2020/06/30 by Sonia M. Buckley, S.M. Buckley, Alexander N. Tait +13
Computer Science · Engineering · Neuroscience · Physics and Astronomy · #Advanced Memory and Neural Computing #Artificial neural network #Biasing #Characterization (materials science) #Detector #Electronic circuit #Function (biology) #Neural Networks and Reservoir Computing #Neural dynamics and brain function #Neuromorphic engineering #Photonics #Synaptic weight #physics.app-ph #physics.ins-det #physics.optics
paper · pdf · doi:10.1103/physrevapplied.14.054008
published as Phys. Rev. Applied 14, 054008 (2020)
openalex created_date 2020/06/19 · arxiv created 2020/11/02 · openalex publication_date 2020/11/05 · arxiv updated 2020/11/11 · openalex updated_date 2026/08/05
Large-scale spiking neural networks have been designed based on superconducting-nanowire single-photon detectors (SNSPDs) as receiver elements for photonic communication between artificial neurons and synapses. Such large-scale artificial neural networks will require thousands of waveguide-integrated SNSPD devices on a single chip. Efficient waveguide-coupled SNSPDs and high-throughput methods for their characterization at this scale are very challenging. Here, we design, fabricate, and measure SNSPDs that are compatible with these large-scale networks. We demonstrate integrated-photonic circuits to simultaneously characterize many waveguide-coupled SNSPDs with a single-fiber input. We achieve up to 15 waveguide-coupled SNSPDs in a single integrated-photonic circuit and a total of 49 waveguide SNSPDs with a detection plateau out of 49 tested. This is perhaps the largest number of SNSPDs integrated in a photonic circuit to date. We employ several types of photonic circuits to enable rapid and reliable characterization of detector performance. We further demonstrate several important synaptic functions of these detectors. These include a binary response in the detectors with average incident photon numbers ranging from less than 10^\ensuremath-3 to greater than 10, indicating that synaptic responses based on these detectors are independent of the number of incident photons in a communication pulse. Such a binary response is ideal for communication in neural systems. We further demonstrate that the response has a linear dependence of the output-current pulse height on the bias current with up to a factor of 1.7 tunability in pulse height, a function that cannot be obtained without a detection plateau.