Skip to main content
eScholarship
Open Access Publications from the University of California

Implantable Neural Recording Front-Ends for Closed-Loop Neuromodulation Systems

  • Author(s): Chandrakumar, Hariprasad
  • Advisor(s): Marković, Dejan
  • et al.
No data is associated with this publication.
Abstract

The goal of neuromodulation is to alter neural activity through targeted delivery of a stimulus to specific sites in the body. A prominent example of neuromodulation is deep brain stimulation (DBS), which has proved effective in mitigating the effects of certain neurological conditions. However, existing neuromodulation treatments lack real-time feedback (simultaneous sensing) to adapt stimulation parameters in response to brain dynamics. Hence, neuroscientists and clinicians aim to perform closed-loop neuromodulation, where stimulation can be optimally controlled in real time for better treatment outcomes. In recent years, the community has emphasized closed-loop neuromodulation as a highly desirable tool for administering therapy in patients suffering from drug-resistant neurological ailments. A miniaturized autonomous implant would be instrumental in ensuring that neuromodulation achieves its full potential.

A key requirement for any closed-loop neuromodulation system is the ability to record neural signals while concurrently performing stimulation. However, neural stimulation generates large differential and common-mode artifacts at the recording sites, which easily saturate existing implantable recording front-ends due to their limited linear input range. To observe the neural response during stimulation, the front-end must faithfully digitize neural signals in the presence of large stimulation artifacts. The front-end must also satisfy strict constraints on power consumption, noise and input impedance, while achieving a small form-factor. State-of-the-art neural recording front-ends do not meet these requirements.

This work presents a recording front-end that can digitize neural signals in the presence of 200mVpp differential artifacts and 700mVpp common-mode artifacts. The front-end consists of a chopper amplifier and a 15.2b-ENOB continuous-time delta-sigma ADC. In the design of the chopper amplifier, new techniques have been proposed that introduce immunity to common-mode interference, increase the DC input impedance (Zin) of the chopper amplifier to 1.5GΩ, and enable the realization of large resistances (90GΩ) on-chip in a small area for filtering electrode offsets. In the design of the delta-sigma ADC, a modified loop-filter is used along with new linearization techniques to significantly reduce power consumption in the ADC. These techniques enable our recording front-end to achieve a dynamic range of 90dB (14b ENOB) in 1Hz - 200Hz, and 81dB (12.7b ENOB) in 1Hz - 5kHz. Implemented in a 40nm CMOS process, the prototype occupies an area of 0.113mm2/channel, consumes 7.3�W from a 1.2V supply, and can digitize neural signals from 1Hz to 5kHz. The input-referred noise is 1.8�Vrms (1Hz - 200Hz) and 6.35�Vrms (1Hz - 5kHz). The total harmonic distortion for a 200mVpp input at 1kHz is −81dB. Compared to state-of-the-art neural recording front-ends, this work improves Zin by 24.2x (for chopped front-ends), the linear-input range by 2x, the signal bandwidth (BW) by 10x, the dynamic range by 12.6dB, and tolerance to common-mode interferers by 6.5x, while maintaining comparable power and noise performance. The ADC alone consumes 4.5�W, has Zin of 20MΩ, BW of 5kHz, and achieves a peak SNDR of 93.5dB for a 1.77Vpp differential input at 1kHz. The ADC’s Schreier FOM (using SNDR) is 184dB, which is 6dB higher than the state-of-the-art in high-resolution ADCs.

Main Content

This item is under embargo until March 26, 2020.