KAIST develops neuromorphic chip that mimics the human brain
A research team from the Korea Advanced Institute of Science and Technology (KAIST) has developed a high-density neuromorphic semiconductor that mimics the human brain.
Professors Choi Yang-kyu and Choi Sung-Yool from KAIST School of Electrical Engineering led the research.
Neuromorphic material is an artificial intelligence function performed in the form of material that mimics the human brain. This stems from the idea that the human brain can perform complicated tasks but only consumes 20 watts of energy. Unlike the traditional von Neumann method, neuromorphic material attracts attention because it can perform AI functions with very low electricity.
The joint research team developed neuromorphic material of neurons and synapses using a single transistor. The material was produced with a standard commercial silicon process, increasing the possibility of commercializing neuromorphic hardware systems.
Han Joon-Kyu, a doctoral student at KAIST School of Electrical Engineering, was the lead author and Oh Jung-yeop was the second author of the research. The research was published in the August online edition of Science Advances under the title “Cointegration of single-transistor neurons and synapses by nanoscale CMOS fabrication for high scalable neuromorphic hardware”.
To materialize neuromorphic material, synapses that remember the connection between neurons that spike when signals synchronize are needed. However, because neurons and synapses that have sway in digital or analog circuits take up a lot of space, there are limits to their degree of integration. Since the human brain uses around 100 billion neurons and 100,000 billion synapses, the degree of integration needs to be improved for use in mobiles and the Internet of Things (IoT).
To do this, various materials and structural bases of neurons and synapses have been suggested. However, most of them could not be manufactured with the standard Si CMOS process, which makes their commercialization and mass production difficult.
To solve the problem, the research team used a single transistor widely used for the standard CMOS Si process to mimic the movements of biological neurons and synapses. Neuromorphic material was produced in the 8 inch wafer made by cointegration.
The newly produced neuromorphic transistor has the same structure as the currently manufactured transistors for memory and system hardware. The transistor can precede the memory function and the logic operation. This research makes perfect sense by showing the possibility of a new neuromorphic simulation. By applying a new operating principle to the transistors currently produced, a new neuromorphic transistor has been realized, with a similar structure but different functions. Like a coin that has a head and a tail, the neuromorphic transistor has proven for the first time in the world that it can function as both a neuron and a synapse.
The degree of integration was greatly increased by replacing the neuron, based on complicated digital and analog circuits, with a single transistor. In addition, this new technology can reduce costs by simplifying the production process with the same synapse structure. While the current neural circuit requires 21,000 units of two-dimensional space, the newly developed neuromorphic transistor only requires six units. The degree of integration is 3,500 times higher.
Using neuromorphic material, the research team simulated brain functions such as gain modulation and coincidence detection. He also proved that facial image recognition and letter shape recognition are possible.
KAIST expects the neuromorphic material to increase integration and decrease costs, moving closer to commercialization.
“Using a single complementary metal-oxide-semiconductor (CMOS) based transistor, we have proven the possibility of simulation of neurons and synapses,” Han said. “By using the commercialized CMOS process, neurons, synapses and co-integrating additional circuits at the same time into a single wafer, we have increased the degree of integration in neuromorphic material, moving one step closer to commercialization neuromorphic material. “