< Back to previous page

Project

Thermal Management of Integrated CMOS-Si Photonics Optical Transceivers

The infrastructure behind the internet are data centres, where data are stored and processed centrally. Transporting large amounts of data inside the data centre requires a significant amount of energy. To improve the energy efficiency of data transport, the classical electrical connections are being replaced with optical fibres. Traditional electrical input-output (I/O) faces multiple bottlenecks: the bandwidth-distance-power trade-off and chip pin count. Silicon photonics, the technology to manufacture integrated photonic circuits using CMOS process technology, promises to alleviate these bottlenecks and provide low propagation loss, high bandwidth optical I/O links. 

Recent advances in packaging technologies allow the tight integration of Si photonics transceiver chips with the host IC in a single chip package. These co-packaged optics enable the high bandwidth required for future data centres as well as the reduction of the package footprint. The Si photonic elements are however significantly affected by changes in local and ambient temperature, e.g. caused by the highly non-uniform power generation in the ASIC or FPGA CMOS chip, the laser sources, and the optical devices themselves. Temperature changes can cause a wavelength shift which leads to an optical power loss due to wavelength mismatch and to laser power degradation, which introduces more challenges for the thermal management of these co-packaged optics. 

In the first part, we study the thermal behaviour of photonic devices used in optical transceivers. Ring-based devices (i.e. ring- modulators and filters) rely on accurate spectral alignment with the laser source. To achieve wavelength locking, the waveguide temperature is controlled with integrated heaters. Through thermo-optical modelling, strategies for improving the heater efficiency are proposed, as well as methods for improving the electromigration lifetime of the heaters. We also study the thermo-optical stability of electro-absorption modulators and ring modulators at high optical power, and propose methods for minimising self-heating. Finally, interferometric devices, such as the Mach- Zehnder interferometer, are also subjected to heater optimisation.

In the second part, the thermal aspects of two different laser sources for optical transceivers are investigated. First, a nano-scale thermal model is developed and validated for the nano-ridge laser, a novel III-V on Si monolithic laser. With the model, device reliability weak spots are identified, and design changes are proposed for self-heating mitigation. Secondly, thermal characterisation of hybrid InP-Si lasers is carried out and benchmarked against different state-ofthe- art integration approaches. The analysis is done both for single-channel and multi-channel laser dies, and the optimal thermal design of a multi-channel light source is investigated. 

In the third and final part, we zoom out and combine multiple photonic devices in integrated photonic circuits. Because there is no off-the-shelf solution for multiscale thermo-optical simulation, new simulation tools are developed. Firstly, thermal-equivalent RC-networks are built for fast circuit-level simulation, considering device-level dynamics and thermal crosstalk. Secondly, multiple machine learning algorithms are trained with finite element simulation data to build extremely compact and black-box models, resulting in three orders of magnitude computational time decrease. The developed models are applied to analyse two optical transceiver designs, for which the impact of 3D electronicphotonic integration is quantified, as well as the impact of thermal crosstalk. In the last chapter all results are combined into a thermo-optical link model, that allows for the calculation of the link efficiency improvement based on the results in this thesis. Current state-of-the-art optical transceivers require 3 pJ/bit, while a thermally optimised design can achieve sub-1 pJ/bit, improving the efficiency more than threefold.

Date:18 Aug 2020 →  9 Nov 2023
Keywords:Photonics, Thermal modelling, Electronics cooling
Disciplines:Photonics, light and lighting, Heat transfer
Project type:PhD project