Neuromorphic Binarized Polariton Networks

Abstract

The rapid development of artificial neural networks and applied artificial intelligence has led to many applications. However, current software implementation of neural networks is severely limited in terms of performance and energy efficiency. It is believed that further progress requires the development of neuromorphic systems, in which hardware directly mimics the neuronal network structure of a human brain. Here, we propose theoretically and realize experimentally an optical network of nodes performing binary operations. The nonlinearity required for efficient computation is provided by semiconductor microcavities in the strong quantum light-matter coupling regime, which exhibit exciton–polariton interactions. We demonstrate the system performance against a pattern recognition task, obtaining accuracy on a par with state-of-the-art hardware implementations. Our work opens the way to ultrafast and energy-efficient neuromorphic systems taking advantage of ultrastrong optical nonlinearity of polaritons.

Energy-Efficient Neural Network Inference with Microcavity Exciton Polaritons

“…Fully functional photonic neurons including nonlinear activation functions were demonstrated [11][12][13][14], including spiking neurons [15,16], as well as neural networks [17][18][19][20][21][22][23][24][25][26][27][28]. Certain networks achieved high performance in challenging machine-learning tasks, such as image and video recognition [29][30][31][32][33][34][35]. Scalable vector-matrix multiplication operations and convolutions, which are at the core of neural network implementations, were demonstrated using photons [36][37][38][39].…”

“…We recently demonstrated hardware neuromorphic systems where strong nonlinearity resulted solely from interactions of exciton polaritons, quantum superpositions of light and matter [30,31,45]. Such superpositions, in the form of mixed quasiparticles of photons and excitons [46,47], are characterized by excellent photon-mediated transport properties and strong exciton-mediated interactions [48,49].…”

“…Such superpositions, in the form of mixed quasiparticles of photons and excitons [46,47], are characterized by excellent photon-mediated transport properties and strong exciton-mediated interactions [48,49]. Their unique properties led to the discovery of remarkable phenomena, including nonequilibrium Bose-Einstein condensation [47,50], superfluidlike states [51,52], interactions of quantum defects [53][54][55], lasing of topological edge states [56,57], as well as demonstration of polariton transistors and switches [49,[58][59][60][61][62], gates [31,63], neurons [64,65], simulators [66], and nonlinear phenomena at the femtojoule level [48,49]. In our proof-of-principle neural network implementations, both photonic and electronic layers are used, but the latter performed linear operations only.…”

Quantum Neuromorphic Computing with Reservoir Computing Networks

“…[ 25,26 ] Reservoir networks have been physically realized with a variety of hardware approaches, [ 27,28 ] including photonic arrays on silicon chips, [ 22 ] nonlinear optical elements coupled to delay lines, [ 23,24,29–31 ] magnetic tunnel junctions, [ 32 ] memristor arrays, [ 33–35 ] and exciton‐polariton lattices. [ 36,37 ] Given that reservoir networks have been an accessible design for different hardware, they seem a natural candidate for building quantum neural networks, where the nodes of classical reservoir networks are replaced with quantum entities. Here we review recent progress in the development of quantum reservoir networks.…”

Dynamics of light-matter coupled quasiparticles

“…Exciton-polaritons also exhibit a significant nonlinearity, which in theory was predicted to itself induce topological behavior [97], antichiral behavior [215], non-Hermitian bi-skin e↵ect [228], and topological solitons [196]. Separate from topological physics, nonlinearity was shown to provide low-energy polaritonic switches [3,7,8], reservoir networks [229][230][231][232], artificial life [233], and realize other information processing elements such as transistors [3][4][5][6], routers [216][217][218], and amplifiers [9,10]. The possible use of topological polariton states at the edges of a strip to couple these elements su↵ers however from an obvious drawback: such states come in pairs, which propagate in opposite directions on the opposite edges of the strip [94,96].…”