Chinese researchers built a RISC-V processor with artificial intelligence in less than 5 hours

A group of Chinese researchers say in Less than five hours have succeeded in designing a new industrial scale processor with RISC-V architecture. They claim that this process is about 1000 times faster than the same processor designed by humans, but its power is equal to an old Intel i486 processor.

A group of Chinese researchers wanted to see if artificial intelligence could produce chips like humans. To test this idea, they first trained an AI model with a set of CPU input and output data. Then one of these trainings Binary Speculation Diagram (BSD) It was built and after the development principles of “Monte-Carlo” and Boolean functions were used to improve the accuracy and efficiency of this processor.

Scientists say that with this method, they have been able to create a processor with an unparalleled level of accuracy through the 65 nm production process. 99.99999999999 percent arrives. The maximum frequency speed of this processor is 300 MHz.

What is the function of artificial intelligence manufacturing processor?

This processor has appeared successful in tests and has demonstrated its performance with the Linux operating system (kernel 5.15). This product performs similar to the processor in Drystone benchmarks i486 had Additionally, this processor seems to perform slightly better than the Acorn Archimedes A3010.

Researchers say that artificial intelligence has been able to discover the von Neumann architecture from scratch by itself, and this is very surprising. The construction of this processor can be important not only from an academic point of view, but also from the point of view of producing new processors.

Researchers say artificial intelligence can reduce time and costs in the semiconductor industry. Additionally, the question has been raised as to whether this technology could be the basis for building a device that can constantly improve itself. The results of the researchers’ findings have been published on the arXiv website.

Another group of scientists had previously been able to design part of a chip just by talking to ChatGPT.