Neurochip3: An Autonomous Multichannel Bidirectional Brain-Computer Interface for Closed-Loop Activity-Dependent Stimulation

Abstract

Toward addressing many neuroprosthetic applications, the Neurochip3 (NC3) is a multichannel bidirectional brain-computer interface that operates autonomously and can support closed-loop activity-dependent stimulation. It consists of four circuit boards populated with off-the-shelf components and is sufficiently compact to be carried on the head of a non-human primate (NHP). NC3 has six main components: (1) an analog front-end with an Intan biophysical signal amplifier (16 differential or 32 single-ended channels) and a 3-axis accelerometer, (2) a digital control system comprised of a Cyclone V FPGA and Atmel SAM4 MCU, (3) a micro SD Card for 128 GB or more storage, (4) a 6-channel differential stimulator with ±60 V compliance, (5) a rechargeable battery pack supporting autonomous operation for up to 24 h and, (6) infrared transceiver and serial ports for communication. The NC3 and earlier versions have been successfully deployed in many closed-loop operations to induce synaptic plasticity and bridge lost biological connections, as well as deliver activity-dependent intracranial reinforcement. These paradigms to strengthen or replace impaired connections have many applications in neuroprosthetics and neurorehabilitation.

Keywords: brain-computer interface; closed-loop stimulation; neural recording; neural stimulation; neurochip; neuroprosthetics.

FIGURE 1

Neurochip3 boards. (A) Assembled boards, from top to bottom: input connections and amplifiers chip, MCU and FPGA, and stimulator control; back plane contains power supply, SD Card, IR communications port, and indicator LEDs. (B) NC3 with cylinder enclosure, signal input connector, and battery pack in cap.

FIGURE 2

Block diagram of major NC3 components. (A) Biophysical analog inputs pass through a 0.07-Hz high pass filter. Data from the analog front end is converted to digital format and delivered to the FPGA. Events from the MCU are relayed over GPIO pins to the FPGA where they can initiate stimulus waveforms previously setup by the MCU over a serial port. The FPGA writes data from the analog front end and MCU to a micro SD Card, and routes selected analog channels to the MCU over a parallel bus. The client computer sends settings to the NC3 over an infrared serial port, and downloads NC3 data from micro SD Cards. TGND, tissue ground; GPIO, general purpose input/output; DAC, digital-to-analog converter. (B) Circuit diagram of one of the positive “V to I converter” blocks of the stimulator. PHVR, Positive High-Voltage Return.

FIGURE 3

Optional battery combiner circuit uses four Schottky barrier rectifiers to allow full discharge of both batteries at only a small loss in total power delivered to the NC3.

FIGURE 4

Matlab-based graphical user interface on a PC. (A) Upper left panel contains controls for interacting with the NC3. Lower left tabbed panel contains settings for recording, event generation, spike discrimination, stimulation, impedance testing, and neural network simulation. Right panel shows data sweeps and summaries for checking performance and setting up correct event generation. It can also be used to review previously collected data. (B) Tabbed panels for record and stimulation settings.

FIGURE 5

Unit potential calculation for the integrate-and-fire neural network. Spiking input from connected units sum into a target unit’s potential Vj(t). If Vj(t) reaches threshold it is reset to 0 and the unit sends a spike to all of its target units, which will cause a Post Synaptic Potential proportional to Wij. These PSPs may be excitatory or inhibitory, depending on the sign of Wij (for more details see Shupe and Fetz, 2021).

FIGURE 6

Connections to brain and artifacts. (A) Single ended recording uses the same reference site for each channel. The NC3 stimulator employs a tissue ground and supports differential stimulation, which can reduce artifacts. (B) Stimulus artifacts that do not exceed the amplifier input range are very brief (<2 ms). (C) Examples of stimulus artifacts that may occur when stimulation saturates the input amplifier. (D) Artifacts are affected by the input filter’s low pass cut-off (fL). Increasing fL from 25 to 300 Hz shortens the artifact duration. (E) Example of a spontaneous spike (left) followed by a stimulus-evoked spike (right).

FIGURE 7

Multiple simultaneous closed-loop stimulation with NC3. As shown for the black circuit, descending pathways are partially lesioned by spinal cord injury. The paretic black muscle generates motor unit potentials that are converted to stimuli delivered to spinal neurons in the pathways to the black motoneuron (at intensities subthreshold for evoking movement). Comparable circuitry pertains to the blue and green motoneurons.

FIGURE 8

Simultaneous stimulation and multichannel recording of neural action potentials with NC3. Stimulus occurs at t = 0. Arrows indicate action potentials evoked by a 15 μA stimulus delivered through a neighboring electrode.

FIGURE 9

Vagal evoked potentials (VEPs) elicited during different behavioral states: (active-wake AW, red); resting-wake (RW, blue); rapid eye movement sleep (REM, green); and non-REM sleep (NREM, magenta). Representative averaged evoked potentials recorded from the following cortical sites are expanded at left: right prefrontal cortex (RPFC2), right ventrolateral nucleus of thalamus (RVL1), right somatosensory-motor cortex (RSM3), and right parietal cortex (RPC2). Time is measured from the first pulse of the 300 Hz train of 5 pulses delivered to the vagus nerve; colored areas designate the time range for specific VEP components (from Rembado et al., 2021).

FIGURE 10

Long-term recording of multiple neurons during sleep and wake. (A) Neural firing rates of three neurons recorded for 30 h during four putative behavioral states: active-wake (AW, blue); resting-wake (RW, orange); rapid eye movement sleep (REM, yellow); and non-REM sleep (NREM, purple). During the time between the two vertical dashed lines the animal room light was turned off (18:00–6:00). (B) Peristimulus time histograms normalized to average firing rate of three different neurons during the four classified states.

FIGURE 11

Implementation of a simple closed-loop BCI via an ANN. (A) Spike activity of simultaneously recorded input neurons (R1-R4) during stimuli delivered at outputs (S1-S4). Labels at top identify stimulus channels producing corresponding stimulus artifacts. (B) Location of recording and stimulating sites on Utah array. (C) Connectivity of spiking SNN showing excitatory (green) and inhibitory (red) connections from inputs (R1–R4) to outputs (S1–S4). Numbers are delays between output spikes and stimuli.