A team of Chinese researchers has successfully utilized artificial intelligence (AI) to design an industrial-scale RISC-V CPU in under 5 hours. The resulting design boasts an impressive 99.99999999999% accuracy and runs at a speed comparable to that of an i486, a chip that debuted in April 1989. The automated AI design process even 'discovers' the von Neumann architecture from scratch, which could have practical applications in the semiconductor industry.
AI Designing an Entire CPU
In a remarkable feat of AI surpassing human intelligence, Chinese researchers at the State Key Laboratory of Computer Architecture (SKLCA) have employed machine learning to design a fully functional CPU. The research team claims that this achievement is approximately 1000 times faster than what a human team could have accomplished. The result of their efforts is a taped-out RISC-V32IA instruction set CPU, fabricated at 65nm and capable of running at up to 300 MHz. Furthermore, the researchers emphasize that their generated Binary Speculation Diagram (BSD) "discovered the von Neumann architecture from scratch."
The Speed of the AI Design Process
While the performance of the generated CPU is on par with an i486, the true significance of this research lies in the speed at which the entire design process was completed. The researchers, led by director Min Pu and colleagues at SKLCA, aimed to determine if machines can design like humans. "Previously, the machines designed by our team were relatively small," explain the researchers. "We wanted to push the boundaries by designing a RISC-V CPU."
The process involved a period of machine learning during which the AI observed CPU inputs and outputs. This data was used to generate a Binary Speculation Diagram (BSD) that serves as a model capable of capturing the entire design space of a CPU. By utilizing principles of Monte Carlo-based expansion and Boolean functions, the researchers were able to enhance the accuracy and efficiency of their generated CPU designs. Notably, the resulting CPU design was solely formed from external input-output observations, without the need for formal program code.
The generated BSD not only discovers the optimal CPU architecture but also automatically generates optimal instruction-level parallelism. Notably, this marks the first time an optimal CPU architecture has been generated from external observations rather than formal program code," state the researchers.
Testing and Applications
The automated AI design of the CPU resulted in a taped-out RISC-V32IA instruction set CPU, fabricated at 65nm, and capable of running at up to 300 MHz. The AI-generated CPU was successfully tested with the Linux operating system (kernel 5.15) and SPEC CINT 2000 benchmarks, confirming its functionality. Additional benchmark tests using the Drystone benchmarks revealed that the AI-generated CPU performed similarly to an i486 and slightly better than an Acorn Archimedes A3010, an ARM-based British computer from the 1980s.
The Use of AI in Computer Processor Design
The use of AI in designing a new RISC-V CPU holds great potential for advancements in computer processor design and has the ability to reduce design and optimization cycles in the semiconductor industry. The researchers even speculate that this research could lay the foundation for self-evolving machines. "Could this be the beginning of a new era of self-evolving machines? Only time will tell," they ponder.
It is worth noting that this is not the first time AI has been employed to advance computer processor designs. In March, Nvidia utilized AI to optimize chip designs, with a particular focus on floor-planning work. The company's DSO.ai software has already been used in over 200 customer chip designs and has the potential to transform the semiconductor industry as a whole.
For more information on the latest computer processor designs, be sure to explore our comprehensive guide. Additionally, discover the best graphics card for your Gaming PC or stay updated on the latest developments in computer hardware.
