Low-Cutoff Frequency Reduction in Neural Amplifiers: Analysis and Implementation in CMOS 65 nm

Fereidoon Hashemi Noshahr, Morteza Nabavi, Benoit Gosselin, Mohamad SawanFrontiers in Neuroscience 2021 · DOI

Abstract

Scaling down technology demotes the parameters of AC-coupled neural amplifiers, such as increasing the low-cutoff frequency due to the short-channel effects. To improve the low-cutoff frequency, one solution is to increase the feedback capacitors' value. This solution is not desirable, as the input capacitors have to be increased to maintain the same gain, which increases the area and decreases the input impedance of the neural amplifier. We analytically analyze the small-signal behavior of the neural amplifier and prove that the main reason for the increase of the low-cutoff frequency in advanced CMOS technologies is the reduction of the input resistance of the operational transconductance amplifier (OTA). We also show that the reduction of the input resistance of the OTA is due to the increase in the gate oxide leakage in the input transistors. In this paper, we explore this fact and propose two solutions to reduce the low-cutoff frequency without increasing the value of the feedback capacitor. The first solution is performed by only simulation and is called cross-coupled positive feedback that uses pseudoresistors to provide a negative resistance to increase the input resistance of the OTA. As an advantage, only standard CMOS transistors are used in this method. Simulation results show that a low-cutoff frequency of 1.5 Hz is achieved while the midband gain is 30.4 dB at 1 V. In addition, the power consumption is 0.6 μW. In the second method, we utilize thick-oxide MOS transistors in the input differential pair of the OTA. We designed and fabricated the second method in the 65 nm TSMC CMOS process. Measured results are obtained by in vitro recordings on slices of mouse brainstem. The measurement results show that the bandwidth is between 2 Hz and 5.6 kHz. The neural amplifier has 34.3 dB voltage gain in midband and consumes 3.63 μW at 1 V power supply. The measurement results show an input-referred noise of 6.1 μVrms and occupy 0.04 mm² silicon area.

Introduction

In the field of neural interfaces, high-fidelity recording of signals across a broad spectrum is essential. This includes high-frequency "spikes" (Action Potentials) and very low-frequency rhythms (Local Field Potentials, or LFPs). AC-coupled amplifiers are the standard for these recordings, where a critical parameter is the low-cutoff frequency ($f_L$). Ideally, for LFPs, $f_L$ should be below 1 Hz to avoid filtering out vital low-frequency brain activity.

However, as CMOS technology scales to 65nm and beyond, a significant physical challenge arises. Thinner gate oxides in smaller transistors lead to increased gate oxide leakage, which reduces the input resistance ($R_i$) of the Operational Transconductance Amplifier (OTA). This reduction in $R_i$ pushes the $f_L$ higher—sometimes to hundreds of Hz—effectively blinding the sensor to low-frequency signals.

Analysis of Frequency Response

The theoretical frequency response of a neural amplifier is typically defined by its feedback network, where $f_L$ is inversely proportional to the feedback resistance ($R_F$) and capacitance ($C_F$). Figure 2 illustrates these fundamental relationships.

Figure 2: Theoretical frequency response of the neural amplifier.

This research proves that in advanced nodes, the OTA's input resistance ($R_i$) becomes the dominant factor. As $R_i$ decreases due to gate leakage, the high-pass corner frequency shifts significantly. Figure 3 demonstrates this effect through simulation, showing how the bandwidth narrows as $R_i$ drops from 1 GΩ to 1 MΩ.

Figure 3: Impact of OTA input resistance on the low-cutoff frequency.

Proposed Solutions

To restore low-frequency performance without increasing the chip's physical footprint with massive capacitors, two strategies were developed:

  1. Cross-Coupled Positive Feedback (CCPF): A simulation-based approach using PMOS transistors as "pseudoresistors" to create negative resistance. This effectively cancels leakage effects and boosts the OTA's input resistance.
  2. Thick-Oxide Differential Pair: For the physical prototype, the team utilized "thick-oxide" MOS transistors in the OTA's input stage. These components have a more robust oxide layer that physically blocks gate leakage, maintaining high input resistance and a low $f_L$.

Experimental Results

The fabricated 65nm chip was validated using in vitro recordings from mouse brainstem slices. Key performance metrics include:

  • Low-Cutoff Frequency: 2 Hz (measured), a significant improvement over standard 65nm designs.
  • Midband Gain: 34.3 dB.
  • Power Consumption: 3.63 μW at 1V, suitable for implantable devices.
  • Area: 0.04 mm², allowing for high-density integration.
  • Noise: Input-referred noise of 6.1 μVrms.

Conclusion

By addressing gate oxide leakage as the root cause of high-pass filtering in scaled CMOS nodes, this work enables high-fidelity neural recording in compact, modern processes. Future developments may include chopper stabilization to further mitigate flicker noise at low frequencies.

Summary updated

Abstract

Scaling down technology demotes the parameters of AC-coupled neural amplifiers, such as increasing the low-cutoff frequency due to the short-channel effects. To improve the low-cutoff frequency, one solution is to increase the feedback capacitors' value. This solution is not desirable, as the input capacitors have to be increased to maintain the same gain, which increases the area and decreases the input impedance of the neural amplifier. We analytically analyze the small-signal behavior of the neural amplifier and prove that the main reason for the increase of the low-cutoff frequency in advanced CMOS technologies is the reduction of the input resistance of the operational transconductance amplifier (OTA). We also show that the reduction of the input resistance of the OTA is due to the increase in the gate oxide leakage in the input transistors. In this paper, we explore this fact and propose two solutions to reduce the low-cutoff frequency without increasing the value of the feedback capacitor. The first solution is performed by only simulation and is called cross-coupled positive feedback that uses pseudoresistors to provide a negative resistance to increase the input resistance of the OTA. As an advantage, only standard CMOS transistors are used in this method. Simulation results show that a low-cutoff frequency of 1.5 Hz is achieved while the midband gain is 30.4 dB at 1 V. In addition, the power consumption is 0.6 μW. In the second method, we utilize thick-oxide MOS transistors in the input differential pair of the OTA. We designed and fabricated the second method in the 65 nm TSMC CMOS process. Measured results are obtained by <i>in vitro</i> recordings on slices of mouse brainstem. The measurement results show that the bandwidth is between 2 Hz and 5.6 kHz. The neural amplifier has 34.3 dB voltage gain in midband and consumes 3.63 μ<i>W</i> at 1 V power supply. The measurement results show an input-referred noise of 6.1 μ<i>V</i> <sub><i>rms</i></sub> and occupy 0.04 <i>mm</i> <sup>2</sup> silicon area.
BibTeX
@inproceedings{noshahr2021low,
  author = {Fereidoon Hashemi Noshahr and Morteza Nabavi and Benoit Gosselin and Mohamad Sawan},
  title = {{Low-Cutoff Frequency Reduction in Neural Amplifiers: Analysis and Implementation in CMOS 65 nm}},
  booktitle = {Frontiers in Neuroscience},
  year = {2021},
  doi = {10.3389/fnins.2021.667846},
}
Figure