The AI "singularity" is approaching, the failure controversy of Moore's Law that lasted for several years has suddenly disappeared, and under the tide of technology, people seem to have reached a certain degree of tacit understanding. Replacing Moore's Law is an industry-circulated AI computing power growth curve, which shows that about every 4 months, the computing demand of AI doubles, and then the discussion of the computing power gap gradually heats up.
At the 2023 Zhongguancun Forum, Beijing Quantum Institute officially released the "Quafu" quantum computing cloud platform, which can control the properties of up to 136 qubits at the same time, making "Quafu" the largest cloud platform with the earliest single-chip bit count in China. Ten days before the release of the cloud platform, quantum computing startup Bose Quantum just released a 100-qubit coherent optical quantum computer, named "Tiangong Quantum Brain".
Artificial intelligence and quantum computing, under the "matching" of computing power, have converged a huge imagination space - will quantum computing become the ultimate solution to computing power in the era of artificial intelligence?
When quantum computing goes to the cloud
The quantum information exhibition area of the cutting-edge information technology booth of the 2023 Zhongguancun Forum left the "C bit" to the quantum computing cloud platform.
"The number of bits shown here is 10 bits, 18 bits, 136 bits." In front of the booth, Yang Zhipeng, a postdoctoral fellow in the quantum computing cloud platform team of Beijing Quantum Institute, from time to time wants to show the operation process of the cloud platform for interested audiences, different from the complex operation in the laboratory, where he only needs to connect to the Internet and click the mouse.
Wang Zhengan, a founding member of the "Quafu" team and a postdoctoral fellow at the Beijing Institute of Quantum Information Science, explained that in practice, superconducting quantum computers are a very complex set of machines, and there are many specific problems to be solved in use. "For example, before conducting an experiment, the entire equipment needs to be calibrated to a relatively good operating state, which may take days or even months. After a period of time, due to the influence of ambient noise, its state will gradually deteriorate, and finally it will reach an unusable state, at which time it will have to be recalibrated. In addition, we have done a lot of work in the manipulation of superconducting quantum computers, the compilation of quantum programs, and the correction of errors. He said.
"The most important thing is that only the equipment is not enough, professional personnel are needed to operate, which involves a lot of professional knowledge and experience, and such talents are also scarce in China." Wang Zhengan said that doing a good job in the complex basic parts of quantum computers so that more people in need can use them through simple operations is also one of the original intentions of the cloud platform project.
Yang Zhipeng introduced: "The quantum computing cloud platform means that users only need to connect to our server through the network, and the server can call the quantum computing device and return the corresponding calculation results, so that researchers can more conveniently use the resources of quantum computing to study their problems." ”
At present, the "Quafu" quantum computing cloud platform has been open to users for testing, and you only need to register an account to use it for free. It is reported that the number of registered users of the cloud platform has exceeded 2,000, including universities, research units, enterprises and enthusiasts.
Behind the simplification of operations, the more practical problem is the cost. It is understood that the low temperature environment required by quantum chips and the supporting measurement and control equipment are expensive, and sharing quantum resources in the form of cloud platforms and providing research means is a relatively simple way to solve high costs.
In Wang Zhengan's view, for the public, the cloud platform mainly plays a guiding role, providing quantum real machines, so that people have more real feelings and experiences of quantum computing; For researchers in research institutes, universities and enterprises, the cloud platform is a cooperative role, which can provide quantum resources to help scientific research on quantum computing. "These two parts complement each other, and the ecology of quantum computing can gradually grow."
Never shy away from talking about "money"
Ecology is a signal. When a technology begins to go out of the lab and talk about ecology, it may no longer be a distant thing to be far from real application. Quantum computing also needs to be "under the ordinary", and the possibility of empowering the industry has skyrocketed.
Quantum computing has amazing potential. Unlike the binary mode in classical computing, which is always in a definite state of 0 or 1, quantum computing can not only contain 0 and 1, but also contain 0 and 1 superposition states at the same time, which also means that a quantum computer with n qubits can be in the nth power of 2 at the same time, and the computing power shows exponential growth.
At present, the quantum computing technology route shows a hundred flowers, the main technical paths include superconductivity, ion trap and optical quantum, etc., in terms of superconductivity, the layout is mainly IBM, Google, Intel and domestic origin quantum, national shield quantum, etc., ion trap route includes Honeywell and domestic Qike quantum.
Bose quantum selection is more "trusting in light". According to reports, Bose Quantum's self-developed "Tiangong Quantum Brain" has 100 computational qubits, which has reached the international leading level, and after dozens of practical case tests, its solution speed is more than 100 times faster than the classical optimization algorithm, and the average solution result is also 120% better than the classical algorithm.
At the same time, the "Tiangong Quantum Brain" also realizes the "full connection" control between hundreds of optical quanta, with complete programmability, that is, corresponding to different application scenarios and different algorithms, the hardware does not need to be modified, and it can be expanded and programmable through software configuration. Chen Liang, vice president of Bose Quantum, believes that this is an important milestone in the development of quantum computing from theoretical quantum superiority to practical quantum advantage.
Chen Liang listed many advantages of optical quantum routes, including long coherence time, strong ability to resist external interference, and the number of qubits is easier to achieve ultra-large-scale, but the most fundamental reason is the feasibility of optical quantum routes in terms of commercialization.
For example, the ultra-low temperature superconducting environment near absolute zero is not required, and the conditions that can operate at room temperature greatly reduce the cost, and the equipment of this route can also upgrade and reuse the existing semiconductor and optical communication equipment production lines, without the need to build new production lines and processing processes, which greatly reduces the threshold for commercialization.
In the process of communication, Chen Liang never shy away from the problem of "money", he bluntly said that Bose Quantum emphasizes the development of "practical" optical quantum computing platform and commercial application landing, that is, to do "practical, usable and useful" quantum computers, the choice of optical quantum route, is also for such considerations.
But for the entire quantum computing field, the arrival of true commercialization may not be so soon. Wang Zhengan mentioned that once the parallel computing power brought by quantum superposition and entanglement is well utilized, classical computers will be completely incomparable. However, there are still very few algorithms that can fully utilize this quantum computing power, and the practical problems that can be solved are now very limited.
"The difficulty at the experimental level is that quantum states are very fragile, and to achieve fault-tolerant quantum computing, sufficient manipulation accuracy and a large number of redundant qubits are required for error correction." Wang Zhengan said that the integration of large-scale qubits and the improvement of their computational fidelity are the main goals of quantum computing.
Who is the Intel of the quantum world
The huge potential of quantum computing also forms a closed loop with the "computing power shortage" caused by the current artificial intelligence competition. In April this year, due to excessive demand, ChatGPT Plus sales were suspended for a while, and the discussion of computing resources in short supply was instantly detonated.
OpenAI's report has shown that training a GPT-3 model with 174.6 billion parameters requires a computing power of about 3640PFlop/s-day, that is, if it is calculated 10 trillion times per second, it also needs to be calculated for 3640 days, corresponding to a training cost of more than 12 million US dollars, and the GPT-3.5 model that has undergone three iterations of GPT-1, GPT-2, and GPT-3 has increased from 117 million to 175 billion, and the amount of pre-training data has increased from 5GB to 45TB.
The same is true in China, before the artificial intelligence large model ignites the explosion of computing power, including automatic driving, meta-universe and other fields of technology research and development has put forward higher requirements for computing power. According to data from the Academy of Information and Communications Technology, from 2016 to 2021, the scale of computing power in mainland China increased by an average of 46% per year, and up to now, the scale of computing power industry represented by computers has reached 2.6 trillion yuan, directly driving the total economic output of 2.2 trillion yuan.
Public information shows that computing power has become the core productivity of the digital economy, and the diversity of data and applications drives different computing architectures to go hand in hand. General-purpose computing technology routes such as ARM and RISC-V have been fully developed, heterogeneous intelligent computing platforms such as GPUs, NPUs, and DPUs have prospered, and the diversity characteristics of computing power have been widely revealed.
Quantum computing may become an important part of breaking the computing power shortage. In February this year, a research report by Huaan Securities showed that ChatGPT has brought changes in the era of large models, and the construction of computing power and network facilities has become a rigid demand. The quantum computing industry is booming. Compared with classical computers, computing power has grown exponentially explosively. At present, quantum computers have been shown to have exponential acceleration capabilities on specific computing tasks, that is, to achieve so-called "quantum supremacy".
Wang Zhengan also said that the combination of quantum computing and artificial intelligence is one of the current research hotspots in the field of quantum computing, because artificial intelligence to break through the bottleneck is likely to require the computing power provided by quantum computing beyond the classical way.
"Computing power is to the era of artificial intelligence, just as land is to the era of agriculture, oil is to the era of internal combustion engines, the future technology competition is indispensable to the battle of computing power, and quantum computing can bring a leap forward in computing power, which is one of the core resources of artificial intelligence competition." Mi Lei, founding partner of China Science and Technology Innovation Star, judged that once quantum computing achieves a breakthrough, the market value of quantum computing companies that stand out is the current Intel, NVIDIA and other companies.
China Science and Technology Innovation Star also pays close attention to cutting-edge technologies in the field of quantum science and technology, and has invested in two quantum computing companies, Origin Quantum and Zhongke Kuyuan. Among them, Origin Quantum is the first quantum computing company in China and the first company in China to develop an engineered superconducting quantum computer and a quantum computer operating system.
However, from the stage of technological development, the field of quantum technology is still in its infancy. Mi Lei believes that investing in quantum technology requires patience, a ten-year cycle layout, and a long-term plan. At present, there is still a gap between the number and quality of patents in quantum science and technology in the mainland than that of European and American countries, and some domestic institutions and enterprises still need to continue to invest in this field.
Wang Zhengan mentioned that from the current trend, quantum computing needs to develop together with classical computing methods. That is, quantum computing may first appear as a means of assisting classical computing in the future, the so-called quantum coprocessor. Only then will there be a pure quantum "killer" application, when quantum computing is the dominant position, and the classical means are auxiliary.
Beijing Business Daily reporter Yang Yuehan