GaN-based bifunctional intelligent sensing: Monolithic integration of fast and slow dynamics

Introduction:

Recently, addressing the inherent timescale mismatch challenge between fast and slow responses in optoelectronic sensors, a collaborative team from Suzhou Institute of Nano-Tech and Nano-Bionics (SINANO), Chinese Academy of Sciences (Yukun ZHAO, Shulong LU, Min JIANG), Fudan University (Lifeng BIAN), and Suzhou University of Science and Technology (Jianya ZHANG) has proposed an innovative monolithic integration scheme. By combining surface defect introduction and local contact interface design with a gallium nitride (GaN) nanowire lift-off technique that eliminates the interference from the underlying silicon substrate, the team integrates fast and slow responses into a single device. This results in a transparent bifunctional device capable of self-driven detection and neural synaptic integration, with omnidirectional (360°) detection capability. As a photodetector, the device demonstrates the millisecond-level response speeds, while it exhibits the second- to minute-level relaxation time as an artificial synapse, achieving an over 1000-fold contrast in response dynamics. The device has been validated in the intelligent perception systems for humanoid robots successfully, advancing the development of multifunctional monolithic optoelectronic devices and providing a solid foundation for further research in related fields.

The work entitled "A dual-mode transparent device for 360° quasi-omnidirectional self-driven photodetection and efficient ultralow-power neuromorphic computing" was published in Light: Science & Applications.

1. Background

Photodetectors, capable of rapidly and sensitively converting optical signals into electrical signals, serve as critical tools for human perception of the world. They are widely applied in aerospace, communications, industry, scientific research and medical fields to acquire, transmit and analyze information. Among them, self-driven photodetectors have drawn significant industry attention due to their ultra-low energy consumption. On the other hand, the human visual system not only detects light but also processes and memorizes the images in the brain, enabling efficient information understanding. Inspired by this, scientists have developed neuromorphic computing systems, with the core of artificial synapse devices. These devices emulate brain neural synapses to simultaneously store and process data, enhancing computational efficiency. Such vision-mimicking devices can perceive optical signals and process them through internal “synaptic” mechanisms, enabling adaptive image recognition and analysis capabilities. Gallium nitride (GaN), renowned for its superior physicochemical stability, stands as an ideal material for optoelectronic device fabrication. Therefore, integrating self-driven detectors with synaptic devices into a single GaN-based device (Figure 1) could simplify system architectures, drastically reduce power consumption and significantly expand application scope.

2. Technical Challenges

Photodetectors require rapid responses to light changes, while artificial synapse devices demand extended durations to process and store signals. Due to this fundamental speed mismatch, photodetectors lack the synaptic capability to memorize images or process optical signals, creating the integration barriers. In essence, the disparate response dynamics (millisecond-scale detection vs. second/minute-scale synaptic relaxation) hinder efficient synergistic operation within a single device. On the other hand, omnidirectional detection necessitates device transparency. However, silicon substrates commonly used for GaN nanowires are opaque across ultraviolet (UV) and visible spectra, compromising overall transparency. This substrate-induced opacity and disparate response dynamics have hindered the development of GaN-based bifunctional devices capable of 360° detection and synaptic integration, these challenges that remain unresolved to date.

3. Innovative Strategy

To address the aforementioned challenges, this study employed electrochemical lift-off techniques to remove silicon epitaxial substrates and constructed an “interface-bulk separation” architecture on a transparent substrate. This architecture comprises graphene/(Al,Ga)N heterojunction functional regions and GaN functional regions, enabling the monolithic integration of a self-driven 360° quasi-omnidirectional GaN-based photodetector with artificial synaptic functionality for the first time. By precisely tuning carrier transport dynamics and oxygen vacancy (V_O) ionization levels through bias voltage control, the team successfully harmonized both fast response and slow relaxation characteristics within a single device. Based on the previous works of this team in the fields of GaN nanowire epitaxy and device fabrication [Adv. Funct. Mater. 35, 2416288 (2025, Front Cover); Commun. Mater. 6, 83 (2025); Small Methods 9, 2400989 (2025)], this work further achieved the controllable design and integrated application of material properties with device functionalities.

4. Innovation and Highlight

This study introduces a trinity solution integrating electrochemical exfoliation, zonal architecture design and bias voltage regulation, achieving monolithic integration of the fast and slow responses with dual functionalities of self-driven omnidirectional detection and neuromorphic computing. The proposed device has also been successfully validated in humanoid robotic systems.

1) Process Breakthrough

Electrochemical lift-off technique was employed to remove silicon epitaxial substrates, resolving the fundamental transparency barrier. Figure 2 illustrates the transparent device fabrication process, completing in the realization of a dual-mode monolithic integrated device.

2) Architectural Innovation

The “interface-bulk separation” architecture features spatially decoupled design as the key to resolving functional conflicts: graphene/(Al,Ga)N heterojunctions form the photodetection functional zone (fast response), while GaN bulk regions constitute the synaptic functional zone (slow relaxation).

3) Functional Integration

As shown in Figure 4, the device achieves 360° quasi-omnidirectional detection with 4 ms rapid response. Its high transparency (peak visible transmittance >70%) enables efficient front/backside optical signal transmission, laying foundations for bidirectional “information transmission” applications.

In the neuromorphic vision system (Figure 5), the periodic optical pulses simulate neurotransmitter inputs, with the dual-mode device acting as a synaptic signal processing unit. The device demonstrates unique advantages in UV imaging scenarios, featuring the strong visible light interference resistance with deep UV/visible light rejection ratio reaching 12908.7. It is suitable for high-brightness or low-contrast environments like outdoor monitoring under intense sunlight. Meanwhile, single synaptic event energy consumption as low as 25 fJ, comparable to biological synapses, provides critical technological support for ultra-low-power neuromorphic chips.

4) Humanoid Robot Application Validation

As illustrated in Figure 6, the team pioneered the validation of application potential of this novel bifunctional device in humanoid robotics, contributing to enhance the intelligent perception and computing capabilities of humanoid intelligent robots while reducing the power consumption.

5. Prospect and Outlook

The bifunctional transparent device proposed in this work offers an innovative solution for intelligent machine vision systems. Its 360° self-driven sensing and on-chip processing capabilities can enhance the environmental adaptability of humanoid intelligent robots, enabling features such as obstacle avoidance without blind spots and real-time decision-making. Simultaneously, it provides scalable units for neuromorphic vision chips, advancing the development of high-efficiency brain-inspired computing hardware. Looking ahead, this work is poised to deliver solid technical support for overcoming the challenges of sensing-processing separated architectures, empowering the next-generation AI terminals.