Russia is developing a processor to accelerate neural networks

    Four Russian companies have teamed up to create the first domestic processor designed to drastically increase the performance of computer neural networks, Izvestia writes . Experts said in the comments to the newspaper that the Russian chip must be quite competitive in the “only emerging world market of neuroprocessors”.

    Neural processors are specialized chips that implement hardware acceleration of the algorithms of artificial neural networks, computer vision, voice recognition, machine learning, and other artificial intelligence methods. The first attempts to manufacture such microcircuits that specialize in pattern recognition were made in 1993., and now you won't surprise anyone like that. And even more so using the GPU to accelerate neural computing, which is almost as efficient as a specialized ASIC.

    Nevertheless, the domestic media write about the “processor for artificial intelligence”, as well as about the unique quantum computer and other revolutionary achievements of Russian scientists and engineers.

    According to Russian experts, separate experimental devices of this type already exist, but the final formation of the world market for neuroprocessors will take "another four to six years." During this time, “small companies and even start-ups will have a chance to gain a foothold in the market.”

    Who specifically develops the processor?

    The development is occupied by four small companies that are part of the industry union of NeuroNet and are funded under the program of the National Technology Initiative. Specific company names are still kept secret and promise to "officially be named in February."

    The National Technology Initiative (NTI) is a state program of measures to support the development of promising industries in Russia, which over the next 20 years can become the basis of the world economy. For the “breakthrough development”, 9 directions were chosen, one of them is NeuroNet.

    Seven “key market segments” are listed on the official NeuroNet website:

    Apparently, for each of these segments, companies need to be found and financed according to the terms of the multi-billion dollar STI program. It is difficult to assess the prospects of other areas, but Russian engineers are quite capable of designing a specialized ASIC to speed up specific calculations, so that at least part of the funding will go to real work.

    As Izvestia writes, the national neuroprocessor “is capable not only of competing with Western models, but will also become 100% domestic,“ trusted ”, that is, guaranteed free from undocumented features and hardware“ bookmarks ”.” The latter is especially important for customers from the Russian military-industrial complex, where neural networks are also becoming widespread - in combat drones control systems, in planning military operations, in precision-guided small arms instrumentation.

    “Russian mathematicians and engineers who develop hardware and algorithms in the field of artificial intelligence and neural networks are the best in the world,” said Alexander Semenov. “Now they have about four years to get ahead of their foreign colleagues and set standards for the future market.”

    According to the head of the laboratory of neural network technologies and computational linguistics at the Moscow Institute of Physics and Technology Stanislav Ashmanov, there are about 2,000 companies in the world today in the race to create a reference neural processor: “Whoever has time to make a chip that will become the industry standard will earn money that is commensurate with - incomes of the current leaders of the market of central processors, such as Intel or AMD, - Stanislav Ashmanov is sure. “So far, out of this pair of thousands of startups all over the world are closest to winning no more than five companies.”

    Neuroprocessors are developed for different areas. Stanislav Ashmanov says that the “race” is going in two directions:

    1. Server chip for powerful servers in data centers.
    2. Economical embedded neuroprocessor for installation on various mobile devices: smartphones, robots, drones, unmanned vehicles.

    According to Ashmanov, Russian developers "have a chance to win in both directions."

    Also popular now: