AI-powered micro-comb lasers: a new approach to transfer portable atomic clock accuracy in integrated photonics

Optical frequency combs are lasers with a frequency spectrum composed of a sequence of lines corresponding to precise trains of optical pulses in time. When miniaturised in portable and energy-efficient platforms, these lasers can provide a precisely beating "optical heart" required by transformative quantum technologies, such as portable optical atomic clocks, gravitational sensors, and dual-comb spectrometers. As outlined in the NQTP strategic plan, these technologies can transform our society by revolutionising healthcare, mobility, financial transactions, and next-generation mobile infrastructures.

In the last fifteen years, the science of miniaturised frequency combs has reached an impressive level of technological maturity. The established platform towards compact and energy-efficient devices is micro-combs, a class of lasers based on miniaturised nonlinear resonators. While extensive efforts have now brought miniaturisation within grasp, micro-combs remain surprisingly hard to control at the high-power emissions regimes required by energy-demanding applications, such as portable atomic clocks and broadband telecommunications. In all these domains, the micro-comb spectrum should be as broad as possible (i.e., octave-spanning), with spectrum lines possessing specific features (e.g., flatness across telecom bands) while carrying enough optical energy to lock to external references.

Meeting these requirements remains particularly challenging in micro-comb platforms, where the ultrafast pulses originate from the interaction and synchronisation of thousands of optical waves. These interactions become increasingly hard to control when approaching high power emissions, leading to chaotic and unpredictable emissions. As a result, state-of-the-art micro-combs are restricted to relatively low optical powers, where nonlinear interactions are easier to tame, and standard stabilisation techniques still apply.

This limitation is surprisingly ubiquitous across many laser technologies, where high-energy emission states have remained substantially uncharted territory. In this regard, highly nonlinear lasers are the photonic counterpart of complex systems like the brain, weather and society. In these systems, a large number of interacting elements and a high degree of nonlinearity provide an essential ingredient to produce high-level functionalities. However, complexity also eludes the definition of universal, interpretable models to understand and control the evolution of these systems. Learning how to tame the extreme richness of complex interactions requires a profound paradigm shift in concepts and methodology. This transformation is today one step closer thanks to the impressive advances in Artificial Intelligence (AI) technologies.

This project's vision is to bring such a paradigm shift in the field of micro-combs by developing a new class of AI-powered lasers capable of "learning" how to optimise their emission in real-time and in real-life experimental conditions. AI is emerging as an ideal tool to stabilise ultrafast lasers in standard emission regimes, delivering improved performance in a fraction of the time. Nevertheless, driving and maintaining a micro-comb laser into an arbitrary, traditionally unstable emission state requires extending AI predictions with sophisticated physical modelling of the system's internal dynamics not necessarily known a priori.

To fill this gap, I will establish and lead an interdisciplinary group of researchers to develop a new methodology based on data-driven discovery, an emerging theoretical framework combining the powerful data-processing capabilities of AI with concepts from dynamical systems and nonlinear control theory. This approach will allow identifying the "hidden" nonlinear effects driving a real-life micro-comb system, opening a unique pathway to apply advanced control strategies and design entirely new generations of micro-combs, inconceivable with existing approach