Scientists of Tyumen State University presented to the world scientific community the development of an unparalleled biomorphic neuroprocessor based on a new component of nanoelectronics - a combined memristor-diode crossbar, the press service of Tyumen State University reports. The research results are published in the Microelectronic journal.
The existing neuroprocessors, according to the scientists of the Tyumen State University, are designed for hardware acceleration of calculations in artificial neural networks on simple neurons and ensure the operation of computer vision, machine learning and other systems with weak artificial intelligence (AI).
Information processing and decision-making in such processors occurs by choosing the most plausible solution based on previously established associations.
The neuroprocessor developed at Tyumen State University, according to its creators, is capable of generating new associations (new knowledge) by a biologically similar mechanism, which allows us to speak of a possible transition from weak to strong artificial intelligence. Strong AI means the ability to comprehend new knowledge.
According to the Tyumen State University, the biomorphic neuroprocessor implements a pulsed hardware neural network based on the developed electrical and original software (Neural Computing and Applications) biomorphic models of the neuron.
The authors claim that their processor is capable, in addition to solving traditional information processing tasks, to reproduce the work of the cortical column of the brain.
According to the professor, the neuroprocessor is, in fact, the basis of a unique new generation system, which is the carrier of artificial intelligence.
The developed specialized hardware can be used to solve technical problems - to increase the speed and energy efficiency of calculations in comparison with the computing facilities existing today (personal computers, servers and supercomputers).
Promotional video:
The effect is achieved through the use of mixed analog-digital computations, including using bipolar memristors integrated into combined memristor-diode crossbars.
To date, scientists from Tyumen State University have successfully demonstrated information processing in manufactured memristor-diode crossbars - weighing, adding and routing pulses, as well as associative self-learning and generation of new associations. Until now, associative self-learning has been demonstrated only in hardware neural networks built on discrete memristors.
At the moment, the research group continues hardware testing of the new system. According to scientists, the launch of a unique neuroprocessor into small-scale production will be possible by 2025. The research was supported by RFBR grants No. 19-07-00272 and No. 19-37-90030.
