Automatic qubit control for large-scale quantum computers enabled by neuromorphic computing and cryogenic bio-inspired hardware

通过神经形态计算和低温仿生硬件实现大规模量子计算机的自动量子位控制

• 批准号: RGPIN-2022-04235
• 负责人: Beilliard, Yann
• 金额: $ 1.89万
• 依托单位: Université de Sherbrooke
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2022
• 资助国家: 加拿大
• 起止时间: 2022-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=754106
• 关键词: Automatic; qubit; control; large; scale

The latest major breakthrough in quantum computing (QC) has been the demonstration of quantum systems with more than 50 superconducting qubits allowing to demonstrate quantum supremacy for the first time. Although significant advances have been made in quantum chip fabrication and quantum algorithms, qubit control and error corrections are still performed mostly by hand with bulky classical electronics located outside the cryostat. This approach, characterized by a "wiring bottleneck" between the qubits and the control electronics, makes it impossible to fabricate truly large-scale quantum computers. One of the next major breakthroughs in QC is to automate the control and error corrections of very large numbers of qubits using integrated cryogenic electronics located inside the cryostat (i.e. in-situ). In recent years, several studies have shown that machine learning (ML) could solve or automate difficult data-driven quantum problems. Artificial intelligence (AI) could thus become a key tool for in-situ qubit control, but major research efforts have first to be conducted on AI-dedicated cryo-electronics and ultralow-power AI solutions with high-performance and minimum heat dissipation. Spiking neural networks (SNN) can achieve state-of-the-art energy efficiency and accuracy with unsupervised learning and reservoir computing. Exploring SNN-based solutions for quantum problems is thus a very promising approach. However, current computing hardware based on the serial and digital von Neumann architecture is fundamentally different from the highly paralleled, analog and asynchronous nature of SNNs. This major software/hardware mismatch induces performance and energy efficiency issues. This problem could be overcome by fully embracing neuromorphic engineering to create circuits emulating brain-like functions in hardware. This approach should lead to energy-efficient SNN-based computing suitable for cryogenic technologies with tight thermal constraints. Before QC can benefit from SNN-based in-situ AI, novel cryo-compatible nanodevices emulating synaptic functions are required. In that scope, emerging multi-terminal resistive memories (i.e. memristors) recently appeared as one of the most interesting candidates to emulate synaptic functions in hardware.  RESEARCH PROGRAM In the long term, I plan to use my expertise in neuromorphic computing, AI-dedicated emerging hardware and quantum engineering to develop fully automatic in-situ control of qubits using ultralow-power SNN solutions running entirely on cryo-compatible memristor-based neuromorphic hardware. The outcomes of this research program would allow Canada to stay ahead in the race for the first general-purpose quantum computers. In the short term, I will kickstart 3 interdisciplinary research projects aiming to develop key enabling technologies for AI-based in-situ qubit control by following a unique research path combining memristor-based neuromorphic engineering and quantum technology.

量子计算（QC）的最新重大突破是展示了具有50多个超导量子位的量子系统，首次展示了量子霸权。虽然量子芯片制造和量子算法已经取得了重大进展，但量子位控制和错误校正仍然主要由位于低温恒温器外部的庞大经典电子设备手动执行。这种方法的特点是量子比特和控制电子设备之间的“布线瓶颈”，使得制造真正的大规模量子计算机成为不可能。QC的下一个重大突破之一是使用位于低温恒温器内部的集成低温电子设备（即原位）自动控制和纠正大量量子位的错误。近年来，一些研究表明，机器学习（ML）可以解决或自动化困难的数据驱动的量子问题。因此，人工智能（AI）可能成为原位量子位控制的关键工具，但主要的研究工作必须首先在AI专用的低温电子学和具有高性能和最低散热的超低功耗AI解决方案上进行。尖峰神经网络（SNN）可以通过无监督学习和水库计算实现最先进的能源效率和准确性。因此，探索基于SNN的量子问题解决方案是一种非常有前途的方法。然而，基于串行和数字冯·诺依曼架构的当前计算硬件与SNN的高度可编程性、模拟性和异步性根本不同。这种主要的软件/硬件不匹配会导致性能和能效问题。这个问题可以通过完全接受神经形态工程来克服，以在硬件中创建模拟大脑功能的电路。这种方法应该导致节能SNN为基础的计算适合于低温技术与严格的热约束。在QC可以从基于SNN的原位AI中受益之前，需要模拟突触功能的新型低温兼容纳米器件。在这个范围内，新兴的多端电阻存储器（即忆阻器）最近出现作为在硬件中模拟突触功能的最有趣的候选者之一。研究从长远来看，我计划利用我在神经形态计算方面的专业知识，AI专用新兴硬件和量子工程，使用完全在低温下运行的超低功耗SNN解决方案开发量子位的全自动原位控制，基于忆阻器的兼容神经形态硬件。这项研究计划的成果将使加拿大在第一台通用量子计算机的竞争中保持领先地位。在短期内，我将启动3个跨学科研究项目，旨在通过遵循基于忆阻器的神经形态工程和量子技术相结合的独特研究路径，开发基于人工智能的原位量子比特控制的关键使能技术。