have semiconductor optical amplifiers inside those channels.”
He added, “What that means is they get much clearer resolution and a quicker response time [in a vehicle] to let you know if there's an object coming and you need to avoid it. Eventually that need will probably go to 32 channels. With the conventional way that they do things today in that industry, it’s cost prohibitive. It's a huge amount of investment that you have to put into backend manufacturing. What photonic integrated circuits do is take things down to a very small die size, so this can be small, compact, less heavy, which for cars is important, and it just enables something that's very different, a different opportunity for these [automotive] guys to really scale down what they need.”
Jack Gold, president and principal analyst at J. Gold Associates, also highlighted the emerging need in the HPC sector. “The challenge most high performance computers now have is that interconnect is the limiting criteria for increasing system performance,” he stated in an email to Fierce Electronics. “Interconnect between the processor, co-processor, accelerators, memory, I/O can significantly reduce the system performance, and new ways to find even higher bandwidth is critical. Existing high speed copper-based connections like Infiniband are reaching their limits. Silicon photonics is the way next-gen high performance systems will be interconnecting. So the race is on to find a way to produce silicon photonics, hard to do well in any event, within the typical fabrication methods of common device manufacturing so that it can easily be incorporated on-chip, and to do so at a reasonable cost.”
OpenLight is not the only company looking at ways to improve the economics and reduce the complexity of silicon photonics production, Gold pointed out, as Intel has been increasing its focus on silicon photonics in recent years, and start-ups like Ayar Labs, which counts both Intel and Nvidia as investors, continue to emerge.
“Over the next two to three years, I expect silicon photonics interconnect to become common in high performance systems, and even start to make its way down the stack into medium and ultimately personal computing systems as the technology matures and the costs come down,” Gold said.