Implanting Thought Inside Chip: Behind the Scene of Neuromorphic Semiconductor Development by Prof. Park Jin-sub and Park Jea-gun

A research team led by Professor Park Jin-sub and Professor Park Jea-gun (School of Electronic Engineering) has developed "neuromorphic chips" that can mimic the high-level cognitive function of humans by designing artificial neural chips and artificial synaptic devices with low power consumption and high integration density. Neuromorphic semiconductors are considered a key future AI technology, as they enable ultra-low-power AI implementation. Notable example applications include IoT devices, wearable technology, autonomous driving, brain-machine interfaces, and medical healthcare.

Until now, there has been no report of an artificial neuron device capable of highly dense integration at low voltage. Through this study, Professor Park's research team has developed an artificial neural chip that not only performs all the functions of a biological neuron but also achieves high-density integration and low power consumption. We spoke with Professor Park Jin-sub, who pushed neuromorphic semiconductor research toward practical applications with more advanced research.

In short, a neuromorphic semiconductor is a "semiconductor that resembles the neural network of the brain." It mimics the structure of neurons and synapses to function like the brain, and it processes information by turning this into a semiconductor. Key components include neurons, synapses, and the algorithms that regulate them. These elements are organically interconnected, allowing efficient information processing like the human brain. When the human brain receives an input signal, each neuron applies a weight and transmits it to the next neuron. These structures enable computation and learning.

Traditional artificial neural chips have been used primarily for speech signal processing and image recognition. Compared to AI semiconductors that rely on conventional artificial neurons, neuromorphic semiconductors go beyond simple computation and are capable of adaptive learning and reasoning. In particular, it is expected to have fast computational speeds that even surpass the learning speed of the brain and large-scale data processing capabilities. "The ultimate goal is to make it closer to the world like robotics, IoT, and healthcare," said Professor Park.

Existing neuromorphic semiconductors are made of silicon so in terms of how signals are transmitted between neurons, they face scaling (the process of making semiconductor devices small) issues and challenges like high power consumption. Capacitors were essential to make neuromorphic semiconductors, but since billions of neurons are needed, capacitor size increases, leading to scaling limitations. To overcome this, Professor Park adopted a capacitorless approach. He added, " By adopting a capacitorless system instead of the traditional capacitor-based method, we designed a structure that enables charge storage and transfer without requiring separate capacitors."

Another challenge was power consumption. While the human brain operates at less than 20W, traditional neuromorphic semiconductors require high voltage. Professor Park successfully reduced the operating voltage to 0.7V using Chalcogenide-based materials. He explained, "Our goal was to bring the voltage at least to lower than 1V, so achieving this milestone means a lot."

"As technology advances, humans must become more cautious. So that our creations do not seize control over us."

The 21st century is witnessing rapid technological advancements. AI is quickly advancing, as a key next-generation technology. The rapid development of AI has brought groundbreaking efficiencies and transformations to human life. However, on the flip side lies a lingering fear reminiscent of science fiction narratives of the past, which are "humans governed by machines."

"One day, AI will catch up even the human creativity," Professor Park predicted. "However, rather than halting development due to vague fears, it is more important to seek new technologies that can address potential risks," he added. "Throughout history, if new technologies introduced new challenges, efforts will always be made to manage these challenges"

Professor Park shared plans to overcome the limitations of conventional memory devices and develop more efficient memory architectures. Currently, he is working on research using chalcogenide. In particular, he has been selected for the Samsung Future Technology Cultivation Initiative, where he leads the development of new memory devices. Through the research that will continue over the next three years, he aims to expand beyond single-layer structures to multi-layer architectures of three or more layers by focusing on integrating devices and advancing 3D stacking technologies.

▲ As the Department Chair of the Department of Nanoscale Semiconductor Engineering, Professor Park accompanies many students' journey of finding joy by leading Advanced Convergence Semiconductors Innovation LAB. ⓒ Professor Park Jin-sub

Professor Park concluded the interview by sharing a message for Hanyang students.

"I believe that passion is driven by "joy." I hope you all find something that you can enjoy. There is an inherent limit in the work that you do just because someone tells you to do so. But there are no limits when you do something because you love it.

You are all capable of making it happen. I hope you find what you enjoy, work on it, and ultimately grow into intellectuals who help others!"