Intel's latest chip will abandon the traditional computing architecture to create a system that can model itself in the way the human brain works.
On Monday, the company introduced a neuronorphic simulation (neuromorphic), codenamed "Pohoiki Beach," capable of processing data like a biological brain by simulating the whole. neural cell network. This approach accelerates the information processing capabilities of semiconductor pieces, with Intel representatives saying: "It will be 1,000 times faster and more than 10,000 times more effective"compared to traditional CPUs in certain AI related tasks.
Pohoiki Beach is made up of 64 smaller "Loihi" chips that can simulate about 8.3 million neurons. This number is likely to be very large, but it is actually only equivalent to the small rodent's brain. As for the human brain, the number of nerve cells can reach nearly 100 billion.
Anyway, the new chip promises to become the foundation for the latest AI algorithms. In particular, Intel said Pohoiki Beach may have a huge advantage in computing speed related to image recognition, self-driving cars and autonomous robots.
Each Nahuku board will contain 16 small chips.
Pohoiki Beach is available free of charge to Intel's mental research community, including more than 60 different partners. However, the company said it hoped to commercialize the technology at a time when Intel was trying to create a faster, more energy-efficient silicon.
"Continuing to increase the benefits of energy efficiency and performance under Moore's Law will need more than just narrowing the process.. "Intel said in its announcement."As complex computing jobs are becoming a normal thing, the need for special architectures designed for specific applications is also increasing.. "
Time will tell if this technology could one day appear on the company's PC processors. But one thing is for sure, that you will see Intel talking more about future morphological chips. Later this year, the company plans to expand this technology to 100 million neurons. However, Intel is not the only one on this road, both Qualcomm and IBM are developing silicon plates in this approach.
Refer to PCMag