Adam WANG SHANGHAI MEGA
SENIOR MEMBER
- Joined
- Aug 7, 2017
- Messages
- 5,082
- Reaction score
- -15
- Country
- Location
Alibaba announced that it has developed the world's strongest quantum circuit simulator "Taizhang"
5月8日，阿里巴巴量子实验室施尧耘团队宣布于近日成功研制当前世界最强的量子电路模拟器，名为“太章”。 基于阿里巴巴集团计算平台在线集群的超强算力，“太章”在世界上率先成功模拟了81（9x9）比特40层的作为基准的谷歌随机量子电路，之前达到这个层数的模拟器只能处理49比特。
量子霸权似乎在上演一场“接力战”。
2月，IBM对外展示了其50个量子比特原型机，内部结构图也曝光；
3月，谷歌公布72位量子比特处理器Bristlecone。
3月底，微软发现天使粒子——马约拉纳费米子（Majorana fermion）存在的有力证据，有望年底前得到可工作的量子比特。
现在，轮到阿里上场了。
5月8日，阿里巴巴量子实验室施尧耘团队宣布于近日成功研制当前世界最强的量子电路模拟器，名为“太章”。
基于阿里巴巴集团计算平台在线集群的超强算力，“太章”在世界上率先成功模拟了81（9x9）比特40层的作为基准的谷歌随机量子电路，之前达到这个层数的模拟器只能处理49比特。
同时，本次模拟任务只动用了阿里巴巴计算平台在线集群14%的计算资源。“太章”的创新算法通信开销极小，得以充分发挥平台在线集群的优势，在过去超级计算机上做不了的模拟任务，比如64（8x8）比特40层的模拟，“太章”只需2分钟即可完成。
阿里巴巴“太章”模拟器与目前主要模拟器模拟谷歌随机电路的结果比较
“太章”模拟的随机量子电路规模与谷歌量子硬件可以实现的规模对比
量子计算可能颠覆当前的计算技术，是科学界和工业界研究的前沿热点。但量子计算的实现十分困难。目前，已经实现的高精度量子处理器也只有20几个量子比特。故而规模稍大的量子算法尚无运行的载体。
模拟器的作用在于“承上启下”，往下可以帮助理解、设计硬件，向上可以承载算法和应用的探索和验证。“太章”首次使得测试和验证被称为“中等规模”50-200比特的的量子算法成为可能, 从而为辅助设计中等规模量子算法、量子软件乃至量子芯片提供了一个有力的工具。
在通常的量子电路模拟方案中，需要存储量子状态的全部振幅，在此海量数据上同时模拟量子运算。这个方法要求不断地在众多的计算节点间交换数据，造成巨大的通讯开销。因此，过去这样的模拟任务往往都在超级计算机上进行。
实验室团队基于施尧耘教授及其合作者Igor Markov在2005年提出的另一种模拟方案，发明了一个简单而有效的方法分解整个模拟任务，然后十分均衡地把这些子任务分配到不同计算节点上。“太章”的通信开销极小，这个优点使之十分适合分布式的计算平台。
“太章”模拟的随机量子电路规模（黑线）与谷歌量子硬件可以实现的规模(红线) 比较（基于谷歌在[Characterizing quantum supremacy in near-term devices]中对7x7的估计）*
作为基准的随机量子电路是谷歌提出为实现“量子霸权”的算法。“量子霸权”指的是量子处理器的规模和精度到达无法被经典计算模拟的程度。谷歌今年3月份提出了未来工作的目标：72比特高精度的量子处理器。“太章”的结果表明这一计划中的处理器如果只运行该基准算法仍不足于达到量子霸权。
本次研究成果也提交到预印本网站arXiv，文章并列第一作者为量子实验室量子科学家陈建鑫博士与实习生张放，作者还有实习生黄甲辰和Michael Newman博士。
阿里巴巴量子实验室由美国密西根大学终身教授、世界著名量子科学家施尧耘担任首席量子技术科学家、量子实验室主任。两次理论计算机最高奖哥德尔奖得主、匈牙利裔美国计算机科学家马里奥·塞格德（Mario Szegedy）于今年年初也加入该实验室。实验室正处于人才引进的高速增长时期。
2016年，谷歌提出通过实现二维阵列MxN对应的量子比特上的一类特定随机量子电路来实现量子霸权的方案，这一类特定随机量子电路通常被称为量子霸权电路。在方案中，认为当该二维阵列上的比特数(MN)达到50, 电路的深度（层数）到达40左右，现有世界上最强大的超级计算机也无法有效模拟这样的电路。
8x8二维网格上一个深度为20的量子霸权电路对应的张量网络展示
谷歌的硬件团队希望将在9量子比特1维阵列中实现的1%读取误差，0.1%单比特门误差，0.6%两比特门误差保持到更大规模的量子系统来实现这样的霸权电路，并通过这个特定任务，实现量子硬件对当前世界上最强大的经典计算资源的超越。此后，若干研究团队纷纷在不同的超级计算机上对该类电路进行模拟。之前，全球最好的研究结果尚未同时达到50比特40层。
nxn二维网格上，计算随机电路输出每一个振幅的执行时间与电路深度的对应关系
在量子计算目前的模型中，有一类是量子电路模型，实现形式是将信息存储在量子比特中，通过类似经典逻辑门的量子门来实现计算。达摩院量子实验室团队量子科学家陈建鑫与实习生张放实现了一种基于分布式的通用量子电路模拟方案，并基于研究的模拟器对谷歌第一版的随机量子电路进行了测试。
利用阿里计算平台的在线集群的少量计算资源(14%左右)实验室团队成功使用“太章”模拟器模拟了9x9 x40也就是81比特40层随机电路，还分别成功模拟
了100比特35层（10x10x35）, 121比特31层（11x11x31）与144比特27层（12x12x27）的随机量子电路。
目前业界主流的模拟方案有两类，一类是存储量子状态的所有振幅，一类是对于任意振幅都可以迅速计算得到结果。第一类模拟方案，基本都在超级计算机上实现，因为存储45比特的量子状态需要Petabyte量级的内存，在存储这么多数据的同时对该量子态进行操作并进行计算，需要不断地在不同的计算节点之间交换数据，这样的通讯开销对于普通云服务是难以承受的。
在阿里巴巴计算平台的在线集群上，实验室团队采用了第二类模拟方案，通过快速有效的计算任意振幅，任务拆分后可以将子任务十分均衡地分配到不同节点，极少的通信开销使得模拟器适配现在广泛提供服务的云计算平台。
在本研究成果之前，对于两种模拟方案，全球尚未有研究团队可以成功模拟谷歌超过50比特40层的第一代随机测试电路。在达摩院量子实验室团队的模拟器内还可以每2分钟计算64比特40层随机电路的一个振幅。本次研究成果也已经以论文的形式在预印本网站arXiv上提交，文章并列第一作者为量子实验室量子科学家陈建鑫与实习生张放，作者还有实习生黄甲辰和Michael Newman博士。
本次研究成果arXiv论文链接： https://arxiv.org/abs/1805.01450
谷歌、IBM、微软量子霸权混战，施尧耘：超导VS离子阱，量子计算进入两极世界
今年三月，在洛杉矶举行的美国物理学会年会上，谷歌展示了一个新的量子处理器Bristlecone。这个基于门的超导系统目的在于研究量子比特技术的系统误差率和可扩展性，以及在量子模拟、优化和机器学习中的应用。
左边是谷歌最新的72量子比特量子处理器Bristlecone。右边是该设备的图示：每个“X”代表一个量子比特，量子比特之间以线性阵列方式相连。来源：Google Quantum AI Lab
谷歌量子AI实验室研究科学家Julian Kelly在Google Research官博发文介绍，Bristlecone遵循是谷歌之前提出的9个量子比特量子计算机的线性阵列技术所对应的物理学原理，而该技术显示的最佳结果如下：
低的读数错误率（1%）、单量子比特门（0.1%）以及最重要的双量子比特门（0.6%）。
该设备使用与9个量子比特的相同的模式进行耦合、控制和读出，但将其扩展为一个包含72个量子比特的正方形数组。
谷歌研究人员计算后认为，量子霸权的目标可以通过使用49个量子比特，一个超过40的电路深度，一个低于0.5%的2个比特误差进行完美的证明。
有观点认为，谷歌之所以公布72位量子比特处理器，是因为量子霸权之路遇到了对手IBM。
2017年11月，IBM宣布成功构建并测量了具有类似性能指标的50个量子比特原型机，今年2月又曝光了其内部结构图。
50个量子比特被普遍认为可以进行普通超级计算机不能完成的任务，IBM此举也是在“量子霸权”上具有里程碑意义的一步。
因为谷歌和IBM的量子处理器都是通过超导来实现量子计算，因此这两家公司在量子霸权上你追我赶。
但是另外一股力量也随时有可能爆发，那就是微软。
微软押注拓扑量子计算，虽然目前还没有做出来相互作用的量子比特，但一直在开发量子硬件以及量子计算机软件开发套件。
微软的逻辑在于，虽然谷歌、IBM都做出了量子比特，但这些都是不精确的量子比特，来自外部环境的微小震动或能量都可能导致计算错误。
而微软的拓扑量子计算机可能能够大大降低噪音。微软量子计算业务发展总监Julie Love曾说：“我们的一个量子比特将会有1000个、甚至10000个嘈杂的量子比特那样强大。”他们认为，微软将在今年年底前得到可工作的量子比特。
施尧耘今年2月发表在新智元上的一篇文章也提到，宇宙中第一个拓扑量子比特将在今年爆发，而微软是有可能做出来的。
事实也证明，微软也离量子霸权更近了一步。
今年3月底，微软的研究人员观察到被称为“天使粒子”的马约拉纳费米子存在的相当有力的证据：电子在他们的导线中分裂成半体。
如果微软希望建造一台能工作的量子计算机，这将是至关重要的。
另外，离子阱量子计算在今年下半年可能会做出手握50-60几个比特的利器，施尧耘认为，这一领域最大的赢家是没有稳定量子比特的Amazon和Facebook，量子计算将进入两极世界时代。
来源：新浪科技
Translations:
On May 8, the team of Alibaba Quantum Labs announced that it has successfully developed the world’s most powerful quantum circuit simulator called “Taizhang” recently. Based on the powerful computing power of Alibaba Group's online computing platform, "Tai Zhang" successfully took the lead in successfully simulating 81 (9x9) bits and 40 layers of Google's random quantum circuits as benchmarks. Previously, this layer of simulator was only Can handle 49 bits.
Quantum hegemony seems to be playing a "relay battle."
In February, IBM demonstrated its 50 qubit prototypes to the outside world, as well as its internal structure.
In March, Google announced the 72-bit qubit processor Bristlecone.
At the end of March, Microsoft found strong evidence of the existence of the Angel particle, Majorana fermion, and it is expected that the working qubit will be available before the end of the year.
Now it's time for Ali to play.
On May 8, the team of Alibaba Quantum Labs announced that it has successfully developed the world’s most powerful quantum circuit simulator called “Taizhang” recently.
Based on the powerful computing power of Alibaba Group's online computing platform, "Tai Zhang" successfully took the lead in successfully simulating 81 (9x9) bits and 40 layers of Google's random quantum circuits as benchmarks. Previously, this layer of simulator was only Can handle 49 bits.
At the same time, this simulation task only uses 14% of the computing resources of the Alibaba Computing Platform online cluster. "Tai-Zhang"'s innovative algorithm has minimal communication overhead and can take full advantage of the platform's online clustering. In the past, supercomputers could not do simulation tasks such as 64-bit (8x8)-bit 40-layer simulation, "Taizhang" only 2 Minutes to complete.
Comparison of Alibaba "Taizhang" Simulator and Main Simulator Simulating Google Random Circuit
Comparison of Alibaba "Taizhang" Simulator and Main Simulator Simulating Google Random Circuit
Comparing the Scale of Random Quantum Circuit Simulated by "Taizhang" with the Scale Computed by Google Quantum Hardware
Quantum computing may subvert the current computing technology and is a hot topic in the scientific and industrial research. However, the realization of quantum computing is very difficult. At present, the high-precision quantum processor that has been implemented has only a few 20 qubits. Therefore, a slightly larger quantum algorithm has no carrier yet to operate.
The role of the simulator is to "bring the link up and down." It helps understand and design the hardware, and it can carry forward the exploration and verification of algorithms and applications. For the first time, "Tai-Zhang" made it possible to test and verify a quantum algorithm called "medium-scale" 50-200-bit, thus providing a powerful tool for assisting in the design of medium-scale quantum algorithms, quantum software, and even quantum chips.
In a typical quantum circuit simulation scheme, it is necessary to store the full amplitude of the quantum state and simultaneously simulate quantum operations on the massive data. This method requires the constant exchange of data among numerous computing nodes, resulting in a huge communication overhead. Therefore, in the past, such simulation tasks were often performed on supercomputers.
The laboratory team, based on another simulation scheme proposed by Prof. Shi and its collaborator Igor Markov in 2005, invented a simple and effective method to decompose the entire simulation task, and then distributed these subtasks to different computing nodes in a well-balanced manner. . "Taizhang" has very little communication overhead. This advantage makes it very suitable for distributed computing platforms.
Comparison of the size of the random quantum circuit (black line) simulated by "Tai Zhang" and the scale (red line) that Google's quantum hardware can achieve (based on Google's estimate of 7x7 in [Characterizing quantum supremacy in near-term devices]) *
Comparison of the size of the random quantum circuit (black line) simulated by "Tai Zhang" and the scale (red line) that Google's quantum hardware can achieve (based on Google's estimate of 7x7 in [Characterizing quantum supremacy in near-term devices]) *
The random quantum circuit as a reference is Google's algorithm for achieving "quantum hegemony". "Quantum hegemony" refers to the extent to which the quantum processor's size and precision can't be simulated by classical calculations. In March this year, Google proposed the goal of future work: 72-bit high-precision quantum processors. The result of "Tai Zhang" shows that the processor in this plan is still insufficient to achieve quantum hegemony if only the benchmark algorithm is run.
The research results were also submitted to the pre-printed website arXiv. The article was tied with the first author for Quantum Lab Quantum Scientist Dr. Chen Jianxin and the intern Zhang Fang, the author and intern Huang Jiachen and Dr. Michael Newman.
Alibaba Quantum Lab is the lifelong professor of the University of Michigan and Shi Quan, a world-renowned quantum scientist, as the chief quantum technology scientist and director of Quantum Lab. Mario Szegedy, a Hungarian-American computer scientist who was also the recipient of the two highest computer science awards for the Gödel Prize, joined the lab earlier this year. The laboratory is in a period of high-speed growth in talent introduction.
In 2016, Google proposed a scheme for realizing quantum hegemony by implementing a specific random quantum circuit on a qubit corresponding to a two-dimensional array MxN. This type of specific random quantum circuit is often called a quantum hegemonic circuit. In the scheme, it is considered that when the number of bits (MN) on the two-dimensional array reaches 50, the depth (number of layers) of the circuit reaches about 40, and the most powerful supercomputer in the world today cannot effectively simulate such a circuit.
A tensor network demonstration of a quantum hegemonic circuit with a depth of 20 on an 8x8 two-dimensional grid
A tensor network demonstration of a quantum hegemonic circuit with a depth of 20 on an 8x8 two-dimensional grid
Google’s hardware team hopes to achieve such a hegemonic circuit by maintaining a 1% read error in a 9-qubit 1D array, a 0.1% single-bit gate error, and a 0.6% 2-bit gate error to a larger scale quantum system. And through this particular task, Quantum hardware is surpassed by the world’s most powerful classic computing resources. Since then, several research teams have simulated these circuits on different supercomputers. Previously, the world’s best research results have not reached 50-bit and 40-story at the same time.
Calculating the relationship between the execution time and the circuit depth of each amplitude output of the random circuit on the nxn two-dimensional grid
Calculating the relationship between the execution time and the circuit depth of each amplitude output of the random circuit on the nxn two-dimensional grid
In the current model of quantum computing, there is a quantum circuit model in which the information is stored in qubits and calculated by a quantum gate similar to a classical logic gate. Quantum scientist Chen Jianxin of the Boomerang Quantum Lab team and intern Zhang Fang implemented a distributed universal quantum circuit simulation scheme and tested the first random quantum circuit of Google based on the research simulator.
A small amount of computational resources (around 14%) using online clustering of Alibaba Computing Platform successfully used the "Taizhang" simulator to simulate a 9x9x40, which is an 81-bit 40-layer random circuit, and successfully simulated separately.
A 100-bit, 35-layer (10x10x35), 121-bit, 31-layer (11x11x31) and 144-bit, 27-layer (12x12x27) random quantum circuit.
Currently, there are two main types of simulation schemes in the industry. One is to store all the amplitudes of the quantum state, and the other is to quickly calculate the result for any amplitude. The first type of simulation schemes are basically implemented on supercomputers because storing 45-bit quantum states requires Petabyte-level memory. The operation and calculation of the quantum state while storing so much data need to be continuously different. The exchange of data between computing nodes, such communication overhead is unbearable for ordinary cloud services.
On the online cluster of Alibaba's computing platform, the laboratory team used the second type of simulation program to quickly and efficiently calculate arbitrary amplitudes. After task splitting, the sub-tasks can be distributed evenly to different nodes with very little communication overhead. The simulator is adapted to a cloud computing platform that is now widely available for service.
Prior to the results of this study, no research team in the world has been able to successfully simulate the first generation of random test circuits with more than 50 bits and 40 layers for Google. An amplitude of 64-bit 40-layer random circuits can also be calculated every 2 minutes in the simulator of the Dharma Quantum Lab team. The research results have also been submitted in the form of papers on the pre-printed website arXiv. The first author of the paper is Quantum Lab Quantum Scientist Chen Jianxin and Intern Zhang Fang, and the author and intern Huang Jiachen and Dr. Michael Newman.
The results of this research arXiv paper link: https://arxiv.org/abs/1805.01450
Google, IBM, Microsoft Quantum Hegemony, Shi Zheng: Superconductivity VS Ion Trap, Quantum Computing Enters the Polar World
In March of this year, at the annual meeting of the American Physical Society in Los Angeles, Google demonstrated a new quantum processor Bristlecone. The purpose of this gate-based superconducting system is to study the systematic error rate and scalability of quantum bit technology, and its application in quantum simulation, optimization, and machine learning.
On the left is Google’s latest 72-bit quantum quantum processor, Bristlecone. On the right is a diagram of the device: Each "X" represents a qubit, and the qubits are connected in a linear array. Source: Google Quantum AI Lab
On the left is Google’s latest 72-bit quantum quantum processor, Bristlecone. On the right is a diagram of the device: Each "X" represents a qubit, and the qubits are connected in a linear array. Source: Google Quantum AI Lab
Julian Kelly, a research scientist at Google Quantum AI Labs, presented in the article published by Google Researcher Bristlecone follows the physics principle of the linear array technology of the nine quantum-bit quantum computers proposed by Google, and the best results shown by the technology are as follows: :
Low reading error rate (1%), single qubit gate (0.1%), and most important double qubit gate (0.6%).
The device uses the same pattern as the nine qubits for coupling, control, and readout, but expands it to a square array of 72 qubits.
Google researchers calculated that the goal of quantum hegemony can be perfectly demonstrated by using 49 qubits, a circuit depth of more than 40, and a 2-bit error of less than 0.5%.
There are views that Google announced the 72-bit qubit processor because the quantum hegemony road encountered rival IBM.
In November 2017, IBM announced the successful construction and measurement of 50 qubit prototypes with similar performance specifications. In February this year, it also exposed its internal structure.
The 50 qubits are generally considered to be tasks that cannot be accomplished by ordinary supercomputers. IBM's move is also a milestone step in Quantum Hegemony.
Because both Google and IBM's quantum processors implement quantum computing through superconductivity, the two companies are chasing after one another in quantum hegemony.
However, there is also the possibility that another force may erupt at any time. That is Microsoft.
Microsoft is betting on topological quantum computing. Although no interworking qubits have yet been made, quantum hardware and quantum computer software development kits have been developed.
The logic of Microsoft is that although Google and IBM have all made quantum bits, these are inaccurate qubits, and tiny vibrations or energy from the external environment can cause calculation errors.
Microsoft's topological quantum computer may be able to greatly reduce noise. Julie Love, the development director of Microsoft's quantum computing business, once said: "One of our qubits will be as powerful as 1,000 or even 10,000 noisy qubits." They think that Microsoft will get a working qubit before the end of this year.
An article published by Shi Zhi in February this year on Xinzhiyuan also mentioned that the first topological qubit in the universe will explode this year, and Microsoft is likely to make it.
The facts also prove that Microsoft is a step closer to quantum hegemony.
At the end of March of this year, Microsoft researchers observed the strong evidence of the presence of Mayorana Fermi, known as the "Angel Particle," that electrons split into halves in their wires.
If Microsoft wants to build a working quantum computer, this will be crucial.
In addition, ion trap quantum computing may be able to hold 50-60 bits of weapon in the second half of this year. Shi Yong believes that the biggest winner in this field is Amazon and Facebook without stable quantum bits. Quantum computing will enter the bipolar world. era.
Source: Sina Technology
5月8日，阿里巴巴量子实验室施尧耘团队宣布于近日成功研制当前世界最强的量子电路模拟器，名为“太章”。 基于阿里巴巴集团计算平台在线集群的超强算力，“太章”在世界上率先成功模拟了81（9x9）比特40层的作为基准的谷歌随机量子电路，之前达到这个层数的模拟器只能处理49比特。
量子霸权似乎在上演一场“接力战”。
2月，IBM对外展示了其50个量子比特原型机，内部结构图也曝光；
3月，谷歌公布72位量子比特处理器Bristlecone。
3月底，微软发现天使粒子——马约拉纳费米子（Majorana fermion）存在的有力证据，有望年底前得到可工作的量子比特。
现在，轮到阿里上场了。
5月8日，阿里巴巴量子实验室施尧耘团队宣布于近日成功研制当前世界最强的量子电路模拟器，名为“太章”。
基于阿里巴巴集团计算平台在线集群的超强算力，“太章”在世界上率先成功模拟了81（9x9）比特40层的作为基准的谷歌随机量子电路，之前达到这个层数的模拟器只能处理49比特。
同时，本次模拟任务只动用了阿里巴巴计算平台在线集群14%的计算资源。“太章”的创新算法通信开销极小，得以充分发挥平台在线集群的优势，在过去超级计算机上做不了的模拟任务，比如64（8x8）比特40层的模拟，“太章”只需2分钟即可完成。
阿里巴巴“太章”模拟器与目前主要模拟器模拟谷歌随机电路的结果比较
“太章”模拟的随机量子电路规模与谷歌量子硬件可以实现的规模对比
量子计算可能颠覆当前的计算技术，是科学界和工业界研究的前沿热点。但量子计算的实现十分困难。目前，已经实现的高精度量子处理器也只有20几个量子比特。故而规模稍大的量子算法尚无运行的载体。
模拟器的作用在于“承上启下”，往下可以帮助理解、设计硬件，向上可以承载算法和应用的探索和验证。“太章”首次使得测试和验证被称为“中等规模”50-200比特的的量子算法成为可能, 从而为辅助设计中等规模量子算法、量子软件乃至量子芯片提供了一个有力的工具。
在通常的量子电路模拟方案中，需要存储量子状态的全部振幅，在此海量数据上同时模拟量子运算。这个方法要求不断地在众多的计算节点间交换数据，造成巨大的通讯开销。因此，过去这样的模拟任务往往都在超级计算机上进行。
实验室团队基于施尧耘教授及其合作者Igor Markov在2005年提出的另一种模拟方案，发明了一个简单而有效的方法分解整个模拟任务，然后十分均衡地把这些子任务分配到不同计算节点上。“太章”的通信开销极小，这个优点使之十分适合分布式的计算平台。
“太章”模拟的随机量子电路规模（黑线）与谷歌量子硬件可以实现的规模(红线) 比较（基于谷歌在[Characterizing quantum supremacy in near-term devices]中对7x7的估计）*
作为基准的随机量子电路是谷歌提出为实现“量子霸权”的算法。“量子霸权”指的是量子处理器的规模和精度到达无法被经典计算模拟的程度。谷歌今年3月份提出了未来工作的目标：72比特高精度的量子处理器。“太章”的结果表明这一计划中的处理器如果只运行该基准算法仍不足于达到量子霸权。
本次研究成果也提交到预印本网站arXiv，文章并列第一作者为量子实验室量子科学家陈建鑫博士与实习生张放，作者还有实习生黄甲辰和Michael Newman博士。
阿里巴巴量子实验室由美国密西根大学终身教授、世界著名量子科学家施尧耘担任首席量子技术科学家、量子实验室主任。两次理论计算机最高奖哥德尔奖得主、匈牙利裔美国计算机科学家马里奥·塞格德（Mario Szegedy）于今年年初也加入该实验室。实验室正处于人才引进的高速增长时期。
2016年，谷歌提出通过实现二维阵列MxN对应的量子比特上的一类特定随机量子电路来实现量子霸权的方案，这一类特定随机量子电路通常被称为量子霸权电路。在方案中，认为当该二维阵列上的比特数(MN)达到50, 电路的深度（层数）到达40左右，现有世界上最强大的超级计算机也无法有效模拟这样的电路。
8x8二维网格上一个深度为20的量子霸权电路对应的张量网络展示
谷歌的硬件团队希望将在9量子比特1维阵列中实现的1%读取误差，0.1%单比特门误差，0.6%两比特门误差保持到更大规模的量子系统来实现这样的霸权电路，并通过这个特定任务，实现量子硬件对当前世界上最强大的经典计算资源的超越。此后，若干研究团队纷纷在不同的超级计算机上对该类电路进行模拟。之前，全球最好的研究结果尚未同时达到50比特40层。
nxn二维网格上，计算随机电路输出每一个振幅的执行时间与电路深度的对应关系
在量子计算目前的模型中，有一类是量子电路模型，实现形式是将信息存储在量子比特中，通过类似经典逻辑门的量子门来实现计算。达摩院量子实验室团队量子科学家陈建鑫与实习生张放实现了一种基于分布式的通用量子电路模拟方案，并基于研究的模拟器对谷歌第一版的随机量子电路进行了测试。
利用阿里计算平台的在线集群的少量计算资源(14%左右)实验室团队成功使用“太章”模拟器模拟了9x9 x40也就是81比特40层随机电路，还分别成功模拟
了100比特35层（10x10x35）, 121比特31层（11x11x31）与144比特27层（12x12x27）的随机量子电路。
目前业界主流的模拟方案有两类，一类是存储量子状态的所有振幅，一类是对于任意振幅都可以迅速计算得到结果。第一类模拟方案，基本都在超级计算机上实现，因为存储45比特的量子状态需要Petabyte量级的内存，在存储这么多数据的同时对该量子态进行操作并进行计算，需要不断地在不同的计算节点之间交换数据，这样的通讯开销对于普通云服务是难以承受的。
在阿里巴巴计算平台的在线集群上，实验室团队采用了第二类模拟方案，通过快速有效的计算任意振幅，任务拆分后可以将子任务十分均衡地分配到不同节点，极少的通信开销使得模拟器适配现在广泛提供服务的云计算平台。
在本研究成果之前，对于两种模拟方案，全球尚未有研究团队可以成功模拟谷歌超过50比特40层的第一代随机测试电路。在达摩院量子实验室团队的模拟器内还可以每2分钟计算64比特40层随机电路的一个振幅。本次研究成果也已经以论文的形式在预印本网站arXiv上提交，文章并列第一作者为量子实验室量子科学家陈建鑫与实习生张放，作者还有实习生黄甲辰和Michael Newman博士。
本次研究成果arXiv论文链接： https://arxiv.org/abs/1805.01450
谷歌、IBM、微软量子霸权混战，施尧耘：超导VS离子阱，量子计算进入两极世界
今年三月，在洛杉矶举行的美国物理学会年会上，谷歌展示了一个新的量子处理器Bristlecone。这个基于门的超导系统目的在于研究量子比特技术的系统误差率和可扩展性，以及在量子模拟、优化和机器学习中的应用。
左边是谷歌最新的72量子比特量子处理器Bristlecone。右边是该设备的图示：每个“X”代表一个量子比特，量子比特之间以线性阵列方式相连。来源：Google Quantum AI Lab
谷歌量子AI实验室研究科学家Julian Kelly在Google Research官博发文介绍，Bristlecone遵循是谷歌之前提出的9个量子比特量子计算机的线性阵列技术所对应的物理学原理，而该技术显示的最佳结果如下：
低的读数错误率（1%）、单量子比特门（0.1%）以及最重要的双量子比特门（0.6%）。
该设备使用与9个量子比特的相同的模式进行耦合、控制和读出，但将其扩展为一个包含72个量子比特的正方形数组。
谷歌研究人员计算后认为，量子霸权的目标可以通过使用49个量子比特，一个超过40的电路深度，一个低于0.5%的2个比特误差进行完美的证明。
有观点认为，谷歌之所以公布72位量子比特处理器，是因为量子霸权之路遇到了对手IBM。
2017年11月，IBM宣布成功构建并测量了具有类似性能指标的50个量子比特原型机，今年2月又曝光了其内部结构图。
50个量子比特被普遍认为可以进行普通超级计算机不能完成的任务，IBM此举也是在“量子霸权”上具有里程碑意义的一步。
因为谷歌和IBM的量子处理器都是通过超导来实现量子计算，因此这两家公司在量子霸权上你追我赶。
但是另外一股力量也随时有可能爆发，那就是微软。
微软押注拓扑量子计算，虽然目前还没有做出来相互作用的量子比特，但一直在开发量子硬件以及量子计算机软件开发套件。
微软的逻辑在于，虽然谷歌、IBM都做出了量子比特，但这些都是不精确的量子比特，来自外部环境的微小震动或能量都可能导致计算错误。
而微软的拓扑量子计算机可能能够大大降低噪音。微软量子计算业务发展总监Julie Love曾说：“我们的一个量子比特将会有1000个、甚至10000个嘈杂的量子比特那样强大。”他们认为，微软将在今年年底前得到可工作的量子比特。
施尧耘今年2月发表在新智元上的一篇文章也提到，宇宙中第一个拓扑量子比特将在今年爆发，而微软是有可能做出来的。
事实也证明，微软也离量子霸权更近了一步。
今年3月底，微软的研究人员观察到被称为“天使粒子”的马约拉纳费米子存在的相当有力的证据：电子在他们的导线中分裂成半体。
如果微软希望建造一台能工作的量子计算机，这将是至关重要的。
另外，离子阱量子计算在今年下半年可能会做出手握50-60几个比特的利器，施尧耘认为，这一领域最大的赢家是没有稳定量子比特的Amazon和Facebook，量子计算将进入两极世界时代。
来源：新浪科技
Translations:
On May 8, the team of Alibaba Quantum Labs announced that it has successfully developed the world’s most powerful quantum circuit simulator called “Taizhang” recently. Based on the powerful computing power of Alibaba Group's online computing platform, "Tai Zhang" successfully took the lead in successfully simulating 81 (9x9) bits and 40 layers of Google's random quantum circuits as benchmarks. Previously, this layer of simulator was only Can handle 49 bits.
Quantum hegemony seems to be playing a "relay battle."
In February, IBM demonstrated its 50 qubit prototypes to the outside world, as well as its internal structure.
In March, Google announced the 72-bit qubit processor Bristlecone.
At the end of March, Microsoft found strong evidence of the existence of the Angel particle, Majorana fermion, and it is expected that the working qubit will be available before the end of the year.
Now it's time for Ali to play.
On May 8, the team of Alibaba Quantum Labs announced that it has successfully developed the world’s most powerful quantum circuit simulator called “Taizhang” recently.
Based on the powerful computing power of Alibaba Group's online computing platform, "Tai Zhang" successfully took the lead in successfully simulating 81 (9x9) bits and 40 layers of Google's random quantum circuits as benchmarks. Previously, this layer of simulator was only Can handle 49 bits.
At the same time, this simulation task only uses 14% of the computing resources of the Alibaba Computing Platform online cluster. "Tai-Zhang"'s innovative algorithm has minimal communication overhead and can take full advantage of the platform's online clustering. In the past, supercomputers could not do simulation tasks such as 64-bit (8x8)-bit 40-layer simulation, "Taizhang" only 2 Minutes to complete.
Comparison of Alibaba "Taizhang" Simulator and Main Simulator Simulating Google Random Circuit
Comparison of Alibaba "Taizhang" Simulator and Main Simulator Simulating Google Random Circuit
Comparing the Scale of Random Quantum Circuit Simulated by "Taizhang" with the Scale Computed by Google Quantum Hardware
Quantum computing may subvert the current computing technology and is a hot topic in the scientific and industrial research. However, the realization of quantum computing is very difficult. At present, the high-precision quantum processor that has been implemented has only a few 20 qubits. Therefore, a slightly larger quantum algorithm has no carrier yet to operate.
The role of the simulator is to "bring the link up and down." It helps understand and design the hardware, and it can carry forward the exploration and verification of algorithms and applications. For the first time, "Tai-Zhang" made it possible to test and verify a quantum algorithm called "medium-scale" 50-200-bit, thus providing a powerful tool for assisting in the design of medium-scale quantum algorithms, quantum software, and even quantum chips.
In a typical quantum circuit simulation scheme, it is necessary to store the full amplitude of the quantum state and simultaneously simulate quantum operations on the massive data. This method requires the constant exchange of data among numerous computing nodes, resulting in a huge communication overhead. Therefore, in the past, such simulation tasks were often performed on supercomputers.
The laboratory team, based on another simulation scheme proposed by Prof. Shi and its collaborator Igor Markov in 2005, invented a simple and effective method to decompose the entire simulation task, and then distributed these subtasks to different computing nodes in a well-balanced manner. . "Taizhang" has very little communication overhead. This advantage makes it very suitable for distributed computing platforms.
Comparison of the size of the random quantum circuit (black line) simulated by "Tai Zhang" and the scale (red line) that Google's quantum hardware can achieve (based on Google's estimate of 7x7 in [Characterizing quantum supremacy in near-term devices]) *
Comparison of the size of the random quantum circuit (black line) simulated by "Tai Zhang" and the scale (red line) that Google's quantum hardware can achieve (based on Google's estimate of 7x7 in [Characterizing quantum supremacy in near-term devices]) *
The random quantum circuit as a reference is Google's algorithm for achieving "quantum hegemony". "Quantum hegemony" refers to the extent to which the quantum processor's size and precision can't be simulated by classical calculations. In March this year, Google proposed the goal of future work: 72-bit high-precision quantum processors. The result of "Tai Zhang" shows that the processor in this plan is still insufficient to achieve quantum hegemony if only the benchmark algorithm is run.
The research results were also submitted to the pre-printed website arXiv. The article was tied with the first author for Quantum Lab Quantum Scientist Dr. Chen Jianxin and the intern Zhang Fang, the author and intern Huang Jiachen and Dr. Michael Newman.
Alibaba Quantum Lab is the lifelong professor of the University of Michigan and Shi Quan, a world-renowned quantum scientist, as the chief quantum technology scientist and director of Quantum Lab. Mario Szegedy, a Hungarian-American computer scientist who was also the recipient of the two highest computer science awards for the Gödel Prize, joined the lab earlier this year. The laboratory is in a period of high-speed growth in talent introduction.
In 2016, Google proposed a scheme for realizing quantum hegemony by implementing a specific random quantum circuit on a qubit corresponding to a two-dimensional array MxN. This type of specific random quantum circuit is often called a quantum hegemonic circuit. In the scheme, it is considered that when the number of bits (MN) on the two-dimensional array reaches 50, the depth (number of layers) of the circuit reaches about 40, and the most powerful supercomputer in the world today cannot effectively simulate such a circuit.
A tensor network demonstration of a quantum hegemonic circuit with a depth of 20 on an 8x8 two-dimensional grid
A tensor network demonstration of a quantum hegemonic circuit with a depth of 20 on an 8x8 two-dimensional grid
Google’s hardware team hopes to achieve such a hegemonic circuit by maintaining a 1% read error in a 9-qubit 1D array, a 0.1% single-bit gate error, and a 0.6% 2-bit gate error to a larger scale quantum system. And through this particular task, Quantum hardware is surpassed by the world’s most powerful classic computing resources. Since then, several research teams have simulated these circuits on different supercomputers. Previously, the world’s best research results have not reached 50-bit and 40-story at the same time.
Calculating the relationship between the execution time and the circuit depth of each amplitude output of the random circuit on the nxn two-dimensional grid
Calculating the relationship between the execution time and the circuit depth of each amplitude output of the random circuit on the nxn two-dimensional grid
In the current model of quantum computing, there is a quantum circuit model in which the information is stored in qubits and calculated by a quantum gate similar to a classical logic gate. Quantum scientist Chen Jianxin of the Boomerang Quantum Lab team and intern Zhang Fang implemented a distributed universal quantum circuit simulation scheme and tested the first random quantum circuit of Google based on the research simulator.
A small amount of computational resources (around 14%) using online clustering of Alibaba Computing Platform successfully used the "Taizhang" simulator to simulate a 9x9x40, which is an 81-bit 40-layer random circuit, and successfully simulated separately.
A 100-bit, 35-layer (10x10x35), 121-bit, 31-layer (11x11x31) and 144-bit, 27-layer (12x12x27) random quantum circuit.
Currently, there are two main types of simulation schemes in the industry. One is to store all the amplitudes of the quantum state, and the other is to quickly calculate the result for any amplitude. The first type of simulation schemes are basically implemented on supercomputers because storing 45-bit quantum states requires Petabyte-level memory. The operation and calculation of the quantum state while storing so much data need to be continuously different. The exchange of data between computing nodes, such communication overhead is unbearable for ordinary cloud services.
On the online cluster of Alibaba's computing platform, the laboratory team used the second type of simulation program to quickly and efficiently calculate arbitrary amplitudes. After task splitting, the sub-tasks can be distributed evenly to different nodes with very little communication overhead. The simulator is adapted to a cloud computing platform that is now widely available for service.
Prior to the results of this study, no research team in the world has been able to successfully simulate the first generation of random test circuits with more than 50 bits and 40 layers for Google. An amplitude of 64-bit 40-layer random circuits can also be calculated every 2 minutes in the simulator of the Dharma Quantum Lab team. The research results have also been submitted in the form of papers on the pre-printed website arXiv. The first author of the paper is Quantum Lab Quantum Scientist Chen Jianxin and Intern Zhang Fang, and the author and intern Huang Jiachen and Dr. Michael Newman.
The results of this research arXiv paper link: https://arxiv.org/abs/1805.01450
Google, IBM, Microsoft Quantum Hegemony, Shi Zheng: Superconductivity VS Ion Trap, Quantum Computing Enters the Polar World
In March of this year, at the annual meeting of the American Physical Society in Los Angeles, Google demonstrated a new quantum processor Bristlecone. The purpose of this gate-based superconducting system is to study the systematic error rate and scalability of quantum bit technology, and its application in quantum simulation, optimization, and machine learning.
On the left is Google’s latest 72-bit quantum quantum processor, Bristlecone. On the right is a diagram of the device: Each "X" represents a qubit, and the qubits are connected in a linear array. Source: Google Quantum AI Lab
On the left is Google’s latest 72-bit quantum quantum processor, Bristlecone. On the right is a diagram of the device: Each "X" represents a qubit, and the qubits are connected in a linear array. Source: Google Quantum AI Lab
Julian Kelly, a research scientist at Google Quantum AI Labs, presented in the article published by Google Researcher Bristlecone follows the physics principle of the linear array technology of the nine quantum-bit quantum computers proposed by Google, and the best results shown by the technology are as follows: :
Low reading error rate (1%), single qubit gate (0.1%), and most important double qubit gate (0.6%).
The device uses the same pattern as the nine qubits for coupling, control, and readout, but expands it to a square array of 72 qubits.
Google researchers calculated that the goal of quantum hegemony can be perfectly demonstrated by using 49 qubits, a circuit depth of more than 40, and a 2-bit error of less than 0.5%.
There are views that Google announced the 72-bit qubit processor because the quantum hegemony road encountered rival IBM.
In November 2017, IBM announced the successful construction and measurement of 50 qubit prototypes with similar performance specifications. In February this year, it also exposed its internal structure.
The 50 qubits are generally considered to be tasks that cannot be accomplished by ordinary supercomputers. IBM's move is also a milestone step in Quantum Hegemony.
Because both Google and IBM's quantum processors implement quantum computing through superconductivity, the two companies are chasing after one another in quantum hegemony.
However, there is also the possibility that another force may erupt at any time. That is Microsoft.
Microsoft is betting on topological quantum computing. Although no interworking qubits have yet been made, quantum hardware and quantum computer software development kits have been developed.
The logic of Microsoft is that although Google and IBM have all made quantum bits, these are inaccurate qubits, and tiny vibrations or energy from the external environment can cause calculation errors.
Microsoft's topological quantum computer may be able to greatly reduce noise. Julie Love, the development director of Microsoft's quantum computing business, once said: "One of our qubits will be as powerful as 1,000 or even 10,000 noisy qubits." They think that Microsoft will get a working qubit before the end of this year.
An article published by Shi Zhi in February this year on Xinzhiyuan also mentioned that the first topological qubit in the universe will explode this year, and Microsoft is likely to make it.
The facts also prove that Microsoft is a step closer to quantum hegemony.
At the end of March of this year, Microsoft researchers observed the strong evidence of the presence of Mayorana Fermi, known as the "Angel Particle," that electrons split into halves in their wires.
If Microsoft wants to build a working quantum computer, this will be crucial.
In addition, ion trap quantum computing may be able to hold 50-60 bits of weapon in the second half of this year. Shi Yong believes that the biggest winner in this field is Amazon and Facebook without stable quantum bits. Quantum computing will enter the bipolar world. era.
Source: Sina Technology
