Scaling up production of photonic integrated circuits

LightCounting publishes a research note from PIC International

Photonic integration promised to bring the economies of scale found in the microelectronics industry to optical components. The integration story has turned out rather differently, as LightCounting previously reported. Early photonic integrated circuit (PIC) products tended to be more complex, high-end pieces produced in low volumes, where integration enabled new functionalities that would have been difficult to implement using discrete optics.

Mass production and low, low costs remains the ambition for integrated photonics, but how can the industry achieve this goal? That question was a common theme across the presentations at PIC International conference, held in Brussels on 10-11 April.

Data centers are undergoing a massive transformation, where the links inside buildings are becoming longer and multiple buildings up to 100km apart are being connected. This has created an important opportunity for silicon photonics. Currently more than 40% of Microsoft’s intra data center links use 100G direct-detect technology based on silicon photonics rather than indium phosphide, according to Radha Nagarajan, CTO of InPhi’s Optical Interconnect business. Microsoft is technology agnostic; silicon photonics offered the right performance at the right price point, he said. More applications like this will emerge over time; this is the basis for LightCounting’s forecast for steady market growth. The latest forecast for integrated optics, including silicon photonics, will be released by LightCounting in the end of May 2018 along with the report titled “Integrated Optical Devices”.

But what about the bigger picture? At the wafer level, the industry has sufficient manufacturing capacity to meet the world’s demand for bandwidth many times over, as Martin Schell, executive director of Fraunhofer HHI explained, in an entertaining presentation. The world has entered what Cisco dubbed “the zettabyte era”, where global IP traffic exceeds one zettabyte annually. A zettabyte is one trillion terabytes or 10 to the power of 21 bytes. It sounds like one heck of a lot of data, until you realize that one indium phosphide wafer could contain as many as 20,000 laser chips; if all those chips were made into modules, a single wafer creates the ability to transmit 3 zettabytes of data per year.

The calculation needs adjusting as each byte of IP data traffic transmitted to end users typically generates 100 times as much traffic inside and between data centers. However, this is offset by the fact that a typically foundry processes many thousands of wafers per year, not just one. “Even if the whole world buys optochips from us, [Fraunhofer HHI] will not know what to do in the afternoon, and we probably have the smallest fab in the world,” Schell joked. From just looking at these numbers, it’s difficult to understand why we have a scaling problem in this industry.

“The nature of the industry itself is the problem”, as he explained. “What the industry needs is volume, but volume today is divided by many applications. Each company might produce between 3,000 and 120,000 pieces per product, per year. The more applications or products there are, the less volume per application, and the more challenging it is to scale production. The volumes are simply too small to justify putting large resources into faster development, to develop and refine specialist processes and equipment.”

It’s a chicken and egg situation. Vendors can’t successfully scale photonic integration manufacturing to reduce costs, because they aren’t making products in sufficient volume; and they can’t break down price barriers to increase demand for their products before they have scaled up the process. How can this deadlock be broken?

Currently, this issue is being addressed through collaboration, or as one attendee put it, “making the pie bigger for everyone, rather than fighting over our share of the pie”. Multi-project wafer runs make it easier for companies to prototype devices by sharing space on a wafer. Then the generic foundry model allows vendors to share manufacturing capacity on the same equipment. There’s an unprecedented amount of activity in this area, particularly in Europe. The American Institute for Advanced Manufacturing Photonic Integration (AIM Photonics) is the US response to the need for easier access to photonics manufacturing. Now with more than 80 members, it is on track for a grand opening of its new facilities this summer.

In the meantime, however, optical companies blazing the silicon photonics trail, such as Intel and Luxtera, have chosen to “fill the fab” by diversifying their own product offerings, rather than partnering with other vendors. Both of these companies are already shipping a lot more than 120,000 pieces per product, per year. LightCounting’s upcoming report will not include product shipments by vendor in order to protect confidential sales data, but it will offer detailed information on the total shipments for more than 100 products, including historical data for 2010-2017 and forecast for 2018-2023.

LightCounting subscribers can log into their account for access to the full research note:





LightCounting is a leading optical communications market research company, offering semiannual market updates, forecasts, and state-of-the-industry reports based on its analysis of primary research with dozens of leading module, component, and system vendors as well as service providers and other users. LightCounting is the optical communications market’s first choice source for the accurate, detailed, and relevant information necessary for doing business in today’s highly competitive environment. For more information, visit: or follow us on Twitter at @LightCounting.