Cryogenic system for the exploration of low-temperature neuromorphic photonic systems

这项NSERC-RTI提案将通过购买低温系统来支持低温神经形态光子学平台的建设，这可能是加拿大的第一个。我们在神经形态光子学方面的工作（Shastri et al.）。Nature Photonics 15, 2021)利用光学物理原理模拟大脑中的神经元和突触（神经元与神经元之间的连接）来创建光子神经网络（即由光供电的芯片），其速度比电子处理器快几个数量级。我们一直在扩展这项工作，以构建与量子技术接口的低温光子神经网络，以实现量子光子机器学习和实时量子态断层扫描的应用。机器学习和量子信息的物理混合将在这两个领域取得进展——量子用于机器学习，机器学习用于量子。本提案要求为购买低温系统提供资金，该系统包括低温恒温器、吸附冰箱、温度控制器和传感器、水冷氦压缩机和真空泵系统。该冷冻系统将使我们能够将硅光子神经网络与单光子量子门、单光子源和超导纳米线单光子探测器（SNSPDs）连接起来。我们正在研究量子信息处理，通过设计神经形态处理器和量子门之间的硬件接口，在单个硅光子芯片上实现混合量子-经典神经网络，以改进用光子神经网络实现的机器学习算法。相反，我们正在研究利用光子神经网络的机器学习来实时控制量子器件。在与Rotenberg Group （Queen’s）的合作中，我们正在将光子神经回路和单光子源之间的接口硬连接起来，这将能够实时处理和控制脆弱的量子光态，这是所有量子光子技术的基础。该冷冻系统将支持15 HQP/年的实验工作，涉及纳米光子学、机器学习和量子信息处理。2019年，全球人工智能和量子计算市场的价值分别为399亿美元和5.071亿美元。如果这些领域合并，它们的整体市场价值可能大于其部分的总和。在这一领域的交叉点工作的HQP将具有独特的定位，可以通过创建初创企业或成为该领域的领先学者来进入联合技术市场。

This NSERC-RTI proposal will support the building of a cryogenic neuromorphic photonics platform-perhaps the first in Canada-through the purchase of a cryogenic system. Our work on neuromorphic photonics (Shastri et al. Nature Photonics 15, 2021) uses principles of optical physics to mimic neurons and synapses (neuron-to neuron connections) in the brain to create photonic neural networks (i.e., chips powered by light) that are orders of magnitude faster than electronic processors. We have been extending this work to build low-temperature photonic neural networks interfaced with quantum technologies to enable applications in quantum photonic machine learning and real-time quantum state tomography. The physical hybridization of machine learning and quantum information stands to advance in both fields-quantum for machine learning, and machine learning for quantum. This proposal requests funding for the purchase of a cryogenic system that includes a cryostat, a sorption fridge, temperature controllers and sensors, a water-cooled helium compressor, and a vacuum pumping system. This cryosystem will allow us to interface our silicon photonic neural networks with single-photon quantum gates with single photon sources and superconducting nanowire single-photon detectors (SNSPDs). We are investigating quantum information processing to improve machine learning algorithms implemented with photonic neural networks by designing a hardware interface between neuromorphic processors and quantum gates enabling hybrid quantum-classical neural network on a single silicon photonic chip. Conversely, we are investigating machine learning with photonic neural networks for the real-time control of quantum devices. In collaboration with Rotenberg Group (Queen's), we are hardwiring an interface between a photonic neural circuit and single photon sources, that will enable processing and control of the fragile quantum optical states, that underlie all quantum photonic technologies, in real-time. The cryosystem will support the experimental work of 15 HQP/year across nanophotonics, machine learning, and quantum information processing. In 2019, the global AI and quantum computing markets were valued at $39.9B and $507.1M, respectively. If these fields merge, their overall market value could be greater than the sum of their parts. HQP working at the intersection of this field will be uniquely positioned to tap into the combined technology market by creating start-ups or become leading academics in this field.