Home

Over the past decades, the computational performance of the computing systems has improved constantly following the Moore's law. Such advance has enabled the growth of datacom applications that span from the storage and data retrieval capabilities of the data centers, to the high performance computing capabilities of the supercomputers. Data centers and supercomputers are computing platforms realized by interconnecting a large number of computing systems (servers or even commodities computers) with an interconnection network. The performance improvement of such computing platforms is now jeopardized by the interconnection networks, currently based on electronics. In particular, bandwidth, wiring density, and power consumption of the electronic interconnection networks are bottlenecks to the further scaling of speed and capacity of the high performance computing platforms.

To counteract these problems, the introduction of optics in the interconnection networks has the potential to scale to higher speed, reduce the power consumption, and increase the reliability, as demonstrated by several projects (including one led by a research partner of this project). Two main challenges are still open impairing the exploitation of the potential offered by photonics: the first issue concerns the possibility to integrate the optical interconnection network on the same silicon substrate that hosts the microelectronic circuitry, the second issue concerns the capability to achieve the high speed required for a fast switching of packets.

The MINOS project aims at exploring the recently discovered stress-induced Pockels effect in strained Silicon for realizing fast and CMOS-compatible optical switching devices, suitable for interconnection networks. Switching elements, such as microring resonators, can be realized on a standard Silicon On Insulator (SOI) substrate by depositing a film of silicon nitride (Si3N4, or SiNx), which induces a tensile stress in silicon lattice. Thus, through the strain-induced electro-optic (Pockels) effect, an electric field (with no flowing current) can be applied to rapidly change the tuning frequency of the microring resonator. This can lead to a significant reduction of the switching time, and the achievement of a breakthrough performance improvement in terms of speed, tunability, and power consumption with respect to currently power-hungry available photonic switching technologies, such as thermo-optic effect (only allowing for slow switching capabilities), or carrier-injection techniques (suffering from very high free-carrier-induced losses and allowing only a limited wavelength tunability). The proposed technology can be easily integrated in a CMOS process - the standard process for microelectronic industry - allowing for higher integration, seamless compatibility with electronic circuitry, lower cost per component and lower power consumption (a negligible current flow is required in Pockels-effect-based switching).

The final objective is to enable the switching of optical packets from input to the output ports of the interconnection network with nanosecond switching time and low power consumption, as required by computing platforms for datacom applications.

Realizing ultrafast optical switches based on microring resonators is a challenging target that can be achieved by addressing five important sub-objectives:

  1. the development of an accurate model to describe the anisotropy induced by the Si3N4 layer and the relative stress-induced Pockels effect,
  2. thorough characterization of the novel material
  3. the definition of the strained-silicon microring fabrication process on SOI platform,
  4. study and experimental realization of photonic devices based on strained silicon,
  5. the design and the simulative and experimental performance assessment of the multiplane optical interconnection networks realized with the above mentioned devices.

The main expected results of MINOS project are:

  • a thorough analysis of the potentialities of the strained Si microrings when used in traditional space switching architectures;
  • novel modeling schemes and numerical tools enabling the study and design of strain-silicon devices exhibiting photoelastic and electro-optic effect;
  • manufacturing of innovative space switching devices with a low energy consumption, and nanosecond switching capabilities;
  • investigation of novel architectures for space switching optimized for the characteristics of the strained Si microring resonators;
  • multiplane architectures able to connect the designed space switch with cards exploiting other switching domains, to offer higher scalability and modularity, as required in computing platforms.

The outcomes of the projects will offer tremendous opportunities for the advancement of silicon photonics and datacom field. The project will be able to reduce the size of the optical interconnection network from a benchmark table to a single chip, and will allow to develop a new technological paradigm for microring devices exploiting the induced electro-optic effect in strained silicon. The integration of the optical devices into a single chip with a CMOS compatible technology will be a significant leap ahead for integration of the optical interconnection networks directly with the semiconductor board of the computing platform.