Silicon photonics has made considerable progress in a relatively short time to emerge as an important systems technology whose time has come.
Just over a decade ago the likes of Intel and IBM were announcing performance records for the basic silicon photonics building blocks — modulators and detectors — used to make optical devices. Now, companies are shipping complex silicon photonics-based integrated circuits as part of their products.
When we started writing our book in late 2014, the frenzied excitement that first greeted silicon photonics had been replaced with an industry pragmatism. Companies realized the scale of the challenges to be overcome to bring the technology to market.
These challenges are not just technical issues but also business ones such as identifying markets, driving down cost and identifying where the technology will have an edge. Silicon photonics is competing with indium phosphide and gallium arsenide, mature technologies that already serve the optical component industry.
We finished the book in late 2016, a good year for silicon photonics. Equipment makers Ciena and Juniper Networks made acquisitions to bring silicon photonics know-how in-house. Ciena bought the silicon photonics arm of Teraxion for $32 million, while Juniper acquired start-up Aurrion for $165 million.
Also last year Acacia Communications, a maker of coherent transceivers based on a silicon photonics IC for long-distance transmission, had a successful IPO. Intel announced its first 100-gigabit transceivers after a decade-plus of lab work. Meanwhile, startups Rockley Photonics, Ayar Labs and Sicoya became more vocal as they moved closer to launching first products. And just last month privately owned company, Elenion Technologies, announced itself. The company has been active for over two years and is already shipping optical engines.
Chip designers need to concern themselves with silicon photonics. High-end chip developers will rightly argue that they are making good progress with their next-generation IC designs without needing silicon photonics. But this situation may not be true much longer.
As Moore’s law comes to an end, scaling designs and systems becomes more challenging. Silicon photonics promises to be a key technology to enable chips to continue scaling. This can be in the form of optical interconnect to get data on and off a chip and as a powerful way to interconnect die in more complex 2.5D and 3D packaged systems.
In 2016, Broadcom announced its Tomahawk II, a 6.4Tbit/second switch chip. The chip has 256 25Gbps serdes. By 2020 two more generations of switch chips are expected, supporting first 12.8Tbits/s and then 25.6Tbits/s capacities. The 12.8Tbits/s can use PAM-4 modulation to enable 50Gbits/s serdes, but at 25.6Tbits/s optical interfaces likely will be required if data is to be moved on and off the chip and across the board. This will represent one key inflection point for silicon photonics.
For now, chip and optical component design are distinct cultures. But the semiconductor and photonics worlds are merging, and once they do change will be rapid. The chip industry will start driving silicon photonics.
The Telecom Infra Project is an example of the two industries merging. TIP is an industry initiative that includes Facebook and ten telecom operators. At its first summit last November the Voyager packet-optical platform was announced. The one rack-unit white box includes a Broadcom Tomahawk 3.2Tbits/s switch chip and two Acacia 400-gigabit coherent optical transceivers.
With the Tomahawk IC and coherent optical transceivers, Voyager is a telling example of a merger between the datacom and telecom worlds. But there is a more subtle development here: the Voyager box shows how optics and an ULSI switch chip are coming together. As mentioned, future switch chips will be early to embrace silicon photonics. Voyager is clearly one platform set for future electronic-optical integration.