Recent Trends and Future Prospects of Neural Recording Circuits and Systems: A Tutorial Brief

Jinbo Chen, Mahdi Tarkhan, Hui Wu, Fereidoon Hashemi Noshahr, Jie Yang, Mohamad SawanIEEE Trans. Circuits Syst. II Express Briefs 2022 · OpenReview

Abstract: Recent years have seen fast advances in neural recording circuits and systems as they offer a promising way to investigate real-time brain monitoring and the closed-loop modulation of psychological disorders and neurodegenerative diseases. In this context, this tutorial brief presents a concise overview of concepts and design methodologies of neural recording, highlighting neural signal characteristics, system-level specifications and architectures, circuit-level implementation, and noise reduction techniques. Future trends and challenges of neural recording are finally discussed.

The Challenge of High-Density Neural Recording

The human brain is a complex network of approximately $10^{11}$ neurons. To decode brain function or treat neurodegenerative diseases, researchers must record from thousands of these neurons simultaneously in real-time. However, designing implantable circuits for this purpose presents a significant engineering challenge defined by three conflicting constraints:

  1. Power & Heat: Implants must consume minimal power—typically less than $50 \mu W$ per channel—to prevent thermal damage to sensitive brain tissue.
  2. Size: Achieving high spatial resolution requires an incredibly small footprint, with the target area per recording channel often being less than $0.05 mm^2$.
  3. Signal Integrity: Neural signals are minute (ranging from microvolts to millivolts) and exist in environments plagued by low-frequency flicker noise and significant DC offsets from the electrodes.

This tutorial provides a roadmap for navigating these trade-offs, shifting from traditional single-channel designs toward massively parallel, intelligent systems.

System and Circuit Methodologies

The paper reviews several key methodologies across the system hierarchy to address these constraints:

  • System Architectures: Designers often choose between traditional Time-Division Multiplexing (TDM), where multiple channels share a single ADC, and Frequency-Division Multiplexing (FDM). Newer "rapid multiplexing" techniques at the electrode level are also emerging to reduce redundant wiring.
  • Neural Amplifiers: A primary choice exists between AC-coupled amplifiers, which use bulky capacitors to block DC offsets, and DC-coupled versions that utilize active feedback loops to save silicon area.
  • Analog-to-Digital Converters (ADCs): While the Successive Approximation Register (SAR) ADC remains a standard, the paper highlights the advantages of Level-Crossing ADCs (LC-ADC). Unlike Nyquist sampling at fixed intervals, LC-ADCs only trigger a sample when the signal crosses a specific threshold, significantly reducing power consumption during periods of neural inactivity.
  • Noise Reduction: Chopper Stabilization is identified as a superior technique for neural applications. By modulating the signal to a higher frequency, it avoids $1/f$ (flicker) noise without the "noise fold-over" (aliasing) issues common in autozeroing techniques.

Key Insights: Leveraging Signal Sparsity

A central insight of modern neural recording is that neural signals are inherently "sparse," consisting of brief electrical spikes followed by long silent periods. Future low-power designs must move toward "event-driven" hardware that ignores this silence.

Level-crossing sampling strategy exploiting signal sparsity to reduce ADC energy consumption.

Beyond sparsity, two other major trends are shaping the field:

  • Direct Digitalization: As CMOS technology scales, digital components shrink faster than analog ones. Moving the ADC closer to the electrode—sometimes eliminating the dedicated pre-amplifier—can save massive amounts of silicon real estate.
  • Compressive Sensing (CS): By using random measurement matrices, signals can be sampled below the Nyquist rate. This shifts the heavy computational burden of reconstruction from the power-constrained implant to an external system.

Key Results and Performance Benchmarks

The paper establishes a "Gold Standard" for target specifications in neural recording: input-referred noise should be less than $10 \mu V_{rms}$ to distinguish signals from biological background noise, and a Common-Mode Rejection Ratio (CMRR) of over $80 dB$ is necessary to filter environmental interference.

Increasing trend of neural recording research publications from 2012 to 2021.

The field is expanding rapidly; the number of neural recording papers on IEEE Xplore tripled between 2012 and 2021. State-of-the-art systems like Neuropixels 2.0 now record from over 10,000 channels in parallel, utilizing the miniaturization techniques discussed in this brief.

Future Frontiers

Despite recent progress, several challenges remain for the next generation of Brain-Machine Interfaces (BMIs):

  • Analog-to-Information (A2I) Conversion: Future chips should perform on-chip feature extraction (e.g., spike sorting) to reduce the wireless bandwidth required for data transmission.
  • Multimodal Recording: Systems must evolve to record electrical, chemical (neurotransmitters), and optical signals simultaneously for true closed-loop modulation.
  • Tissue-Electrode Interface: While circuit noise is being addressed, the long-term physical stability of the connection between silicon and brain tissue remains a hurdle for chronic, decades-long implantation.

Summary updated

Abstract

Recent years have seen fast advances in neural recording circuits and systems as they offer a promising way to investigate real-time brain monitoring and the closed-loop modulation of psychological disorders and neurodegenerative diseases. In this context, this tutorial brief presents a concise overview of concepts and design methodologies of neural recording, highlighting neural signal characteristics, system-level specifications and architectures, circuit-level implementation, and noise reduction techniques. Future trends and challenges of neural recording are finally discussed.
BibTeX
@article{chen2022recent,
  author = {Jinbo Chen and Mahdi Tarkhan and Hui Wu and Fereidoon Hashemi Noshahr and Jie Yang and Mohamad Sawan},
  title = {{Recent Trends and Future Prospects of Neural Recording Circuits and Systems: A Tutorial Brief}},
  journal = {IEEE Trans. Circuits Syst. II Express Briefs},
  year = {2022},
  url = {https://openreview.net/forum?id=YjtXTYau7q},
}
Figure