Recent Advances in Artificial Sensory Neurons: Biological Fundamentals, Devices, Applications, and Challenges

Shuai Zhong, L. C. Su, Mingkun Xu, Desmond K. Loke, Bin Yu, Yishu Zhang, Rong ZhaoNano-Micro 2024 · DOI

TL;DR

This review paper explores the exciting field of Artificial Sensory Neurons (ASNs), which mimic biological sensory receptors to convert external stimuli into spike-based signals for energy-efficient neuromorphic computing. It highlights recent advancements in device technologies, diverse sensory modalities, and key challenges for their practical implementation.

What Problem Does This Paper Address?

Traditional sensors and analog-to-digital converters (ADCs) used in neuromorphic systems are often power-hungry and occupy significant space, hindering their application in edge computing and the Internet of Things (IoT). The paper addresses this by focusing on Artificial Sensory Neurons (ASNs), which integrate sensing and spike conversion capabilities using emerging devices. This approach aims to overcome the limitations of conventional systems by offering higher energy efficiency and a more biologically plausible way of processing sensory information.

Methods

The paper provides a comprehensive review of ASNs, covering:

  • Biological Fundamentals: It delves into the mechanisms of biological signal transduction for various senses (visual, tactile, thermal, gustatory, olfactory, auditory, nociceptive, collision avoidance).
  • Emerging Devices: It discusses various novel devices used to implement ASNs, including memristors (diffusive and Mott), single transistor latch-based field-effect transistors (STLFETs), and 2D memtransistors.
  • ASN Implementations: It categorizes ASNs based on the sensory modality they emulate (tactile, thermal, acoustic, gustatory, olfactory, biochemical, visual, multimodal, nociceptive, collision avoidance).
  • Performance Metrics: It outlines key metrics for evaluating ASNs, such as power consumption, sensitivity, dynamic range, linearity, response time, resolution, and stability.
  • Applications: It explores potential applications in machine learning, robotics, prosthetics, and autonomous systems.

Datasets

Not discussed in the paper. The paper is a review and does not present new experimental results or datasets.

Key Insights and Ideas

  • In-Sensor Computing: ASNs represent a significant step towards "in-sensor computing," where sensing and signal processing (spike generation) are integrated into a single device, drastically reducing power consumption and latency.
  • Beyond CMOS: Emerging devices like memristors and STLFETs offer unique advantages over traditional CMOS technology for neuromorphic applications, including non-volatility, analog behavior, and compact form factors.
  • Biological Inspiration is Key: Understanding the intricate mechanisms of biological sensory receptors is crucial for designing effective ASNs that can accurately mimic natural perception.
  • Multimodal Integration is the Future: The paper emphasizes the growing importance of multimodal ASNs that can simultaneously process information from multiple senses, leading to more robust and intelligent systems.
  • Challenges Remain: Despite significant progress, challenges in device uniformity, scalability, environmental stability, and efficient integration with neural networks still need to be addressed for widespread adoption.

Key Results

The paper does not present specific quantitative results from a single experiment but rather synthesizes findings from numerous studies. Key qualitative findings include:

  • Demonstrations of ASNs capable of encoding various stimuli (pressure, temperature, light, chemicals, sound, ions) into spike trains.
  • Development of ASNs that mimic specific biological phenomena like nociceptive sensation (threshold, relaxation, allodynia, hyperalgesia) and collision avoidance.
  • Successful integration of ASNs into systems for tasks like gesture recognition, tumor classification, gas identification, and robotic arm control.
  • Performance comparisons of different emerging devices (memristors, STLFETs, 2D memtransistors) across metrics like area, CMOS compatibility, uniformity, power consumption, and spike frequency.

Limitations and Future Work

The authors acknowledge several limitations and suggest future research directions:

  • Device Level:
    • CMOS Compatibility: Many emerging devices are not yet fully compatible with standard CMOS fabrication processes, hindering large-scale integration.
  • Uniformity and Stability: Diffusive memristors suffer from significant device-to-device variations, while Mott memristors and other devices can be susceptible to thermal instability.
  • Power Consumption: While ASNs are generally more energy-efficient than conventional ADCs, further reduction in power consumption is needed, especially for Mott memristors.
  • Scalability: Achieving high-density integration of ASNs remains a challenge.
  • ASN Level:
    • Sensor Area: The area of the sensor component often dominates the overall ASN footprint, leading to inefficiency. Developing 2-in-1 sensing-and-encoding devices is crucial.
  • Dynamic Range and Sensitivity: Many ASNs have narrower dynamic ranges and lower sensitivity compared to biological counterparts, limiting their application in certain scenarios.
  • Response Time: The response time of ASNs is often limited by the sensor component, which can be orders of magnitude slower than the neuromorphic devices.
  • Resolution: Achieving high resolution in ASNs is challenging and often requires complex processing.
  • Multimodal Fusion: While progress has been made, achieving true multimodal fusion and accurate interpretation of combined sensory inputs is still an area of active research.
  • Array Level:
    • Crosstalk: Neighboring sensors in an array can lead to unwanted signal interference.
  • Thermal Effects: The use of massive resistors and capacitors in arrays can lead to significant thermal issues.
  • Multimodal Arrays: Demonstrating multimodal functionality at the array level is still nascent.
  • System and Application Level:
    • Integration with Neural Networks: Developing efficient algorithms for training spike-based neural networks and creating suitable neuromorphic hardware is essential.
  • Biocompatibility: For applications like neural interfaces, ensuring the biocompatibility of ASNs is critical.
  • Understanding Biological Complexity: Mimicking the full complexity of biological neural systems, especially at the molecular level, remains a long-term goal.

Future work should focus on addressing these limitations through interdisciplinary collaboration, material innovation, device engineering, and advanced circuit design to unlock the full potential of ASNs for next-generation intelligent systems.

Summary updated

Abstract

Spike-based neural networks, which use spikes or action potentials to represent information, have gained a lot of attention because of their high energy efficiency and low power consumption. To fully leverage its advantages, converting the external analog signals to spikes is an essential prerequisite. Conventional approaches including analog-to-digital converters or ring oscillators, and sensors suffer from high power and area costs. Recent efforts are devoted to constructing artificial sensory neurons based on emerging devices inspired by the biological sensory system. They can simultaneously perform sensing and spike conversion, overcoming the deficiencies of traditional sensory systems. This review summarizes and benchmarks the recent progress of artificial sensory neurons. It starts with the presentation of various mechanisms of biological signal transduction, followed by the systematic introduction of the emerging devices employed for artificial sensory neurons. Furthermore, the implementations with different perceptual capabilities are briefly outlined and the key metrics and potential applications are also provided. Finally, we highlight the challenges and perspectives for the future development of artificial sensory neurons.
BibTeX
@inproceedings{zhong2024recent,
  author = {Shuai Zhong and L. C. Su and Mingkun Xu and Desmond K. Loke and Bin Yu and Yishu Zhang and Rong Zhao},
  title = {{Recent Advances in Artificial Sensory Neurons: Biological Fundamentals, Devices, Applications, and Challenges}},
  booktitle = {Nano-Micro},
  year = {2024},
  doi = {10.1007/s40820-024-01550-x},
}
Figure