An optical switch will route an incoming optical signal from an input to a destination output without converting the light to electrical signal. It has the potential to enable ultra-high capacity optical networks that can deliver large volumes almost at the speed of light. An optical switch can be significantly miniaturized using silicon photonics, which is based on the same technology used to make computers chips. Fast, nanoscale silicon photonic optical switches can also enable emerging applications in computing architectures such as optically interconnected processors and memories.
So far the bulk of the research efforts on silicon photonic switches have been focused on integrating as many devices on a single chip. Instead of putting the whole switch on a single chip, our work proposes, to divide it over multiple chips. Like computers have moved from a single processor to multi-core architectures our switch uses multiple chips to guide the light to the desired output. We demonstrated a proof-of-concept for 8 inputs - 8 outputs switch fabric by assembling multiple photonic chips offering multiplexing functionalities. The proposed switch can guide the light from any input to any output, can scale to a large number of inputs and outputs, it is straightforward to control, and is compatible with the standard silicon chips manufacturing process.
This modular or “divide and conquer” approach for photonic integrated circuits offers the advantage of piece-wise manufacturing, packaging, assembly, and fewer electrical controls. It paves the way to create large scale systems from standard chips, which can be stacked like LEGO blocks to create 2.5 dimensional photonic systems.
More information : D. Nikolova, et. al., "Modular architecture for fully non-blocking silicon photonic switch fabric", Nature Microsystems & Nanoengineering 3, 16071 (2017), doi:10.1038/micronano.2016.71
High radix, low latency, energy efficient optical switches are essential to sustain the rapid growth in aggregate bandwidth requirements in data centers and high performance computers. Higher port count switches not only permit to interconnect more hosts directly but they also allow more compact topologies involving less links and switches to interconnect large networks. Starting with detailed device models which I have developed and experimentally verified, I analyzed the scalability of silicon photonic microring based switch fabrics with the widely used Benes topology. I was able to estimate the crosstalk and losses of a single chip switch fabric. This allowed me to determine the achievable port count and bandwidth density for these switch fabrics. The surprising conclusion of this was work was that the on-chip waveguide crossings contribute significantly to the scaling of a two dimensional switch fabric
2D switch fabric from double microring switching elements with Benes topology ; top left inset shows picture from a fabricated device; top right - schematics of the operation of the switching element; right inset - picture of a on-chip waveguide crossing; bottom - miltiple crossings model.
The silicon photonic technology is still young and only recently emerging for commercial systems adoption. Optical interconnects based on silicon photonic devices have the potential to deliver huge amount of bandwidth at low energy and cost. Understanding the capacity limits of these interconnect is of vast importance. In this paper, we present a design and modeling approach for obtaining the maximum achievable aggregate bandwidth in a silicon photonic link with microring modulators and filters. It is based on a uniquely comprehensive modeling platform for efficiently exploring the design space of silicon photonic interconnects, from the physical layer to the link-level analysis. We concentrate our efforts on microring-resonator-based links, as they have the lowest footprint among the current most developed silicon photonic interconnect devices. Furthermore we derived the optimal device parameters. We obtained an upper bound on the maximum aggregate throughput achievable with a microring-based silicon photonic link. This is of paramount importance for link designers, informs device designer for optimal parameters and also can allow comparison with other existing and future devices.
Vision for optically interconnected racks enabled by silicon photonic microring modulators and filters
When hundred thousands of flows are competing for the same switch output, scalable, low complexity scheduling algorithm are needed to ensure low latency and fairness. In wire-line access networks where the infrastructure is shared by relatively low number of users achieving high utilization but at the same time guaranteeing stringent quality of service has led to the use of multiple channels over the same cable/fiber. I proposed the input and output queuing concepts for scheduling a single user transmission simultaneously over multiple channels. I also analyzed and implemented the bonded deficit round robin algorithm which realizes these concept and demonstrated that it can guarantee strict quality of service requirements even when the channels use varying bit rates.
Input queuing where the packets are buffered in common queue and scheduled for transmission owhen a channel becomes free
The realization of silicon photonic devices is made possible due to the confinement of light in nanometer structures. Another example of tight confinement of the light field are surface plasmons-polaritons which are coupled light-electron plasma waves, propagating at the interface between a dielectric and conducting material. They are extremely sensitive to the dielectric properties of the materials which makes them very interesting also for sensing applications and have given rise to the field of plasmonics. The combination of plasmonics with magneto-optical materials is particularly interesting because it introduces a nanoscale interaction between light felds and magnetisation, hence opening up the possibility of using either one of these fields to control the other. My particular contribution in this area was to derive the influence of magnetization on the light propagation in plasmonic waveguides. I was able to demonstrate how the magnetic field can be used to switch dipole emission on and off and to spatially direct the light out of a plasmonic cavity. This raises the novel possibility of using magnetic fields to control light propagation in nanostructures and using light to sense the magnetic properties in nanoscale domains.
Magnetic switching of the dipole emission in an 80 nm thick cavity of ferromagnetic dielectric surrounded by silver layer.
I am a scientist and an engineer with wide interests in photonics, optical enineering, deep learning and biology.