Intelligent sensors

For large-scale implementation, each air quality sensor must provide reliable data during its whole lifetime. However, sensors are not perfect, and they age. In ULISSES, we developed machine learning algorithms for networked Cloud-connected sensors that allow the sensors to stay calibrated with support from their peer or superior sensor friends. Using Cloud intelligence, each sensor can learn from its own and other sensors’ history and self-estimate its own reliability. Moreover, instead of reporting a value as the sensor reading, the sensors provide ranges of likely readings with different probabilities. Such a data representation makes it possible for the sensors to learn from each other and agree on even more confident measurements of the air quality. This work resulted in a number of publications [11, 14], including parts of a PhD-thesis. In the close future, we will see more and more “intelligent” gas sensors that learn from their own history and their friends to provide more reliable data.

Photonic waveguide technology for gas sensors

Today, gas sensors based on infrared spectroscopy are bulky - several cubic centimeters in size - and power hungry. In ULISSES, we aim at realizing chip-integrated gas sensors of just a few cubic millimeters. Such miniaturization is impossible using traditional free-space optics technology because, due to the physics involved sensors become less sensitive the smaller you make them. Instead, we use MOEMS technology, and particularly integrated waveguides, which in general public terms are very tiny optical fibers [16, 17]. Our waveguides have a cross-section smaller than the infrared light they guide. In this way, the light travels along the waveguide partly inside it and partly outside it. This last fraction of light is thus in contact with the gas around the waveguide and it is affected by the gas concentration. Waveguides that are several centimeters long can be fabricated on a very small chip over an area of just a few square millimeters. Until now, we developed and tested all the microfabrication methods [1, 6, 8, 10, 13], including chip-scale packaging [13], required for such micro-optical sensor chips. We recently started the very challenging first fabrication run, where all essential parts of the gas sensor are put together and integrated on the same chip. Just like you, I am eager to see the first prototype at work!

2D materials: Active devices

Optical gas sensors require a light source and a light detector. 2D materials such as graphene and platinum diselenide have unique properties that allow them to convert infrared light into a detectable electric response. In ULISSES, we developed a new graphene photodetector suitable for waveguide integration. It uses electrical bias to form a thermoelectric junction that can efficiently convert the electrons excited by light into an electrical current [4, 9]. Moreover, we developed a new growth technique for the production of high-quality platinum diselenide directly on top of waveguides [5, 7]. The research work on 2D-material photodetectors also enabled us to fabricate and test novel waveguide-integrated infrared thermal light sources.

2D materials: Processing technologies

Typically, the placing of graphene on semiconductor wafers or chips is a manual process that consists in “fishing” the thin graphene layers that are floating like jellyfish in a liquid bath with the target substrate. This is of course not suited for high-quality mass fabrication. In ULISSES, we developed wafer-scale dry transfer processes that result in good coverage with very few cracks, also on surfaces with waveguide topography, and are much better suited for mass production. Finally, we made significant progress in encapsulating graphene to protect it from the environment without damaging it [6].

In conclusion, ULISSES is producing significant advancements on all these fronts to enable the mass implementation of air quality sensors. Every little achievement on this road is part of the puzzle, but it is also a scientific breakthrough in itself, and will find application in many other fields as well.