Strain-controlled power devices as inspired by human reflex

The rapid advancements in artificial intelligence (AI) have revolutionized various aspects of our lives, particularly in robotics and autopilot technologies. Nature offers numerous examples of efficient and responsive systems, inspiring the development of bio-inspired electronic devices. Existing sensor-actuator systems, however, often rely on complex circuitry (A/D and D/A converters, electrical isolation, CPU intervention) to translate mechanical signals into electrical control. This complexity hinders the development of power devices capable of real-time, rapid response power modulation crucial for applications like self-driving cars and robots requiring immediate power adjustments based on environmental changes. This research aims to develop a power device that directly responds to external stimuli, mimicking the speed and efficiency of the human reflex. The use of III-V materials, particularly AlGaN/AlN/GaN HEMTs, offers a promising pathway due to their high carrier density, mobility, and breakdown field. The piezotronic effect, where mechanical strain modifies electrical transport properties, is leveraged to achieve this direct power modulation. This bio-inspired approach seeks to create a simpler, more efficient, and faster system for power control in AI applications.

The literature extensively documents the development of bio-inspired electronics, including e-skin, e-nose, cochlear implants, prosthetics, and even artificial larynxes. These devices often mimic specific biological functionalities. However, the focus on power devices directly modulated by mechanical stimuli and capable of real-time control remains relatively limited. Studies have explored the piezotronic effect in III-nitride materials, demonstrating the potential for strain-based modulation of electrical properties in nanowires and HEMTs. Existing research highlights the potential of AlGaN/AlN/GaN HEMTs for high-power applications, but the direct integration of strain-based control for real-time power modulation is a novel advancement. The human reflex, specifically the knee-jerk response, serves as a compelling biological model for rapid, involuntary response to a stimulus. This study leverages the speed and efficiency of this reflex arc to create a more responsive and less complex power device for AI applications.

The study designs a strain-controlled power device (SPD) using an AlGaN/AlN/GaN HEMT in a cantilever architecture. The device fabrication involves III-nitride epitaxial layer growth via metal-organic chemical vapor deposition (MOCVD) on a silicon substrate. Inductively coupled plasma etching (ICP) is employed for precise fabrication of the cantilever structure. A fully dry etching process is used to minimize contamination and ensure better integration with silicon-based circuits. Ohmic and Schottky contacts are formed using Ti/Al/Ni/Au and Ni/Au metal stacks, respectively. The characterization of the fabricated SPD involves field-emission scanning electron microscopy (SEM), focused ion beam scanning electron microscopy (FIB), high-resolution transmission electron microscopy (HRTEM) with energy-dispersive X-ray spectroscopy (EDS), and confocal Raman spectroscopy. Electrical characteristics are measured using a Keysight B1500A semiconductor characterization system. Strain is applied to the cantilever using a probe needle, and the resulting changes in current and power are measured. Acceleration-feedback control experiments utilize a combined system of a semiconductor characterization tester and a linear motor, enabling the evaluation of real-time power modulation in response to varying acceleration levels. COMSOL Multiphysics is used for simulation of strain distribution within the device.

The fabricated SPD demonstrates significant output power modulation (2.30–2.72 × 10³ W cm⁻²) under weak mechanical stimuli (0–16 mN) at a gate bias of 1 V. The device exhibits a maximum transconductance of 7.5 mS mm⁻¹. Raman spectroscopy confirms the presence of tensile strain relaxation in the GaN layer after the etching process. The output power density increases with applied strain, reaching 1.39 × 10³ W cm⁻² at -5 V gate bias and 2.72 × 10³ W cm⁻² at 1 V gate bias under a 16 mN strain. The gate voltage provides programmable control over the sensitivity of the output power to strain. Furthermore, acceleration-feedback control experiments show real-time adjustment of output power in response to acceleration changes. At 15 V supply voltage, the change in power density (ΔP) ranges from 72.78 W cm⁻² to 132.89 W cm⁻² as acceleration increases from 1 G to 5 G. Theoretical calculations show a correlation between strain, 2DEG concentration, and output current, supporting the experimental findings. The reproducible strain-dependent analysis demonstrates consistent device behavior over multiple cycles. The 2DEG sheet carrier concentration increases with increasing strain, which explains the increased output current and power density.

The findings demonstrate the successful creation of a strain-controlled power device that directly modulates output power density in response to mechanical stimuli. The high output power densities achieved, combined with sensitivity to weak forces, represent a significant advance over conventional sensor-actuator systems. The bio-inspired design, mimicking the human reflex, provides a simpler and more efficient alternative for power control in AI applications. The ability to program the sensitivity of the device through gate voltage control adds versatility. The demonstration of real-time acceleration-feedback control further highlights the device’s potential in applications requiring dynamic power adjustment. The findings strongly suggest a new approach to creating intelligent power devices with improved response times and reduced complexity.

This study successfully demonstrates a bio-inspired strain-controlled power device (SPD) exhibiting high output power density and real-time response to mechanical stimuli. The integration of the piezotronic effect with an AlGaN/AlN/GaN HEMT in a cantilever architecture allows for direct, efficient, and programmable power modulation. The successful implementation of acceleration-feedback control opens exciting avenues for application in various AI systems. Future research could focus on further miniaturization, integration with complex systems, and exploration of novel materials for enhanced performance and broader applicability.

The current SPD design shows a slightly reduced current compared to a standard HEMT, likely due to strain release and imperfections introduced during the fabrication process. The study primarily focuses on a specific cantilever geometry, and further research should investigate the impact of different geometries on device performance. The acceleration-feedback control experiments are conducted in a controlled laboratory setting; further research is necessary to evaluate the device's performance in real-world scenarios and under varied environmental conditions.