(16th-December-2019)
Today, I am talking about the new computer architecture for finding the best combination. Quantum computer does calculation to find out the best combination pares very fast compare traditional Von Neumann type architecture computer. Then, some of manufactures are doing challenges. Following article is come from Fujitsu. They are The first is “a technology that minimizes data movement by non-Neumann-type processing”.
Uses non-Neumann type processing, which is different from Neumann type (program execution type), and updates the set of optimization variables (bits) so that the evaluation value of the optimization problem decreases. To perform an operation, first load the problem from memory, then perform as many optimization operations as necessary, and finally read the result. Since there is no read / write to the memory during the operation, the time and energy loss are minimized. Also, by suppressing data movement between basic circuits, data movement to upper layers is almost eliminated.
Technology to minimize data movement with non-Neumann operations
The second is “high-speed technology in the basic optimization circuit”. In the basic optimization circuit, the search from a certain state to a more optimal state is repeatedly performed using a probabilistic method. The probability of finding the next state is improved by calculating the values of each evaluation result for multiple next candidates in parallel. If you become stuck in the middle of the search, you can make it easier to move to the next state by repeatedly adding a certain value to the evaluation value to increase the escape probability. This makes it possible to find the optimal solution at high speed.
Basic optimization circuit acceleration technology
The “Neumann type” is a calculation method in which a program is stored in a storage device as data, and the program is sequentially read and sequentially executed. Most computers are currently used. In recent years, due to dramatic improvements in computer performance, the weakness of reading instructions from memory has become more prominent, and the “non-Neumann type” using another basic design has been considered. Neurocomputers that use cranial nerve models as models, quantum computers that apply the behavior of elementary particles in quantum mechanics, and DNA computers that use DNA as computing elements are non-Neumann types.
Compared to conventional software processing, it operates approximately 10,000 times faster
This time, when a basic optimization circuit that can handle combinations represented by 1024 bits was installed in an FPGA and evaluated, it operated at a speed about 10,000 times faster than software processing that runs on a conventional processor.
Comments