PhD defense
Coherent optical spectroscopy of InGaAs/GaAs quantum dots doped with a single Mn atom
C2N - Centre de Nanosciences et de Nanotechnologies, , PalaiseauPhD defense
In the context of quantum technologies, InGaAs/GaAs:Mn quantum dots containing a single magnetic atom (a Mn atom) present a level structure and optical selection rules of great interest for the implementation of quantum protocols based on spin-photon entanglement. To confirm the potential of this system, we carried out coherent optical spectroscopy experiments. To this end, we developed a low-temperature (2K) dark-field confocal microscopy setup, based on modal filtering rejection of the reflected resonant laser in a cross-polarized configuration, to perform resonant Raman scattering measurements. By combining 1- and 2-laser experiments addressing the transitions of a V-like system, and an analysis of resonances, Autler-Townes splittings and induced transparency, with a model based on Bloch's equations, we have shown that the coherence of the addressed spin states is largely dominated by their radiative lifetime. Our work also identifies the difficulties that still need to be overcome to make this system truly interesting as a spin-photon interface.
Figure : Principle of 2-laser coherent spectroscopy applied to the levels of an InGaAs/GaAs quantum dot containing a single magnetic Manganese atom.
(in french)
C2N - Centre de Nanosciences et de Nanotechnologies, , PalaiseauPhD defense
Generation of entangled states of spins and photons with a semiconductor quantum dot
IPVF, , PalaiseauPhD defense
Semiconductor quantum dots embedded in photonic cavities are promising emitters of single photons for photonic quantum technologies. In addition, a semiconductor quantum dot can confine a single charge carrier (electron or hole) whose spin can be used as a local memory, leading to interesting applications. In 2009, it was proposed to use such charged quantum dots to emit large entangled states comprising the confined spin and many photons. These so-called cluster states are of particular interest for scalable, measurement-based photonic quantum computing. The challenge in generating these quantum states lies in the combination of efficient photon generation with control over the confined spin: while both aspects have been demonstrated separately, their combination has remained so far elusive.
In this thesis, we use semiconductor quantum dots to experimentally generate entangled states of spin and photons. To do so, we rely on an off-resonant, phonon-assisted excitation technique to generate single photons in an efficient manner, while preserving access to the spin via the emitted photons polarization. We explore the physics of single spins in the quantum dot, characterizing in particular their coherence properties. We then use a charged QD-cavity device to generate three-partite entangled states at high rates, comprising the spin and two indistinguishable photons, thereby demonstrating for the first time the implementation of the 2009 proposal with a quantum dot in a 3-dimensional photonic cavity. Finally, we also explore the use of neutral quantum dots to generate hyperencoded quantum states, where a single photon is in a superposition of several degrees of freedom.
figure : a. Scanning electron microscopy image of the device. A train of linearly-polarized pulses leads to entanglement between the spin and successively emitted photons. b. Energy levels and optical selection rules of the charged quantum dot under small (< 100mT) transverse magnetic field. c. Projection of the measured Bloch vector of the second emitted photon while scanning t23, after measurement of the last photon in R (yellow) or L (blue).
(in french)
C2N - Centre de Nanosciences et de Nanotechnologies, , PalaiseauPhD defense
(in french)
C2N - Centre de Nanosciences et de Nanotechnologies, , PalaiseauPhD defense
(in french)
C2N - Centre de Nanosciences et de Nanotechnologies, , PalaiseauPhD defense
Multi-photon quantum interference and entanglement
Telecom Paris, , PalaiseauPhD defense
In this PhD work, we study key features for optical quantum information processing, namely the possibility to perform quantum interferences and generate entanglement with many photons. We use a semiconductor quantum-dot (QD) embedded in a monolithic micropillar optical cavity as a bright source of pure and indistinguishable single photons. Interfacing such high-performance single-photon sources with reconfigurable glass photonic circuits fabricated using femtosecond laser micromachining, we manipulate up to four photon on chip. We first explore the measurement of multiphoton indistinguishability, a non-trivial task for state of n > 2 photons. It is defined as the overlap of the multiphoton state with the pure state made of n identical photons. We use a scalable interferometer design that has N = 2n modes and includes a cyclic array of beam splitters to experientially quantify the indistinguishability of a 4-photon state on an 8-mode integrated version of this interferometer. We measure a 4-photon indistinguishability of 0.81±0.003, providing a first reference value and benchmark of four-photon indistinguishability. Then, we demonstrate high-fidelity high rate generation of a 4-partite GHZ states on chip, and perform a first proof-of-principle of a real-world application by demonstrating a 4-partite quantum secret sharing protocol. We achieved a fidelity of F=86% to the target GHZ state, and a state purity of P=76.3%. We have certified the genuine entanglement of the generated state with a semi device-independent approach within 39 standard deviations. In our quantum key distribution protocol, we reach a QBER of 10.87%, just below the 11% threshold to ensure a secure communication. Finally, we study remote two-photon interference with photons generated from independent remote electrically tunable bright single-photon sources deterministically fabricated using the in-situ lithography technique. We measure 2 photon interference from four pairs of remote independent bright single- photon sources using pulsed excitation schemes. Us- ing resonant excitation, we show that we can reach up to V=54.8±1% without any spectral filtering, and V=69±1% when the single-photon are filtered to reduce the impact of charge noise.
(in french)
C2N - Centre de Nanosciences et de Nanotechnologies, , PalaiseauPhD defense
(in french)
C2N - Centre de Nanosciences et de Nanotechnologies, , PalaiseauPhD defense
Physics-Grounded Neuromorphic Computing : From Spiking Neurons to Learning Algorithms
C2N - Centre de Nanosciences et de Nanotechnologies, , PalaiseauPhD defense
In our digital era, marked by an exponential growth in computational power and memory capacity, we are confronted with a pressing challenge: the escalating energy consumption of information technology. The increasing demand for data-intensive services, notably artificial intelligence and cloud computing, underscores the urgent necessity for energy-efficient computing solutions that are environmentally sustainable and foster innovation. This thesis explores the potential of memristors for neuromorphic computing to achieve energy-efficient AI.
Because Spiking Neural Networks could offer the promise of low-energy learning, we first focused on hardware neurons composed of volatile NbOx filamentary memristors. These components emerge as appealing alternatives to conventional CMOS devices because of their scalability and spiking behaviors. These devices were characterized and reproduced numerous neuronal spiking and bursting behaviors, such as Leaky-Integrate-and-Fire characteristics and phasic bursting.
We then focused on the algorithmic side and tackled the challenge of adapting the Equilibrium Propagation (EqProp) algorithm to physical systems. EqProp, rooted in physics rather than calculus, offers an attractive prospect—harnessing the inherent physics of hardware systems for on-chip learning. This work revolved around addressing the challenges posed by continuous-valued gradients in a memristor-based environment, where the mode of programming is a series of pulses.
Next, we tested the resilience of the discretized version of EqProp by replacing the ideal software synapses with HfOx memristor data, reaching more than 91% accuracy on the MNIST dataset.