The development progress of quantum computers in the world is much faster than we perceive it in Japan. Starting this year, the technology of quantum computer computation will grow by leaps and bounds. IBM has recently announced the introduction of a 4000+ qubit quantum gate machine in 2025. How will such machines be evaluated and applications made? In fact, quantum supremacy, proposed by Google in 2018, has been overtaken again by classical computers three years later. The technology behind this is tensor networks. To solve the mysteries of physics, quantum computation is reduced to the form of nodes called tensors. The exact simulator called a state vector simulator, which has been considered common sense in the quantum computer industry, is close to a brute force calculation and contains a lot of unnecessary information. By limiting the simulation to the necessary information, quantum computer applications can now be made faster and applications can be created. This has made it possible to perform quantum computer calculations with hundreds to tens of thousands of qubits of quantum gate calculations, whereas previously 40 or so qubits were considered to be the maximum. Deep learning technology supports this tensor network technology. This major breakthrough in quantum computing is supported by the latest AI technology, which we now know can be shared through calculations called tensors. Google&#39;s machine learning framework is called tensorflow. This tensor computation is the heart of deep learning, and this huge investment and technology in machine learning can now be directly incorporated into quantum computers. Parallelization processes such as jax pytorch tensorflow cupy are actually starting to pay off in quantum computation in quantum computers in a big way in application development. NVIDIA&#39;s newly released cuQuantum has achieved quantum gate quantum computation at 1688 qubits, which significantly breaks the previous world record for quantum computer simulations, and now performs combinatorial optimization of over 1000 qubits on a quantum gate machine. In addition, QAOA with 289 qubits will be performed on an actual Rydberg atom quantum computer, which is scheduled to appear this year, and evaluated on a tensor network. Naturally, it is the hardware that supports these new technologies, and the know-how and computational power of deep learning, which has been driven by GPUs, will flow into the quantum computer industry, and we can expect significant growth and development.

The Only Reason Why GPUs Will Be Required for Quantum Computer Computations Starting in 2022

Yuichiro Minato