The Age of Photonics
In a recent blog post summarizing the OIDA Executive Forum, I wrote a bit about how the telecommunications industry has moved from the TDM era to the WDM era to what can now be called the silicon photonics era. Silicon photonics is a very interesting topic that I’d like to revisit a bit.
Silicon is an amazing material for electronics. It is abundant, relatively inexpensive, and has electronic properties that make it ideal for making transistors. It forms a natural oxide (silicon dioxide – glass) that has insulating properties and is transparent to most wavelengths. The overwhelming majority of chip fabrication facilities worldwide are based on silicon and the technology is advancing to smaller and smaller feature sets that have more capabilities and use less power every year. Moore’s Law is based on the capabilities of silicon.
As optics become more complex with the advent of multi-level, multi-phase transmitters and coherent receivers, a lot of extra circuitry is now required to make the links work. The right solution for the digital signal processors (DSPs) that are a core part of modern receivers is to build them in silicon. Silicon is ideal for digital circuitry, there is a large base of fabrication facilities that understand the technology, and the latest innovations keep the power requirements low. Until relatively recently, however, the choice for the transmitter was less clear.
The problem with silicon on the optical front is that silicon cannot be used to make lasers. Semiconductor lasers are made from materials with a “direct bandgap”. Silicon has an “indirect bandgap” structure. Without getting too much into the details, a direct bandgap semiconductor can efficiently turn electrons into photons while an indirect bandgap semiconductor cannot. Since silicon is indirect and cannot be used to make lasers, any photonics work has to come up with some other way to generate light.
One approach is to make transmitter transistors and circuitry out of direct bandgap materials like gallium arsenide (GaAs) or indium phosphide (InP). For example, the folks at Infinera have placed their bet on InP-based photonic circuits. The issue with InP is one of cost. The material is hugely more expensive than silicon and the dearth of fabrication facilities that work in InP makes the price higher. Likewise, while there are extensive libraries and tools for creating circuitry out of silicon, these do not exist for InP. InP PICs are fascinating, and I applaud Infinera for championing the technology, but there might be a reason that no one else has chosen this path.
The path that is much more common today is to integrate direct gap semiconductor light sources with silicon circuitry. The silicon circuitry can be used both as the receiver DSP and as a way to manipulate the transmitted signal. My graduate research, many years ago, was focused on lifting off light sources and receivers from the substrate where they were grown and transferring them onto silicon circuits and micromachines. Today the integration is less process intensive and more automated. While still not ideal, the solutions are very good – as evidenced by the rapid evolution of silicon photonics solutions. And they are getting better every day.
With silicon as the engine inside of modern optics, a new world of capabilities is being opened. Optics can now start to take advantage of Moore’s Law as well as the enormous research that has already been done on programmability, power savings, and improved performance.
Several years ago, when 100G was first being introduced, predictions were made that the cost of 100G would limit it to only a few long distance applications. The rapid developments in silicon photonics blew that prediction out of the water. We have now deployed 200G and 400G solutions. 600G will be deployed by the end of the year. Laboratory experiments have passed the terabit mark. Modern silicon photonics solutions are not just higher speed, but are fully programmable so that the optimal line rate can be deployed based on measured fiber characteristics. The new DSPs are providing more feedback on network performance to allow better network maintenance and optimization. Silicon photonics, like almost every innovation, is moving from the expensive high end applications down the chain into more everyday applications. In a few years we might consider silicon photonics to be just as normal as 10G SFPs are today.
We are just at the start of the silicon photonics era, and there is a lot of room to grow. The innovations over the next years should be very exciting. Who knows what the next era will be – no one predicted this one – so let’s enjoy the start of this era before it, too, becomes old news.