- What is the problem that Luxtera is solving?
Communications links, or interconnects, are the biggest bottleneck in networks and computers.
Moore’s Law keeps on marching, driving the need for ever-faster networking. The next generation of ethernet runs at 100Gb/s and can be created by four lanes of 28Gb/s (28,000,000,000 bit per second) and at 28Gb/s electrical signals in copper wires can only travel a small distance before fading out completely. Some companies are trying to push the limits on this approach, but physics constrains the results to short distances over special wiring. In addition, these chips consume large amounts of power, a precious commodity in any computer or switch.
Optical fiber on the other hand is the ideal medium for communications over most distances. The fiber itself is very cheap, and light travels through it for miles even when launched with tiny amounts of power. Optical fiber also has the capability to carry data at rates up to hundreds of times faster than 28Gb/s. At each end of the fiber, an optical transmitter/receiver (“transceiver”) is required to interface to the computer or switch. Unfortunately, these optical transceivers currently are extremely expensive. As a result, optical fiber communication has been largely confined to the capital-intensive long distance telecommunications infrastructure.
To summarize: at 28Gb/s and beyond the established approach of using electricity in copper wiring just won’t work, and the ideal approach of using light in optical fiber is just simply too expensive. We believe that the industry is at a “tipping-point” where Luxtera’s disruptive technology can tip the balance from copper wiring to fiber optics for speeds of 28Gb/s and beyond.
- So how is Luxtera’s technology going to “tip” the data networking industry from copper to fiber?
Right now, optical transceivers are built using a combination of expensive technologies. The transmit laser chip and receive photo-detector chip are built out of Silicon Germanium, Gallium Arsenide or Indium Phosphide and packaged into individual high-precision cylindrical packages that slip over the ends of the fiber. Flexible circuits run from these cans to a printed circuit board where dozens of components are soldered down, interconnected by wires carrying data at very high speed signals. The result is a very costly product due to the expensive components, complex assembly and exhaustive testing required.
Luxtera’s CMOS Photonics™ technology is disruptive because it integrates complex optics together with complex electronics in a single CMOS chip (CMOS is the standard manufacturing process used by the silicon chip industry). We are a so-called “fabless” company that designs each chip, and then sends the design data to our development partner, Freescale Semiconductor (previously the semiconductor division of Motorola). Freescale builds the chip in the same CMOS process (SOI – Silicon On Insulator) they use for building high performance microprocessors. This is a powerful business model because it lets Luxtera take advantage of the many billions of dollars that are invested into CMOS manufacturing every year.
With CMOS Photonics™ we can build single-chip transceivers with price/performance far beyond that of conventional technologies. Not only will these transceivers be disruptive at the time they enter the market, but also CMOS Photonics™ gives us the ability to innovate rapidly in many dimensions through the continuous integration of additional functionality. This is simply impossible with existing optical technologies.
- Is it possible for Luxtera to build products that run faster than 10Gb/s?
Absolutely. It’s well known that a single optical fiber can carry many independent streams of data, each on different colors, or wavelengths. This technique is known as Dense Wavelength Division Multiplexing, or DWDM. Long distance telecom systems make extensive use of DWDM to pass many (for example 128) streams of 10Gb/s data on a single fiber. Existing DWDM systems occupy multiple racks of extremely expensive equipment. Luxtera has integrated the key elements required for implementing DWDM into a single CMOS chip. DWDM is not required for Luxtera’s 10Gb/s products, but will provide a seamless product path to100Gb/s ethernet and beyond.
Luxtera’s CMOS Photonics™ technology will scale with Moore’s law for electrical functionality and faster than Moore’s Law (through DWDM) for communications bandwidth. Why is the speed of 28Gb/s (Gigabits per second) important?
The current generation of ethernet, 10 gigabit ethernet, moves data at 10 billion bits per second (10Gb/s) over copper wires or optical transceivers. Broadband to the home is rapidly approaching 100Mbps and clouds services for the home and business are just beginning, which will significantly increase the data usage of smartphones and PCs. 100Gb Ethernet adoption at the edge of the network is projected to explode over the next few years.
- Why is the speed of 10Gb/s (Gigabits per second) important?
The current generation of ethernet, gigabit ethernet, moves data at 1 billion bits per second (1Gb/s) over copper wires. Gigabit ethernet ports already ship in the majority of all computers, and are creating a huge wave of demand for faster links to aggregate the data they produce. This effect occurs in every ethernet generation, driving demand for the next, which provides 10 times more bandwidth: in this case, 10Gb/s. Currently 10G ethernet is shipping only in very low volumes in very costly data switches; as prices fall it will begin shipping on servers – which are in themselves an 8 million unit/year market.
- Aren’t optical fibers expensive?
Not at all. Optical fibers are made out of silica, which is made out of sand. Wires are made out of copper, which after all is a metal sufficiently precious for us to use in coins. So fiber is actually very inexpensive and it is already made in huge amounts for long distance use.
- How do you generate light in silicon?
Actually, we don’t. Silicon is a bad material to try and build lasers in. Some silicon lasers have been demonstrated, but these are completely impractical. As it turns out there’s no need to build a silicon laser: lasers are already very inexpensive (remember, there’s already one in every PC – inside the CD/DVD player). The challenge has been finding an inexpensive way to attach the lasers to silicon. Solving this problem, and the related one of inexpensively attaching optical fibers to silicon, is a key piece of Luxtera’s intellectual property. We think of a laser as being just like a DC power supply – only it provides a steady stream of photons rather than electrons.
- This sounds like science fiction. When will you be able to build actual products?
We are in production today through our partnership with Molex, where they ship 40Gbp/s Active Optical Cables, recognized widely for their best in class power consumption, performance and reliability. This product has been a great pipe cleaner to validate the technology and to begin the ramp of very high volume (million units per year) type designs with silicon photonics.
Luxtera’s CMOS Photonics™ technology platform is real (not science fiction), high-performance, low-cost and scalable for decades to come.
- CMOS is all about building electronics. How do you build photonics in CMOS?
Previously, it was impossible to build photonics in CMOS, but Luxtera’s CMOS Photonics™ platform is built on a series of disruptive technical breakthroughs that make CMOS ideal for building photonic products.
Firstly, we should note that although silicon is opaque to visible light (you can’t see through a silicon wafer) it is in fact completely transparent to infrared light. Moreover, fortunately for us, infrared light is perfect for communicating through glass optical fibers. So, in a way, silicon chips and optical fiber were just meant to be together.
Secondly, right now the wires, transistors, and such in a CMOS chip are about 65 nanometers in width. In fact, you can think of a CMOS fab (or factory) as a very precise factory for building huge volumes of things at the nanoscale. Consider this: a single CMOS fab produces about a million billion working transistors a year – all of them around 65 nanometers in size.
Thirdly, 65 nanometers is much smaller than the wavelength of light. At Luxtera we’re designing and building silicon nanoscale structures that manipulate light in previously impossible ways. An example is our recently announced 28Gb/s optical modulator. This device can turn a light beam on or off 28 billion times a second under electrical control – but only because of the very small size of the channels, or waveguides, that we use to guide the light inside the modulator.
There are two other great things about working in CMOS. Firstly, we can build complex electrical circuits right alongside our photonic circuits. This lets us monitor the performance of a photonic circuit in an intelligent manner, and continually “tune” it for optimum performance independent of manufacturing and temperature variations. Secondly, the silicon industry continually drives line-widths downwards – soon chips will be manufactured at 28 and 22 nanometers. This works well for us because as line-widths get smaller, CMOS Photonics™ devices get even smaller, faster, and more efficient.