Quantum Optical Neural Networks for Quench Prevention

用于预防猝灭的量子光神经网络

• 批准号: 10073463
• 金额: $ 45.43万
• 依托单位: DUALITY QUANTUM PHOTONICS LTD
• 依托单位国家: 英国
• 项目类别: Feasibility Studies
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=10073463
• 关键词: Quantum; Optical; Neural; Networks; Quench

The need for secure, clean, reliable, and sustainable sources of energy has grown in both importance and urgency. Part of the solution to meet these needs is nuclear fusion. While experimental progress in fusion has evidenced its viability, a range of engineering challenges must be met and coordinated before fusion reactors can operate reliably for long periods, and to deliver a net energy gain.Among these challenges is the processing of large real-time data sets from cryogenically cooled superconducting magnetic coils that maintain the plasma from which energy is released. Superconductivity can break down if a hotspot forms in part of a coil; the subsequent rapid warming and loss of plasma confinement results in damage and downtime. To prevent this, hotspots must be rapidly located so individual coils can be protected.Hotspots can be detected using a process called optical frequency domain reflectometry (OFDR). Laser light is sent down an optical fibre that is co-wound with a coil; a hotspot affects some of the light reflected back along the fibre; its detection allows the hotspots to be located. However, precisely locating hotspots in multiple coils within fractions of a second, requires the rapid processing of vast amounts of data. This information processing challenge is a barrier to clean energy from fusion.As information processing has matured beyond the central processing unit (CPU), a variety of tailored control and computational hardware has emerged including graphics processing units (GPUs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), Neural Networks (NNs) and quantum computing. Each of these sacrifices a general purpose (classical) computing capability to enable much greater power for particular information processing tasks.The people at Duality Quantum Photonics have pioneered integrated photonics as a platform for both Optical Neural Nets (ONNs) and quantum information processing. Quantum Optical Neural Nets (QONNs), the combination of these two paradigms, in integrated photonics, provide an appealing platform for a range of information processing tasks, including the processing of real-time data required to sustain fusion energy generation.In this project, Duality will partner with the private fusion energy company Tokamak Energy, and with the UK Atomic Energy Authority, to design and fabricate QONNs in photonic chips to process OFDR data for the rapid location of hotspots. The project will demonstrate how quantum computing can help tackle some of the information processing challenges that stand in the way of net gain fusion energy.

对安全、清洁、可靠和可持续能源的需求变得越来越重要和紧迫。满足这些需求的部分解决方案是核聚变。虽然聚变实验的进展已经证明了其可行性，但在聚变反应堆能够长期可靠运行并提供净能量增益之前，必须满足和协调一系列工程挑战。这些挑战包括处理大量实时数据由低温冷却的超导磁线圈组成，该线圈维持等离子体并从中释放能量。如果线圈的一部分形成热点，超导性就会被破坏。随后的快速升温和等离子体限制的丧失会导致损坏和停机。为了防止这种情况发生，必须快速定位热点，以便保护各个线圈。可以使用称为光频域反射计 (OFDR) 的过程来检测热点。激光沿着与线圈共绕的光纤发送；热点会影响一些沿着光纤反射回来的光；它的检测可以定位热点。然而，要在几分之一秒内精确定位多个线圈中的热点，需要快速处理大量数据。这种信息处理挑战是融合清洁能源的障碍。随着信息处理的成熟超越了中央处理单元（CPU），出现了各种定制的控制和计算硬件，包括图形处理单元（GPU）、专用集成电路（ASIC）、现场可编程门阵列（FPGA）、神经网络（NN）和量子计算。其中每一个都牺牲了通用（经典）计算能力，以便为特定信息处理任务提供更大的能力。对偶量子光子学的人们开创了集成光子学作为光学神经网络（ONN）和量子信息处理的平台。量子光学神经网络（QONN）是这两种范式的结合，在集成光子学中，为一系列信息处理任务提供了一个有吸引力的平台，包括处理维持聚变能量生成所需的实时数据。在这个项目中， Duality 将与私营聚变能源公司 Tokamak Energy 以及英国原子能管理局合作，设计和制造光子芯片中的 QONN，以处理 OFDR 数据，从而快速定位热点。该项目将展示量子计算如何帮助解决阻碍净增益聚变能源的一些信息处理挑战。