(A) study on bio-inspired artificial mechanoreceptors with built-in synaptic functions

국문 초록 (Abstract)

시냅스 기능을 내재한 생체영감형 인공 기계수용기는 촉각과 청각 두 감각 축 에서 감지, 기억, 필터링, 전처리를 센서 자체에서 수행하는 모노리식형 아키텍처를 구현하였다. 마찰전기 수용층과 그래핀 기반 시냅스 트랜지스터를 하나의 수직적 구조로 일체화하여, 기계 자극이 만드는 전위를 시냅스 후 전류로 변환하고, 이 신 호에 기반해 적응과 기억을 말단 수용체 단계에서 수행하도록 개발하여, 방대한 양 의 데이터를 센서 자체적으로 전처리가 가능해 빅데이터 산업에서 데이터를 병렬적, 대량으로 처리가 가능하도록 구현하였다. 이를 통해 데이터의 병목현상이 발생하는 본노이만 구조 대비 데이터, 전력 효율을 높이고 지연을 줄일 수 있어, 향후 고속, 고효율, 자율 AI 시스템 구현의 핵심 기술의 새로운 패러다임을 개발하였다. 촉각 영감형 인공 기계수용기는 인간의 느린적응과 빠른적응 경로를 하나의 플랫폼 에서 하드웨어로 개발하였다. 지문 유사 미세구조와 이온겔 게이트 절연체를 수직 적층하고, 마찰전기-정전용량 결합으로 느린적응과 빠른적응 신호를 오직 평면전극 유무에 따라서 분리되도록 재현하여 설계를 단순화했다. 자극의 주파수, 시간, 힘, 161 접촉 횟수로 시냅스 가소성을 정밀 조절하였고, 이온겔 게이트 절연체의 농도를 조 절하여 시냅스 가소성의 정도, 즉 기억의 정도를 조정해 장기기억 혹은 단기기억의 기능이 조절되도록 하였다. 시냅스 가소성의 기억 특성을 활용하여 한글 글씨 쓰는 순서를 예측하였고, 데이터를 100 % 사용하여 물체를 구별하는 기존 센서들과는 달리, 본 인공 기계수용기 연구에서는 시냅스 가소성을 활용하여, 자극이 종료된 후 의 전류 값을 사용해 10 %미만의 데이터를 사용하여 여러 종류의 거칠기, 미세질 감을 95 %이상의 정확도로 구별하였다. 청각 영감형 인공 기계수용기는 소리가 귀로 들어오면 증폭하거나 소리를 자체적으 로 줄여 신호 감쇄 역할을 하는 인간의 소리 감지 기능을 하는 유모세포의 기능을 모방하였다. P(VDF-TrFE) 기반 마찰전기 층을 나노입자 및 이온 첨가제로 설계 해 약한 음압에서도 높은 출력전압을 획득할 수 있도록 설계하였고, 이를 이온겔 게이트절연체와 모노리식하게 적층하여 외부 증폭기 없이 시냅스 후 전류를 생성하 였다. 또한, PEIE의 소량 첨가로 전하 중립점을 낮추고 그래핀 채널의 양극성을 활 용하여 평상 음압에서는 흥분성 시냅스 후 전류(EPSC), 큰 음압에서는 억제성 시 냅스 후 전류(IPSC)로 전환되는 자가 억제를 구현해 생물학적 유모세포의 기능을 구현하였다. 또한, 사람 목소리 입력에 대해 민감한 감지를 보여주어 단어 인식, 방 향, 강도 등 다양한 활용가능성을 제시하였다. 본 연구는 별도의 외부 전원없이 기계적 자극을 게이트 전압으로 활용해 소자 구동 이 가능하고, 센서와 메모리 소자의 일체화로 데이터 전송의 과부화를 줄여 방대한 양의 데이터를 저전압, 저전력, 고효율로 처리하는 것이 핵심이며, 뉴로모픽 감각시 스템, 웨어러블 전자피부 등 다양한 분야에 새로운 기술적 가능성을 제시하였다.

다국어 초록 (Multilingual Abstract)

Artificial sensory systems require not only high-fidelity transduction of external stimuli but also memory and adaptive processing to reduce data redundancy at the sensor level. Here, we propose a unified bio-inspired mechanoreceptor platform that embeds synaptic functionality directly into the sensing element and demonstrates its applicability across tactile and auditory modalities.
Tactile perception. Inspired by human cutaneous afferents where slowly adapting (SA) and fast adapting (FA) pathways, together with synapse-like interactions at peripheral terminals, perform pre-processing. We present an array of artificial synaptic mechanoreceptors by vertically integrating synaptic field-effect transistors (FETs) with reduced-graphene-oxide (rGO) channels, an ionogel gate dielectric, and an elastomeric fingerprint-like receptive layer in a single stack. Tactile stimulation induces triboelectric–capacitive gating between the receptive layer and ionogel, governing excitatory postsynaptic current (EPSC) patterns that reproduce SA/FA characteristics for signal preprocessing. Using fused SA/FA postsynaptic features, the array achieves handwriting style, surface pattern, and texture recognition with high data efficiency via machine learning.
Auditory perception. In biology, stereocilia in hair cell deflection modulates tip-link tension to gate mechano-electrical transduction channels, yielding a direction sensitive, generating the receptor potential with rapid adaptation. In artificial auditory mechanoreceptors, a tribo-positive and negative pair are monolithically stacked on a synaptic rGO-FET so that acoustic vibration produces an alternating, potential while simultaneously enabling synaptic functions. The triboelectric layer based on P(VDF-TrFE) with BaTiO3, Al2O3, TiO2 nanoparticles and ionic additives (NaCl, KCl) amplifies the output voltage, which modulates the rGO-FET biased near its Dirac point. Dirac-point shifting enables both EPSC and inhibitory postsynaptic current (IPSC) responses, providing a hardware pathway for noise cancellation, selective frequency/level filtering, and sound recognition. With human voice inputs, the system exhibits high sensitivity detection and indicates the feasibility of estimating sound direction and intensity as well as discriminating alphabets and words.
Collectively, these results establish a scalable route to multimodal neuromorphic sensing and, more broadly, exemplify a Physical AI architecture in which perception, adaptation, and low-level inference are embedded directly in the sensing hardware.

Tactile perception is mediated by a combination of preprocessing of signals from both slowly adapting (SA) and fast adapting (FA) afferents, which have synapse-like interactions between mechanoreceptors and their dendrites or terminals and transmission of the signal to the brain. However, emulating stimulus reception and preprocessing functions of SA and FA afferents, such as adaptation and sensory memory, is of great interest for artificial tactile sensing systems. Here, inspired by the human tactile afferent systems, an array of artificial synaptic mechanoreceptors (ASMRs) with built-in synaptic functions is proposed as realized by vertically integrating synaptic transistors having a reduced graphene oxide channel, an ionogel gate dielectric, and an elastomeric fingerprint-like receptive layer in an all-in-one platform. Tuning of triboelectric-capacitive gating between the receptive layer and gate dielectric in response to tactile stimulation governs excitatory post-synaptic current pattern-enabled functions of signal preprocessing through the built-in synaptic function by endowing ASMRs with SA and FA characteristics. The ASMRs array was demonstrated for discriminative reception of tactile stimuli, including the discrimination of handwriting style and recognition of surface patterns and textures with machine learning using the fused post-synaptic values of FA and SA ASMRs, rather than all the signal data, providing high data efficiency. The proposed ASMRs have great potential for intelligent tactile skin with high data efficiency.

This chapter presents a bio-inspired artificial hair cell designed for edge intelligence. The device follows the cochlear-mimicking structure in which inner hair cells convert stereocilia deflection into receptor potentials and outer hair cells supply level-dependent amplification and tuning. We realize this behavior by monolithically stacking a tribo-positive and tribo-negative pair on a synaptic transistor with a reduced graphene oxide channel and a PU ionogel as dielectric layer. Acoustic vibration generates a triboelectric potential that directly gates the transistor and produces postsynaptic currents. The channel is biased near its Dirac point so that small polarity shifts yield both excitatory and inhibitory responses, enabling adaptive gain control, temporal filtering, and local memory without external processors.
The triboelectric layer is based on P(VDF-TrFE) co-engineered with BaTiO3, Al2O3, and TiO2 nanoparticles, and ionic additives such as NaCl and KCl, which amplifies the acoustic potential under weak stimuli. This stacked architecture couples transduction and synaptic modulation in one path and generates the short-term or long-term plasticity that encodes recent acoustic history. Using human-voice inputs, the system detects low-level signals with high sensitivity and extracts compact features. Experiments indicate feasibility for estimating sound direction and intensity and for discriminating alphabets and words using in-sensor features, which increases data efficiency and power efficiency.
The results demonstrate an artificial hair cell that unifies sensing, memory, and preprocessing in a single, flexible stack. This platform advances edge-intelligent audition by embedding perception and inference in the hardware and establishes a path toward scalable, multimodal Physical AI front ends.

목차 (Table of Contents)

• Abstract
• Chapter 1. INTRODUCTION 1
• 1.1 Bio-inspired artificial-intelligence perception 1
• 1.1.1 Tactile perception system in human - mechanoreceptors 3
• 1.1.2 Mechanism of tactile perception system in human 5
• 1.2 Background of auditory perception 6
• 1.2.1 Auditory perception system in human Hair cell 6
• 1.2.2 Mechanism of auditory perception system in human 7
• 1.3 Background of research 8
• 1.3.1 Triboelectric enhancement by surface roughness: Effects and Mechanism 8
• 1.3.2 Synaptic devices: principles, mechanisms, advantages, and prospects 11
• 1.3.3 Advantages of rGO devices compared with other memory devices 14
• 1.4 Current issues and Challenges 16
• 1.4.1 Complexity, energy efficiency, flexibility 16
• 1.4.2 Functional issues at the receptor level 16
• 1.5 Summary 17
• Chapter 2. Bio-inspired artificial mechanoreceptors with built-in synaptic functions for intelligent tactile skin 19
• 2.1. Abstract 19
• 2.2. Introduction 20
• 2.3. Human tactile systems and artificial mechanoreceptors 23
• 2.3.1. Concept of artificial synaptic mechanoreceptors (ASMRs) inspired by human tactile sensory organs 23
• 2.3.2. Mechanism of synaptic-like connection in our sensory organ and ASMRs (mimicking the SA-afferent system) 30
• 2.3.3. Mechanism of synaptic-like connection in our sensory organ and ASMRs (mimicking the FA-afferent system) 32
• 2.3.4. Synaptic properties of IG-gated rGO-FET as a synapse-like connection 34
• 2.3.5. Tactile reception characteristics of SA and FA ASMRs 53
• 2.3.6. Array of ASMRs with built-in synaptic function for bio-inspired intelligent tactile skin 77
• 2.4. Conclusions 118
• Chapter 3. Bio-inspired artificial hair cell for edge intelligence system 120
• 3.1 Abstract 120
• 3.2. Introduction 122
• 3.3. Artificial mechanoreceptor inspired by human auditory system 125
• 3.3.1. Artificial auditory sensing with triboelectric effect 125
• 3.3.2. Intelligent artificial hair cell with rGO-FET 133
• 3.4 Conclusion 140
• Chapter 4. Summary & Conclusion 142
• References 144
• 논문요약 160