Japanese IT giant's semiconductor quantum chip is about to be mass-produced, and Intel is following closely in a different way丨Semiconductor quantum computing is on the eve of commercialization

This article is authorized to be reproduced from DeepTech (ID: mit-tr)

Not long ago, the industry still believed that semiconductor quantum technology was still far away from us. Even D-Wave CEO Vern Brownell said in an exclusive interview with DT that it would take at least 10 years to commercialize semiconductor quantum computing. However, with the Japanese IT giant Fujitsu's Digital Annealer quantum computing chip about to go into mass production, and Intel's breakthrough in silicon spin qubit technology, and China also demonstrating the development of semiconductor quantum computing, quantum computing may be able to enter ordinary people's homes earlier through semiconductor technology.

Obviously, quantum computing is important because it has the ability to quickly solve "human-scale" problems that were difficult to solve using traditional computing architectures in the past, such as finding a solution to cancer and better personalized medical methods. It will play an extremely important role not only in the energy field, the most popular AI simulation, but also in uncovering more secrets of the universe.

The quantum physics phenomena that serve as the basis of quantum computing are actually universal physical phenomena in nature and can occur in many different materials, chemicals or natural environments. Therefore, there is more than one way to achieve them. For example, the research scope of quantum computing has advanced from superconducting quanta to photonic quanta, and even semiconductor quanta based on digital annealing technology have been mass-produced. In other words, as long as the phenomena caused by the material can observe quantum physical characteristics, it is possible to use them for calculations.

However, after nearly 20 years of development, the commercialization of quantum computing based on superconducting technology is still greatly limited in practical application due to the insufficient maturity of the software ecosystem and the extremely high difficulty of mass production. Although we argue about who will hold quantum supremacy based on the scale of each company's quantum bits, in fact, the biggest limitation of quantum computing is not the lack of computing power, but the difficulty in popularizing it, which makes it difficult for the ecological development to move forward effectively.

Also because of the current limitations of quantum computing, if the existing semiconductor production technology can be used to solve the problems of scale expansion and mass production of quantum computing chips, and get rid of the huge cooling architecture required for superconducting quantum computing, then quantum computing may be able to enter general computing applications earlier than expected, and accelerate the maturity of the related ecosystem, becoming the computing core of PCs, smart homes, cars, and even various networked devices, and completely changing human life.

At present, in terms of the progress of quantum technology in the semiconductor field, the more well-known ones include the Digital Quantum Annealer technology launched by Fujitsu, the silicon spin qubit technology based on silicon semiconductor technology launched by Intel, and the three-quantum dot semiconductor bit based on quantum dot technology proposed by the Key Laboratory of Quantum Information of the Chinese Academy of Sciences.

• Digital quantum annealing technology

Fujitsu has collaborated with the University of Toronto in Canada to develop the Digital Annealer as an alternative to D-Wave’s quantum annealing computing architecture, which requires a carefully controlled low-temperature environment to function. Fujitsu uses conventional semiconductor technology that works at room temperature and can be mounted on a circuit board small enough to be inserted into a data center rack.

Figure｜The digital annealing quantum processor based on semiconductor technology is currently used in Fujitsu's cloud services.

The digital annealer is a specialized chip that uses a non-von Neumann architecture to minimize data movement when solving combinatorial optimization problems. It consists of 1024 "bit-updating blocks" with on-chip memory to store weights and biases, logic blocks to perform "bit flips," and interface and control circuitry.

Instead of using the computing power of the digital annealer through traditional programming, problems are uploaded in the form of weight matrices and bias vectors in order to transform the problem into an "energy landscape" and solve it using phenomena simulated by physics. To achieve this goal, Fujitsu has partnered with 1QB Information Technology, a leader in quantum computing software based in Vancouver, Canada, which provides both the software to run the system and a software development kit for customers to write their own energy landscapes.

Hidetoshi Nishimori, professor of physics at Tokyo Institute of Technology and co-author of the world's first paper proposing the theory of quantum annealing, explains this operation metaphorically: "In digital annealing, the system jumps from one state to another in search of a better solution, like a person wandering in a complex landscape full of hills and valleys, looking for the lowest point."

Nishimori added that this technology is the opposite of traditional quantum annealing, in that the system searches for the best solution in a massively parallel manner, taking all states into account at the same time. Fujitsu also claimed that its CMOS-based digital annealer, although only equipped with 1024 qubits, has performance comparable to D-Wave's latest 2000 qubit quantum annealing system.

Nishimori pointed out that the weights between blocks of bits on the Fujitsu machine can express problems with higher precision than the D-Wave system, because it is much more difficult to control this precision with quantum bits. The digital annealer has 16 bits of precision between bits, while the D-Wave system has only 4 bits of precision. However, Nishimori also mentioned that D-Wave's quantum annealers will have the potential to surpass digital annealers in the long run because they have a large amount of quantum parallelism, which is enough to make up for the shortcomings of weaker precision performance.

At the same time, Fujitsu said that the 1,000-qubit solution is currently being used on its own cloud servers, and its goal is to mass-produce digital annealers with 8,192 qubits in 2019, while its long-term goal is to move towards millions of qubits.

The company began offering cloud services in Japan on May 15. Fujitsu is also working with the University of Toronto to research the application of digital annealing machines, and later this year Fujitsu will begin selling digital annealing servers, tower hosts and chips for in-house quantum computing. The company also plans to launch cloud services in North America, Europe and Asia by the end of this year. Fujitsu said it aims to generate 100 billion yen (about $900 million) in revenue for the service by 2022.

• Silicon spin qubits

Intel and Dutch quantum computing company QuTech cooperated to launch programmable dual quantum computing based on silicon chips at the beginning of this year, which uses spin quantum units. The advantage of spin quantum units is that they do not require harsh environmental conditions, such as extremely low temperatures. In essence, spin quantum units are electrons activated by microwave pulses. The unique advantage of silicon-based spin quantum units is that they operate at the electronic level, so they can work closely with existing computing work platforms.

Figure ｜ A quantum processor with two qubits on a silicon chip.

In fact, the concept of silicon spin qubit is very simple. When a stable current passes through a traditional transistor, a single electron in the transistor can switch between the two states of 0 and 1. In terms of the spin of the electron, a single electron in the transistor can have one of two states: spin up or spin down, which are the two states of the qubit. Therefore, what Intel is doing is mainly creating a series of single-electron transistors through its process and making them produce quantum states.

However, Intel is still working on integrating more qubits on a single chip. Currently, they can only maintain 26 qubits per single chip, which is obviously far from superconducting qubits. However, if we use the whole wafer as a comparison benchmark instead of a single chip, the number of silicon spin qubits has reached tens of thousands, and the qubit density is no less than that of traditional superconducting qubits.

Figure | Intel's silicon quantum chip trial product

Of course, it is not practical to use wafers as a unit of comparison. Intel also stated that the company's superconducting quantum technology is mature enough to be integrated into the system, but silicon spin quantum will take several years to develop.

However, Intel also mentioned that, taking the history of processor development as an example, it took 10 years from the first integrated circuit to the first processor 4004 with 25,000 transistors. In fact, the progress was quite fast. They also see the future development potential of silicon spin quantum and believe that it will not be difficult to develop to a single chip with more than 1,000 quantum bits within 5 years.

In the long run, if we can achieve millions of quantum bits on a single semiconductor chip and realize universal quantum computing at room temperature, it will be a radical change to the entire digital industry and even the social existence mode.

• Three quantum dot semiconductor bits

The three-quantum-dot semiconductor bit launched by the Key Laboratory of Quantum Information of the Chinese Academy of Sciences at the beginning of this year is one of the technical applications of quantum dots. It is a solid-state computing method belonging to many quantum computing types. It is mainly made of GaAs or AIGaAs or similar materials to make quantum dots. Quantum dots refer to the confinement of electrons and holes in extremely small substances of only a few nanometers, thus producing controllable light, electricity, spin and other properties. Usually these properties are related to the size, shape and material of the quantum dots. When light is in the form of quantum dots, there are different spin quantum states such as photons, electrons and atoms that can be used as the basis of quantum computing.

Figure ｜ The three-quantum dot semiconductor bit logic gate launched by the Key Laboratory of Quantum Information of the Chinese Academy of Sciences at the beginning of this year.

The experiment used semiconductor nanofabrication technology to prepare an asymmetric coupled three-quantum dot structure, and then used the atomic shell structure filling principle of electrons to resolve the complexity of the multi-electron energy level structure and construct a hybrid quantum bit with quasi-parallel energy levels. While ensuring the bit coherence time, by adjusting the electrode voltage of the third quantum dot, it was clearly observed that the bit energy level was continuously adjustable in the range of 2 to 15GHz.

However, although this technology is a type of semiconductor quantum technology, its semiconductor material is not silicon, so it is quite different from existing semiconductor processes.

Quantum bits based on superconducting circuits and quantum well-based quantum bits are in the leading position in terms of the number of controllable quantum bits because of their large circuit size and relatively easy implementation. Currently, Google has reached 72 quantum bits, and Intel and IBM have also launched quantum computing architectures with 49 quantum bits and 50 quantum bits, respectively.

However, their large size makes the integration of large numbers of quantum bits in the future very problematic, which in turn will affect the implementation of some quantum algorithms for practical applications.

Although all existing superconducting quantum computing methods can provide unprecedented computing power, the technology requires extremely high design and maintenance costs: in order to achieve correct output for problems beyond the scope of traditional computing, superconducting quantum computing needs to be kept close to absolute zero, and various shielding designs are used to avoid the influence of magnetic interference, thermal noise and mechanical vibration, so that the quantum bits can maintain superposition state and quantum entanglement, which becomes the basis for realizing quantum computing.

Due to the instability of quantum bits, the accuracy of quantum computing is also problematic. Generally speaking, the fidelity is generally not high, which means that the existing quantum computing architecture must spend a lot of effort on error correction to ensure that there are no errors in the observation of quantum phenomena. This also makes the already bloated quantum computing architecture even larger.

These traditional quantum computing problems are still quite difficult to solve even though the laboratory has already exceeded 72 qubits. This is why Microsoft chose to use Majorana particles based on topological architecture as the core of quantum computing. Since Majorana particles are electrically neutral and rarely interact with other particles, their state is relatively stable. This means that to build a 1Qubit topological quantum computer, only one Majorana particle is needed, without the need for additional error correction design. In theory, this will be a very competitive quantum computing architecture.

However, this Majorana particle, also known as the angel particle, still exists only in theory and has not yet been discovered.

Is it impossible to create a quantum computing architecture that is both small and easy to scale? No, as mentioned before, Fujitsu and Intel have already made breakthroughs in related fields. The Key Laboratory of Quantum Information of the Chinese Academy of Sciences, led by Academician Guo Guangcan of the University of Science and Technology of China, is promoting the semiconductorization of quantum computing through quantum dot technology in the research of semiconductor quantum computing chips.

Since the digital annealing technology based on physical simulation is more mature in theory and its application is more focused on specific calculations rather than general purposes, its commercialization is moving the fastest, just like the quantum annealing of its big brother D-Wave. However, its disadvantage is that the entire computing system is not compatible with the current computing architecture, making it more difficult to operate the ecosystem.

Intel is planning to bring silicon spin quantum into the general daily computing environment as a powerful weapon to replace the traditional computing architecture, so universality is a point that cannot be abandoned. However, because of this, it may take several years for it to be truly commercialized.

Putting aside the current controversy over quantum supremacy, traditional superconducting quantum computing has already entered commercialization, and quantum computing based on semiconductor technology has also exceeded market expectations and is beginning to emerge in technology research and development. Although, with the exception of the oldest annealing technology, the commercialization schedule for other technologies aimed at general quantum computing may take several years.

Even so, the progress of these technologies reveals that the computing architectures we use in our daily lives may be moving towards a possible future of quantum computing. Although it is still difficult to imagine at present, it is certain that with the huge computing power brought by quantum computing, the world we live in will be very different from now.

"Xinshiye's media"