AI Semiconductors That Learn Like the Brain Solve 'Reliability' Issue [Reading Science]

Implementing Low-Power Neuromorphic Devices
Through the Combination of Organic and Inorganic Semiconductors

A core semiconductor technology that will accelerate the era in which artificial intelligence (AI) learns and makes decisions like the human brain has been developed by a Korean research team. It is being evaluated as a breakthrough that could resolve the device reliability issues that have been considered the biggest challenge for “brain-inspired AI semiconductors,” which perform both computation and storage simultaneously.

The National Research Foundation of Korea announced on January 13 that a research team led by Professor Joo Jinsoo at Korea University has implemented a heterojunction memtransistor by combining an organic semiconductor with a two-dimensional inorganic semiconductor, thereby achieving both current control stability and reproducibility in neuromorphic devices.

Bottom-contact (BC) TCTA/MoS2 memtransistor structure and characteristics.
a. (Top) Schematic of BC TCTA/MoS2 memtransistor. (Bottom) Schematic of biological neuron system.
b. Energy band alignment structure of the heterojunction active layer in BC TCTA/MoS2 memtransistor.
c. Optical microscope image of BC TCTA/MoS2 memtransistor.
d. Current transfer characteristics (ID vs. VG) graph of BC TCTA/MoS2 memtransistor.
e. Current output characteristics (ID vs. VD) graph of BC TCTA/MoS2 memtransistor.

Figure description and provided by: Taekjun Kim, PhD, Korea University (currently at Samsung Electronics)

‘Von Neumann’ Architecture Limitations... Brain-Inspired AI as an Alternative

Current computers are based on the “von Neumann architecture,” in which the processor and memory are separated. This structure causes bottlenecks during data transfer and has the drawback of high power consumption. With the spread of generative AI and the resulting surge in computational demand, these structural limitations are becoming even more pronounced.

Neuromorphic computing, which is attracting attention as an alternative, operates like the human brain by performing computation and storage simultaneously and transmitting signals only when necessary to maximize energy efficiency. However, the core device for realizing this-memtransistors-has faced criticism for unstable current control due to material properties and declining reliability with repeated operation.

Stabilizing Current Flow by Combining Organic and Inorganic Semiconductors

To overcome these limitations, the research team designed a heterojunction structure by combining the organic semiconductor TCTA with the two-dimensional inorganic semiconductor molybdenum disulfide (MoS₂). By assigning the high-resistance state to the organic semiconductor and the low-resistance state to the inorganic semiconductor, they precisely controlled the flow of current.

In particular, by introducing a bottom-contact structure where the electrode contacts the underside of the active layer, they stabilized charge transport characteristics. This allowed them to realize learning properties similar to real brain synapses, such as long-term plasticity and spike-timing-dependent plasticity (STDP).

Another achievement of this research is the ability to control synaptic operation using the drain voltage while utilizing the gate voltage as an additional input, thereby enabling multi-input processing and flexible learning functions.

“A Practical Solution for Brain-Inspired AI Semiconductors”

Professor Joo Jinsoo explained, “This research is significant in that it implemented a memtransistor capable of precise current control through the meticulous design of semiconductor junctions. It is a case where both the reliability and reproducibility of the device have been achieved in the field of electronic systems that mimic brain neural networks.”

This research was supported by the Mid-Career Research Program of the Ministry of Science and ICT and the National Research Foundation of Korea. The results were published online in the international journal Advanced Science on January 5.

The title of the paper is “Heterosynaptic Memtransistors Based on Switching Operation Mechanism Using Designed Organic/Inorganic Heterostructures for Neuromorphic Electronics.”