Optical Fiber Communications: the State of the Industry and
Predictions for the Future of Silicon Photonics

Copper out, Silicon in? AECs and DACs are running out of steam

We hear a lot of discussion about AECs (Active Electrical Cables) and DACs (Direct Attached Cables), and whether they are running out of steam. DACs have the advantage of being low in power, but increasingly the reach is becoming limited, especially as the industry looks forward to 200Gb/s per channel where many DACs may no longer be able to service the traditional server-switch application. Power consumption, cable bend-radiuses and weight of the copper cables are increasing limitations of AECs, and this creates a potential opportunity for AOCs (Active Optical Cables).

AOCs are lighter and offer superior flexibility and aren’t vulnerable to electromagnetic interference. In addition, AOCs (especially Silicon Photonics-based AOCs) also have virtually no reach limitation in the data center application. For these and many other reasons, AOCs look increasingly poised to become the preferred alternative.

AOC based on silicon photonics or VCSELs?

VCSELs have long been the preferred technology for AOCs, due to the power and cost structure they offer. But they’re now almost a generation behind equivalent single-mode solutions. For example, 100G VCSELs are just entering production now, and we are not likely to see 200G per channel VCSEL AOCs until 2025 or 2026.

VCSELs are also increasingly looking less attractive from the point of view of cost efficiency. By the end of 2023, it is expected that AOCs using silicon photonics will be available that run at 200G per lane. This means that an 800G AOC will be available that supports 4 transmit lanes and 4 receive lanes, running off of a single laser and single PIC chip. This will require 8 single-mode fibers (SMF) in total.

In contrast, an equivalent VCSEL solution in this timeframe will require 8 transmit, and 8 receive lanes running at 100G. The implication is that this solution will require 8 VCSELs (lasers) and 16 total fibers. This cost advantage is further compounded by the fact that silicon photonics use more affordable SMF fiber, rather than MMF (and fewer fibers).

In addition, the supply chain for VCSELs is increasingly precarious. There are very few vendors who can make VCSELs at these high speeds, and our checks have confirmed that the costs of the high-speed VCSEL arrays today are very high due to poor yields and insufficient competition.

Insights from thought leaders

We have also been able to participate in many conversations around a new proposed linear interface from the host ASIC to the optical module, eliminating the existing retimed/DSP paradigm that is common in high-speed optical modules today.

Lightcounting hosted an in-depth webinar on this topic on March 21, 2023, where a number of industry thought leaders discussed this interface. Several presentations showed that silicon photonics modules with linear drive can result in a minimum of 25% power savings at system level and up to 50% at transceiver level.

Coupled with silicon photonics, this linear interface can be a cost-effective alternative to AECs. We expect that there will be an ongoing debate in the industry as to whether a full linear interface makes sense, or if a single retimer/DSP can be used in the module (for example on the transmit side as opposed to both transmit and receive which is the norm today). Either way, we believe this trend will reduce power and cost inside the module, and Dustphotonics plans to be a key participant in this trend. Furthermore, we believe the AOC is the most straightforward starting point for the linear interface, especially given the shift away from AECs and DACs that we mentioned earlier.