Introduction: Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have created a highly scalable neuromorphic hardware inspired by the brain by synergistically integrating single-transistor neurons and synapses.
Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have created a highly scalable neuromorphic hardware inspired by the brain by synergistically integrating single-transistor neurons and synapses. Using standard silicon complementary metal-oxide-semiconductor (CMOS) technology, this neuromorphic hardware promises to reduce chip costs and simplify the manufacturing process.
A research team led by Yang-Kyu Choi and Sung-Yool Choi has created highly scalable neuromorphic hardware neurons and synapses based on single transistors, demonstrating their ability to recognize text and facial images. The findings were published in the journal *Science Advances*.
Neuromorphic hardware has garnered significant attention due to its artificial intelligence capabilities, yet it consumes less than 20 watts of power because it mimics the human brain. For neuromorphic hardware to function, it must have a neuron that generates a spike when integrating a signal, and a synapse that remembers the connection between two neurons, much like a biological brain. However, the large space required for neurons and synapses built on digital or analog circuitry limits hardware efficiency and cost. Since the human brain consists of approximately 10¹¹ neurons and 10¹⁴ synapses, applying it to mobile and IoT devices would necessitate increased hardware costs.
To address this problem, the research team simulated the behavior of biological neurons and synapses using a transistor and co-integrated them onto an 8-inch chip. The resulting neuromorphic transistor has the same structure as currently mass-produced memory and logic transistors. Furthermore, for the first time, neuromorphic transistors have demonstrated that they can achieve a "two-sided structure," functioning as both a neuron and a synapse, much like a coin has two sides.
Professor Yang-Kyu Choi stated, "By replacing neurons and synapses in complex digital and analog circuits with a single transistor, hardware costs can be significantly reduced." First author Joon-Kyu Han stated, "We have demonstrated that neurons and synapses can be implemented with a single transistor." "By using standard CMOS processes to co-integrate single-transistor neurons and synapses on the same chip, the hardware cost of neuromorphic hardware has been reduced, which will accelerate the commercialization of neuromorphic hardware."
This research was supported by the National Research Foundation (NRF) and the Integrated Circuit Design Education Center (IDEC).
