Quantum-Inspired Reconfigurable Processor

量子计算机有望以高速解决NP问题，例如因子分解，因为它可以同时验证所有候选解。然而，量子计算机的最终系统仍不清楚。要实现实际操作，不仅需要硬件，还需要软件进行算法和基准测试。因此，量子计算仿真器是必要的，以检查算法。然而，大规模量子算法的仿真是困难的，因为在量子计算机中利用量子叠加，并且需要2^N个存储器来仿真N个量子比特。通过用经典位替换部分量子位并使用量子宏操作，我们提出了一种新的量子计算仿真器。在这个模拟器中，通过只关注核心计算部分，我们为一些计算过程分配了经典位，并使用酉宏操作进行频繁的量子操作。因此，我们成功地减少了内存使用。在实现上，我们在40 MHz时钟的FPGA上实现了192个并行PE的双指令多数据（DIMD）处理器，并设计了一个C语言软件控制处理和宏操作的运算转换器。在这个量子计算机仿真器中，我们用量子计算机中著名的Shor算法对一个64位整数进行了因式分解实验。虽然这种计算相当于在量子计算机上使用320量子比特的计算复杂度，但作为实验的结果，1,300秒用于计算，并且这种仿真只需要40 k比特的存储器

A quantum computer is expected to solve the NP problem, such as factorization, at high speed, since it can verify all the solution candidates simultaneously. However, the final system of the quantum computer is still unclear. It is noted that not only hardware but also software with algorithm and benchmark tests is indispensable to realize practical operation. Hence, a quantum-computing emulator is necessary to inspect the algorithm. However, emulation of large-scale quantum algorithm is difficult, since quantum superposition is utilized in the quantum computer and 2^N memories are necessary to emulate N qubits. By replacing some qubits with classic bits and using quantum-macro operations, we have proposed a new quantum-computing emulator. In this emulator, by focusing on only a core calculation part, we have assigned classical bits for some calculation process and using unitary-macro operations for frequent quantum operations. As a result, we have succeeded to reduce memory usage. In implementation, we have fabricated a DIMD (dual-instruction multiple-data) processor with 192 parallel PE on FPGA with 40MHz clock and an operation converter which is a C-language software controlling processing and the unitary-macro operations. In this quantum-computing emulator, we have made an experiment on the factorization with a 64 bits integer by using Shor's algorithm, which is well-known algorithm for the quantum computer. Although this calculation is equivalent to computational complexity to use 320 qubit for on the quantum computer, as a result of experiment, 1,300 seconds are used for computation and only 40k-bit memory is required for this emulation