LONDON – STMicroelectronics NV has said it agreed with CMOS photonics specialist Luxtera Inc. to develop a dedicated silicon photonics process at its 300-mm research and pilot production wafer fab in Crolles, France. Production at Crolles would then enable the two companies to provide silicon photonics components and systems.
The deal is intended to take silicon photonics mainstream and allow the integration of silicon photonic technology with system-on-chip (SoC) ICs.
Luxtera (Carlsbad, Calif.) was a 2001 fabless semiconductor spin off from Caltech that builds complex electro-optical systems in mainstream CMOS processes. The technology has yet to find wide deployment but is expected to be of use in high-speed computing, CPU interconnect, data-storage and on-chip clock distribution.
Under the terms of the deal ST gains rights to use Luxtera's silicon photonics technology that will be implemented in an ST photonics process and subsequent generations of photonics processes. ST will provide Luxtera with silicon as a foundry supplier as well as being able to manufacture in its own name.
The silicon photonics process will offer scalability of electro-optical transceivers for data rates of 100-Gbits per second, 400-Gbits per second and beyond. It will support light at wavelengths of 1310-nm, 1490-nm and 1550-nm.
ST did not indicate how quickly the process would be developed or what the critical dimension capabilities would be. Nor did ST say when it would be able to manufacture photonic chips using the process for itself and for Luxtera.
"This will bring silicon photonics into the mainstream of important technologies such as optical networking, ultra-fast computer processors and other applications via the commercial volume availability of a best-in-class silicon photonics IP platform," said Flavio Benetti, general manager of mixed process division at STMicroelectronics, in a statement.
In the same statement Greg Young, president and CEO of Luxtera, said: "We can now offer our customers a high-volume, capable source of supply and an aggressive long-term photonic process technology roadmap. This will advance our base technology and enable the integration of optical transceivers with SoCs from advanced CMOS nodes to deliver photonic-enabled SoCs for large scale systems."
I believe that Intel was successful in making Light Peak work from a technical standpoint - just not as cheaply as they could implement Thunderbolt. And I guess their customers (Apple?) decided that Thunderbolt was good enough for now. Perhaps Light Peak will supplant Thunderbolt in the next generation interface in another few years.
Good point @resistion...E-O-E conversion will be always required...if you have to send 10 Gb/s over 1km the power dissipation of the photonic solution is lower, at 100m comparable and at 10m higher than that of the purely electronic one...so there is ways to go before that technology can be used at the chip level...but right now there are many repeaters and clock trees that are used on-chip that could be displaced with something arguably simpler in the future, optical clock tree will be very simple...Kris
theres so much to say on this nano photonics subject it fascinates me :)
but its hard to cut through the crud and pull out the gems in general reporting not least when we will actually see a group of products you can buy and play with.
i prefer to look at the core device or process they are describing and think how one might put that and related findings to practical use to make an exact copy of what's already available in the normal scale only nano scale in this case.
i dont want to see this simply as "smarter wire" or so different that people think it needs a totally different wat of doing things, for instance the "omparing light wavelength with MOSFET channel size" im not quite sure what that means but this pic overcomes that i think ?
see http://www.spacemart.com/nanotech.html for many other things they can do and how you might use them to make a 3d stacked SOC etc
"Nano-coating doubles rate of heat transfer" seems obvious,out that and a nano thermoelectric generator in layers to both cool and reclaim power to supplement local power nano ram.
"weld nanowires with light"
and lots more
"Metal nanoparticles shine with customizable color" for laser.
"Light-emitting nanocrystal diodes go ultraviolet"
i think in interesting as the wavelengths of 1310-nm, 1490-nm and 1550-nm above imply they are ONLY thinking generic Wavelengths Fiber Optics in the infrared range over glass not the lower UV range Plastic optical fiber (POF)at 650 nm.
odd in so far as nano plastics and so called bio chips are available today, so its perfectly possible to combine these and the other nano processes into one assembly and see the damn things ASAP LOL, did i say there's lots to say :)
I guess we were on different pages @me...I am slowly understanding what you are saying...you are comparing light wavelength with MOSFET channel size...yes, photon is larger in that sense, and it leads to a statement that photonic devices will be probably always larger then electronic one...so yes photodiode on silicon will be always larger than transistor...but the signal from laser or modulator going to a photodiode will be send much faster than from transistor to transistor (if they are far away enough) and that is a main point of photonic interconnects on chip...Kris
I am pretty sure that photon is smaller than silicon chip @me (assuming you can isolate one and measure how large it is which is rather impossible according to quantum physics)....but putting size deliberations aside the discussion should be around a point "how long the interconnect has to be in order for optical transmission to make sense?"...that number used to be several km (long haul optical fibers) and it is now several cm (optical back-planes)...so likely it will become several mm in the future at which point application on chips might make sense...Kris
I believe that Intel continues to invest heavily in this technology, this is the future, photons are faster than electrons...the only debate is when it will be deployed commercially not whether it will be deployed...Kris
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.