Periodic Reporting for period 4 - PhotonICSWARM (Photonic Integrated Circuits using Scattered Waveguide elements in an Adaptive, Reconfigurable Mesh.)
Período documentado: 2021-10-01 hasta 2022-09-30
In PhotonicSWARM, we are building general-purpose photonic chips, where light can be controlled in software. Such chips can then be used for a diversity of functions, from fiber-optic communications over sensors to radio-frequency filters for 5G networks.
Photonic integrated circuits, or PICs, manipulate light on the surface of a chip through waveguides. Today, PICs are mostly used in optical communication, either in long-haul fiber networks or to connect servers inside a datacenter. But new applications are being developed that use the same technology to create extremely accurate sensors, spectrometers, and processors for high-speed microwave signals. Photonic chips are also one of the key technologies to develop artificial neural networks or optical quantum computers. However, the adoption of this powerful technology for new applications is extremely slow.
Today, photonic chip technology is increasingly making use of the same technology to make electronic circuits: silicon photonic makes it possible to pack thousands of optical building blocks together on a chip, and make these chips in very large volumes. This lowers the barriers for manufacturing, but of course this only makes sense if there is a market for large volumes of chips. Today, the need for photonic chips is orders of magnitude lower than electronic chips, and this makes it the process to develop a new photonic chip product very costly and time-consuming (millions of Euros and > 5 years). This long development cycle puts a major break on innovation.
A key reason for this long cycle is that today, photonic chips are always developed for a particular purpose; they are so-called application-specific photonic integrated circuits (ASPIC). To developing a new ASPIC, you need to design, simulate and fabricate a chip, but these processes are not yet mature enough to guarantee first-time-right designs (as in electronics). Multiple iterations are needed, taking up a lot of time and money.
The programmable PICs developed in PhotonicSWARM should significantly change this. Instead of an application-specific chip that can perform only one function, a generic programmable PIC is not designed can be reconfigured on the fly to using control electronics and software. This way, they resemble programmable electronics, such as microprocessors, digital signal processors (DSP) or field-programmable gate arrays (FPGA). Today, such electronics can be purchased off-the-shelf and programmed directly, shortening the development time from more than a year to weeks. PhotonicSWARM has been laying the foundations for a similar ecosystem for photonics that can open up the technological field of photonic chips for innovators and the maker community.
Photonic integrated circuits, or PICs, manipulate light on the surface of a chip through waveguides. Today, PICs are mostly used in optical communication, either in long-haul fiber networks or to connect servers inside a datacenter. But new applications are being developed that use the same technology to create extremely accurate sensors, spectrometers, and processors for high-speed microwave signals. Photonic chips are also one of the key technologies to develop artificial neural networks or optical quantum computers. However, the adoption of this powerful technology for new applications is extremely slow.
Today, photonic chip technology is increasingly making use of the same technology to make electronic circuits: silicon photonic makes it possible to pack thousands of optical building blocks together on a chip, and make these chips in very large volumes. This lowers the barriers for manufacturing, but of course this only makes sense if there is a market for large volumes of chips. Today, the need for photonic chips is orders of magnitude lower than electronic chips, and this makes it the process to develop a new photonic chip product very costly and time-consuming (millions of Euros and > 5 years). This long development cycle puts a major break on innovation.
A key reason for this long cycle is that today, photonic chips are always developed for a particular purpose; they are so-called application-specific photonic integrated circuits (ASPIC). To developing a new ASPIC, you need to design, simulate and fabricate a chip, but these processes are not yet mature enough to guarantee first-time-right designs (as in electronics). Multiple iterations are needed, taking up a lot of time and money.
The programmable PICs developed in PhotonicSWARM should significantly change this. Instead of an application-specific chip that can perform only one function, a generic programmable PIC is not designed can be reconfigured on the fly to using control electronics and software. This way, they resemble programmable electronics, such as microprocessors, digital signal processors (DSP) or field-programmable gate arrays (FPGA). Today, such electronics can be purchased off-the-shelf and programmed directly, shortening the development time from more than a year to weeks. PhotonicSWARM has been laying the foundations for a similar ecosystem for photonics that can open up the technological field of photonic chips for innovators and the maker community.
Programmable PICs consists of a mesh of optical waveguides which are coupled together using waveguide couplers that can be electrically controlled. Our initial demonstration, the first programmable PIC in silicon photonics, was a simple 4x4 beam coupler that mapped 4 inputs onto 4 outputs in any arbitrary combination. Such ‘forward-only’ circuits are useful in applications that require matrix operations, such as neural networks and quantum computing. But they are limited when it comes to more advanced functions such as optical wavelength filters.
Therefore we switched to ‘recirculating’ meshes, where the light is routed in loops. We studied different connectivity schemes for such meshes, and how accurately they need to be controlled to avoid unwanted parasitic waveguide paths. Because all elements on the photonic chip need controlled, we developed an electronics layer based on commercial FPGAs that allows us to digitally control a large number of the on-chip tuners.
We built multiple iterations of these new programmable chips, using different waveguide connectivity, actuation mechanisms (heaters, MEMS, liquid crystals) and control schemes. However, not everything worked right out of the box. While these chips are conceived to shorten development time and eventually bypass the need for ASPICs, their own design, fabrication and testing still goes through the traditional slow cycle, with the risk that the chip does not work as expected after it returns from the fab. As these are very complex circuits, we experienced different failure mechanisms on multiple prototypes. In the first 3 generations we encountered a combination of optical, electrical and thermal issues that limited the functionality of the chip, and a 4th-generation chip is still in fabrication.
We managed to demonstrate different core technologies for these programmable photonic circuits, ranging from low-power actuators, tolerant building blocks and subcircuits (e.g. for the tunable coupler gate that is used throughout our circuit), and control routines to calibrate and configure routing and filtering functions in the waveguide mesh. This also included the use of high-speed modulators, where we added programmability to make them reconfigurable and optimize their characteristics, so they can be used to process high-speed microwave signals. We demonstrated these smaller results as part of dedicated application-specific circuits, but their use in a fully-programmable photonic circuit still needs to be demonstrated.
Even though the hardware side is not yet fully functional, we have developed a software framework to design, simulate, control and configure programmable photonic circuits. This framework already allows us to experiment with new algorithms to configure and control the circuit, such as powerful graph-based routing or filter synthesis routines. The simulation tools also allow new researchers and developers to already assess if programmable photonic circuits could be suitable for their applications.
We have published our results in multiple peer-reviewed journal publications and at scientific conferences, and our introductory videos on Youtube have proven very popular. We have also looked into different application spaces where these programmable photonic chips could bring the most value, and we are now exploring avenues to bring this technology to the market in the ERC proof-of-concept grant LIQUORICE.
Therefore we switched to ‘recirculating’ meshes, where the light is routed in loops. We studied different connectivity schemes for such meshes, and how accurately they need to be controlled to avoid unwanted parasitic waveguide paths. Because all elements on the photonic chip need controlled, we developed an electronics layer based on commercial FPGAs that allows us to digitally control a large number of the on-chip tuners.
We built multiple iterations of these new programmable chips, using different waveguide connectivity, actuation mechanisms (heaters, MEMS, liquid crystals) and control schemes. However, not everything worked right out of the box. While these chips are conceived to shorten development time and eventually bypass the need for ASPICs, their own design, fabrication and testing still goes through the traditional slow cycle, with the risk that the chip does not work as expected after it returns from the fab. As these are very complex circuits, we experienced different failure mechanisms on multiple prototypes. In the first 3 generations we encountered a combination of optical, electrical and thermal issues that limited the functionality of the chip, and a 4th-generation chip is still in fabrication.
We managed to demonstrate different core technologies for these programmable photonic circuits, ranging from low-power actuators, tolerant building blocks and subcircuits (e.g. for the tunable coupler gate that is used throughout our circuit), and control routines to calibrate and configure routing and filtering functions in the waveguide mesh. This also included the use of high-speed modulators, where we added programmability to make them reconfigurable and optimize their characteristics, so they can be used to process high-speed microwave signals. We demonstrated these smaller results as part of dedicated application-specific circuits, but their use in a fully-programmable photonic circuit still needs to be demonstrated.
Even though the hardware side is not yet fully functional, we have developed a software framework to design, simulate, control and configure programmable photonic circuits. This framework already allows us to experiment with new algorithms to configure and control the circuit, such as powerful graph-based routing or filter synthesis routines. The simulation tools also allow new researchers and developers to already assess if programmable photonic circuits could be suitable for their applications.
We have published our results in multiple peer-reviewed journal publications and at scientific conferences, and our introductory videos on Youtube have proven very popular. We have also looked into different application spaces where these programmable photonic chips could bring the most value, and we are now exploring avenues to bring this technology to the market in the ERC proof-of-concept grant LIQUORICE.
With the second-generation chip, we intend to show that it is indeed possible to program such a circuit for multiple applications: it can function as a spectrometer or switch, and with its built-in modulators and detectors it can function as an optical transmitter or receiver circuit, or even a microwave signal processor.
In parallel, we are looking to alternative architectures and tuning mechanisms that will allow us to eliminate the heaters we currently use for the tunable couplers and phase shifters. These are very power-hungry. One promising alternative approach is the integration of liquid crystals on the surface of the silicon chip.
Even when we are only starting to use our second-generation chip, we are already planning the design and fabrication of the next generation chip, where we intend to scale up the complexity and look into new circuit architectures. These architectures implemented in silicon photonics, with power-efficient tuners, electronic drivers, control algorithms and programming strategies, form complete technology stack. This way, PhotonicSWARM is laying the foundation for an ecosystem that will enable a future with off-the-shelf multifunctional programmable PICs.
In parallel, we are looking to alternative architectures and tuning mechanisms that will allow us to eliminate the heaters we currently use for the tunable couplers and phase shifters. These are very power-hungry. One promising alternative approach is the integration of liquid crystals on the surface of the silicon chip.
Even when we are only starting to use our second-generation chip, we are already planning the design and fabrication of the next generation chip, where we intend to scale up the complexity and look into new circuit architectures. These architectures implemented in silicon photonics, with power-efficient tuners, electronic drivers, control algorithms and programming strategies, form complete technology stack. This way, PhotonicSWARM is laying the foundation for an ecosystem that will enable a future with off-the-shelf multifunctional programmable PICs.