TL;DR
This paper introduces a novel, ultra-low-power, low-noise operational transconductance amplifier (OTA) designed for implantable neural recording systems. By combining a recycling folded-cascode topology with positive feedback and a mixed N/P-type MOSFET input stage, the amplifier achieves high gain and bandwidth while consuming minimal power, making it ideal for amplifying faint neural signals.
What Problem Does This Paper Address?
Implantable neural recording microsystems are crucial for understanding brain activity, diagnosing neurological diseases, and developing neuroprosthetics. A key component of these systems is the analog front-end amplifier, which must amplify weak neural signals (typically up to 1 mV) in the 0.1 Hz to 10 kHz range while filtering out noise. Existing challenges include achieving high gain, low noise, low power consumption, and a small chip area simultaneously, as these parameters often involve trade-offs. Specifically, designing an amplifier that can effectively capture both action potentials (APs) and local field potentials (LFPs) with minimal power draw for long-term implantation is a significant hurdle.
Methods
The authors propose a single-stage, fully differential Recycling Folded-Cascode (RFC) Operational Transconductance Amplifier (OTA). Key innovations include:
- Recycling Folded-Cascode Topology: This architecture reuses current sources to improve efficiency and reduce power consumption.
- Positive Feedback: Cross-coupled transistors are used to introduce local positive feedback, significantly boosting the DC voltage gain and improving linearity.
- Mixed N-type and P-type MOSFET Input Stage: Unlike traditional all-P-type inputs, this design uses both N-type and P-type MOSFETs to expand the input common-mode range, which is beneficial for neural signal acquisition.
- Subthreshold Operation: Input transistors are biased in the subthreshold region to minimize flicker noise and power consumption.
- CMFB Circuit: A Common-Mode Feedback circuit is implemented to set the output common-mode voltage, ensuring stable operation.
The design was implemented in a 0.18 µm CMOS process.
Datasets
Not discussed in the paper. The performance is evaluated through post-layout simulations.
Key Insights and Ideas
- The Power of Positive Feedback in OTAs: The paper demonstrates how carefully implemented positive feedback can dramatically increase DC gain and improve linearity in an OTA, a crucial aspect for amplifying weak neural signals.
- Mixed Input Stage for Wider Range: Using both N-type and P-type MOSFETs in the input stage is a clever way to extend the common-mode input range without significantly compromising noise performance, which is vital for capturing diverse neural signals.
- Subthreshold Operation for Low Noise and Power: Biasing input transistors in the subthreshold region is a well-established technique for reducing flicker noise and power, but its effective application here, combined with other techniques, yields impressive results.
- Trade-offs are Inevitable, but Manageable: The "analog design octagon" (Figure 2) highlights the inherent trade-offs. This work shows that by intelligently combining multiple techniques, a favorable balance can be struck for neural recording applications.
Key Results
The proposed RFC OTA demonstrates excellent performance metrics:
- Open-Loop Performance:
- DC Voltage Gain: 97.23 dB
- Unity-Gain Bandwidth (UGBW): 2.91 MHz (with 1 pF load)
- Input-Referred Noise (IRN): 4.75 µVrms
- Total Power Dissipation: 5.43 µW (including bias circuitry)
- Chip Area: 0.0055 mm²
- Integrated Neural Recording Amplifier (NRA) Performance:
- Simulated IRN: 5.73 µVrms (1 Hz to 10 kHz bandwidth)
- Power Consumption: 5.6 µW
- Output Voltage Swing: 0.976 Vpp
- Total Harmonic Distortion (THD): -62.68 dB at 1 kHz
These results are achieved using a 0.18 µm CMOS process with a 1.8 V supply. The performance is competitive with or superior to existing state-of-the-art neural recording amplifiers, particularly in terms of power efficiency and noise.
Limitations and Future Work
- Process Variations: While corner analysis and Monte Carlo simulations were performed, the authors acknowledge that real-world performance can vary. Future work could explore more robust design techniques to mitigate these variations further.
- Noise Optimization: Although the IRN is low, further optimization could be achieved by increasing the W/L ratio of input transistors (at the cost of area and power) or by implementing advanced noise reduction techniques like chopper stabilization or correlated double sampling, which would add complexity.
- Technology Scaling: The paper uses a 0.18 µm CMOS process. While mature and cost-effective, newer technologies might offer further improvements in power and noise. Future work could investigate the design in more advanced process nodes.
- Experimental Validation: The results presented are based on post-layout simulations. Experimental validation on fabricated chips would be the next crucial step to confirm these findings.
Summary updated