A Core Part of an AI Semiconductor
A team of Korean researchers has developed a core part of an artificial intelligence (AI) semiconductor, which is needed to develop computers like the human brain and nervous system. This part saves a new computer 10 times more power compared to running a current AI computer system.
The Korea Institute for Science and Technology (KIST) announced on March 29 that a joint research team led by Dr. Song Kyung-mi, Dr. Joo Hyun-soo, Director Jang Joon-yeon and Dr. Woo Sung-hoon of IBM at its Next Generation Semiconductor Research Institute has developed a core part for the world's first neuromorphic computer using a skyrmion. The researchers predict that if the part is made smaller and several of it is connected to each other, it will lead to the development of an AI CPU that combines a CPU and a memory of the current PCs.
"We came up with the world's first skyrmion-based artificial synapse component which had been suggested only through theories," Dr. Song explained.
The researchers electrically controlled a skyrmion to mimic the movement of electron spins in a synapse that transmit excitement and stimulation from neurons of the human brain.
The team conducted a Modified National Institute of Standards and Technology (MNIST) recognition test using this artificial synapse material and found that it obtained a high recognition rate of 90 percent with 15,000 times of learning alone. Other artificial synapse materials required hundreds of thousands of iterative learning sessions to achieve a 90 percent recognition rate. This means that the technology developed by the team requires less than 10 percent of power.
"The component closely mimics the human brain that regulates synapses by the amount of neurotransmitters by controlling synapses’ weights according to the number of electrically controlled skyrmions," Dr. Song said.
The results of this study were published on the March 16 online edition of the global journal Nature Electronics.