For those of you waiting with baited breath to see the Fabless community with the likes of Broadcom or Qualcomm challenge Intel in the data center space, you were just dealt a significant blow. Intel debuted their long awaited silicon photonics modules for what they called “lightning-fast” connectivity in data centers. Intel announced they are in volume production of their 100G PSM4 (Parallel Single Mode fiber 4-lane) and 100G CWDM4 (Coarse Wavelength Division Multiplexing 4-lane) transceiver products targeted for use in data communications applications, in particular switch-to-switch optical interconnects in data centers. Both of these products can communicate 100G signals up to two kilometers distance.
This is undoubtedly just the beginning but it is a significant beginning. While the holy grail is to eventually integrate this kind of functionality onto the processors themselves, this is a large step forward as Intel claims that for the first time, these transceivers integrate both electrical and optical components onto a single die, including a wafer-bonded hybrid III-V GaN laser. This technology, should make a for a lower cost and easier to make solution. The news was delivered as part of the keynoted address at the Chipzilla Developer’s Forum being held in San Francisco this week.
Earlier this year at the Optical Fiber Communications Conference (OFC), Robert Bloom, Director of Strategic Marketing and Business Development for the Silicon Photonics Products Division of Intel gave a briefing on Intel’s vision for the future of data center connectivity. In his presentation Robert showed the expected data center spending levels for 100G and 400G interconnects through the rest of the decade growing from $1.2B in 2016 to $5.1B in 2020. The growth is impressive and is not limited to just crossing the spine of the data center. By 2020 it is expected that these connection speeds will be used virtually everywhere, in the racks, across rows, across the data center and even between data centers. To do this the drive to integrate the photonics closer to the processors will pick up speed. According to Bloom’s presentation 400G is just around the corner with the transceivers moving from pluggable components as shown by Ms. Bryant to being embedded on the boards and eventually integrated with the switches and servers themselves.
It was also disclosed that Microsoft is an early adopter of Intel’s new technology and will be using it in their Azure data centers. Microsoft has also started testing field-programmable gate arrays (FPGAs) from Intel’s Altera business group in its data centers. Intel completed the acquisition of Altera in December of last 2015. This is yet another stake in the heart of the fabless guys entering this space. It makes you wonder how long it will be before Intel integrates the speed of the photonics with the flexibility of an FPGAs for Software Defined Networking capabilities. Now that would be a killer app.
The question still remains as to when the fabless guys are going to jump into the fray. Just last week Juniper Networks acquired Aurrion, a photonics IP provider, however one of the comments on that article claimed that it was a bit of a stretch to call this fabless in the photonics space as Juniper were most likely going to be getting silicon from a foundry source that wasn’t going to open its offerings up to the general public. None the less, the pressure is growing and with the move towards integration of photonics closer and closer to the processors, it will only be a matter of time before the chip volumes will reach a level where the fabless guys can’t afford to ignore photonics any longer.