Quantum photonic neural networks to predict instabilities in tokamaks

实现净零排放是减缓气候变化的关键和必要步骤，也是限制全球变暖的综合战略的重要组成部分;与此同时，对安全可靠的能源的需求也变得越来越重要。核聚变有可能实现碳中和，产生可持续能源，而不会产生大量温室气体排放。但是，尽管实验进展越来越多地证明了聚变的可行性，但在聚变反应堆能够长期可靠运行之前，必须满足一系列工程挑战，以提供净能量增益。这些挑战之一是防止能量释放的等离子体中断。除了降低效率外，中断事件还可能损坏聚变反应堆，并导致发电厂严重停机。然而，聚变等离子体的行为是复杂的，并且必须快速处理从诊断中生成的大数据集，这使得在足够的时间内预测中断事件以允许缓解行动非常具有挑战性。可以定制专门的计算硬件来实现机器学习并提供真实的时间数据分析。流行的专用处理器包括图形处理单元（GPU）、专用集成电路（ASIC）、现场可编程门阵列（FPGA）和光学神经网络（ONN）。每一个都牺牲了通用计算能力，以实现更大的能力来执行特定的信息处理任务。最近，量子技术因其为某些信息处理任务提供指数加速的能力而受到关注。作为支持量子信息处理和机器学习的平台，集成光子学具有通用性、鲁棒性和可制造性。在该项目中，DQP将开发量子光子神经网络（QPNN），以快速分析等离子体诊断的大型数据集并预测中断事件。DQP将使用最适合光子量子信息处理的新材料在芯片中设计和制造QPNN，并且对辐射和大磁场具有弹性。DQP的QPNN将在英国原子能机构专家的建议下开发。QPNN最终可以大规模生产，从而可以真实的实时地单独监测和控制大量的聚变反应堆部件。DQP汇集了光子芯片制造、量子信息处理和人工智能的几种前沿方法。将这套强大的技术应用于提供净正聚变能的重大工程挑战，开启了一个令人兴奋的科学技术新时代，并为我们所有的未来提供清洁的可持续能源。

Achieving Net Zero emissions is a critical and necessary step towards mitigating climate change and an essential component of a comprehensive strategy to limit global warming; meanwhile, the need for secure and reliable sources of energy has grown in importance. Nuclear fusion has the potential to be carbon-neutral and produce sustainable energy without significant greenhouse gas emissions. But although experimental progress increasingly evidences the viability of fusion, a range of engineering challenges must be met before fusion reactors can operate reliably for long periods, to deliver a net energy gain. Among these challenges is preventing disruptions of the plasma from which energy is released. In addition to reducing efficiency, disruption events can damage fusion reactors and cause significant power plant downtime. However the behaviour of a fusion plasma is complex and large data sets generated from diagnostics must be rapidly processed, making it very challenging to predict disruption events in sufficient time to allow mitigating action. Specialised computational hardware can be tailored to implement machine learning and provide real time data analysis. Popular special purpose processors include graphics processing units (GPUs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), and Optical Neural Nets (ONNs). Each of these sacrifices a general purpose computing capability to enable much greater power for particular information processing tasks. More recently, quantum technologies have attracted attention for their ability to deliver exponential speedups for certain information processing tasks. As a platform for supporting both quantum information processing and machine learning, integrated photonics is versatile, robust, and manufacturable. In this project Duality Quantum Photonics (DQP) will develop Quantum Photonic Neural Nets (QPNNs) to rapidly analyse the large data sets from plasma diagnostics and predict disruption events. DQP will design and fabricate QPNNs in chips using new materials that are best suited to photonic quantum information processing and are resilient to radiation and large magnetic fields. DQP's QPNNs will be developed with advice from experts within the UK Atomic Energy Agency. QPNNs can ultimately be produced at scale so that large numbers of fusion reactor components can be individually monitored and controlled in real time. DQP have brought together several cutting edge approaches to photonic chip manufacture, quantum information processing and AI. Bringing this powerful set of technologies to bear on the grand engineering challenge of delivering net positive fusion energy opens an exciting new era of science and technology, and clean sustainable energy for all our futures.