An AC-Coupled Wideband Neural Recording Front-End With Sub-1 mm²×fJ/conv-step Efficiency and 0.97 NEF

Abstract

This paper presents an energy-and-area-efficient ACcoupled front-end for multichannel recording of wideband neural signals.The proposed unit conditions local field and action potentials using an inverter-based capacitively-coupled low-noise amplifier, followed by a perchannel 10-bit asynchronous SAR ADC. The adaptation of unit-length capacitors minimizes the ADC area and relaxes the amplifier gain so that small coupling capacitors can be integrated. The prototype in 65nm CMOS achieves 4× smaller area and 3× higher energy-area efficiency compared to the state of the art with 164 µm×40 µm footprint and 0.78 mm 2 ×fJ/conv-step energy-area figure of merit. The measured 0.65 µW power consumption and 3.1 µVrms input-referred noise within 1Hz-10kHz bandwidth correspond to a noise efficiency factor of 0.97. Index Terms-AC coupling, front-end, inverter-based amplifier, monotonic switching, neural recording, successive approximation register analog-to-digital-converter, unit length capacitor.

An Area- and Energy-Efficient 16-Channel, AC-Coupled Neural Recording Analog Frontend for High-Density Multichannel Neural Recordings

“…Analog frontend integrated circuit architectures for neural recordings to achieve a decent performance while maintaining the balance between power and area consumptions have, recently, been extensively researched. As summarized in [11], one can roughly classify these into the three distinct circuit architectures: (1) AC-coupled amplifier and analog-to-digital converter (ADC), (2) DC-coupled amplifier and ADC, and (3) DC-coupled direct ADC. The first architecture may be one of the most popular and widely adopted topologies [4,6,10,[24][25][26][27].…”

“…This work shows a decent channel FoM and E-A FoM of 147.87 fJ/c-s and 1.77 fJ/c-s•mm 2 , respectively, while some essential performance for neural recordings, such as the bandwidth and IRN, are comparable or better than other recent works. Particularly, this work shows the smallest footprint of 0.012 mm 2 per channel except for that in [11]. While [11] considers only the active circuits into their estimation, this work takes the active circuits and LNA input pads that will be coupled to a neural probe into consideration, thus, this work could use the area more effectively than [11].…”

“…As such, the requirements are not easy to achieve and are even harder when it is needed to achieve the requirements are not easy to achieve and are even harder when it is needed to achieve them simultaneously since they are interrelated in the context of integrated circuit design trade-offs [5]. To break the trade-offs, a lot of research into the neural recording frontend circuit design have been conducted [6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21]. Some works have accomplished an ultralowpower operation [18], others have realized the frontend circuit within a tiny area [8,19], and, recently, a few works have tried to achieve both simultaneously [9,11].…”

“…To break the trade-offs, a lot of research into the neural recording frontend circuit design have been conducted [6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21]. Some works have accomplished an ultralowpower operation [18], others have realized the frontend circuit within a tiny area [8,19], and, recently, a few works have tried to achieve both simultaneously [9,11].…”

Analysis and Reduction of Nonlinear Distortion in AC-Coupled CMOS Neural Amplifiers with Tunable Cutoff Frequencies

“…In overall, the architecture shown in Figure 1 remains the gold standard for modern neural amplifiers [26][27][28]. Many designs based on this circuit idea, with excellent noise performance and low power consumption, have been reported; for a review see [29,30].…”