IEEE Circuits and Systems Magazine - Q4 2020 - 32

Wireless closed-loop optogenetics is a powerful paradigm in experimental
brain research that allows to regulate neural circuits in real time through
the combined utilization of smart implantable sensors and actuators.
utilization in an experimental setting with freely moving animals [12]-[14]. A hardwired CL system processing up to 32 channels (6.4 Mbps) in parallel and running
on a host computer is presented in [12], but this system
has a computing latency of 8 ms, which corresponds to
the duration of four APs typically, thus missing some
important circuit features and resulting in data losses.
Although these systems can perform complex and computationally intensive neural analysis, they are not practical to conduct freely moving experiments.
B. Existing Hardware Solutions
A vast majority of systems that provide parallel electrophysiological recording and optical stimulation have a
hardwired connection with the host computer, which
limits their utilization to fixed experimental settings
typically focusing on anesthetized animals (CL hardwired experimental settings with rats, primates and
unanesthetized humans have been reported [24],
[25]). Advanced systems have been designed to detect
the presence of APs, classify them according to their
shapes, and use the overall activity level to issue proportional optical stimulation via a controller [12], [15].
Since neurons are known to generate APs of similar
shape over time [26], signal classification allows the
estimate of the number of neurons nearby the recording sites, and to detect when these neurons are active.
Therefore, in a closed-loop scheme, the users can select which neuron can trigger the stimulation, or not.
Another method is to count the number of APs within a
time window, in order to detect the presence of bursts
of APs [13]. The presence of " bursts " of APs have shown
to play an essential role to understand the communication between neurons, and how a neural network is
recruited [27], [28], [55].
Another less computationally intensive technique
consists of closing the loop between the stimulation
with the neural recording modules by analyzing the low
frequency content of the neural signal, i.e. the local field
potential (LFP) signals, instead of looking at the higher
frequency APs. Proportional integral (PI) or proportional integral derivative (PID) regulators are sometimes
used to control the phase and the amplitude of LFP
signals [32]-[34]. These systems have found important
applications in the management of Parkinson's disease,
epilepsy, or spinal cord injury, using electrical stimulation [25], [32]-[34].
32â

III. Wireless CL System Based on
Short-Latency Neural DSP
A. Concept
In this wireless CL system, a link between the stimulator and the recording circuit is established by using a
" burst " neural activity classifier, combining the two key
techniques presented in [12] and [13], inside a wireless
system. First, the electrodes sense the low-voltage electrical activity of a group of neurons, i.e. APs, the voltage
of which is further conditioned and digitized by a neural
recording interface. Then, the single-cell neural signals
are passed through an AP detector [29]-[31]. Instead
of storing the raw APs, each detected waveform can be
separately compressed to reduce the amount of data to
be processed and stored. Each AP is then sorted according to its waveform, providing an approximation of how
many neurons are close to the electrodes and when these
neurons are active [26]. Then, using this information, the
firing rate, or the " bursting " activity of each neuron can
be estimated, and the stimulation pulses can be triggered
when the firing rate reaches a preconfigured value.
B. Hardware Overview
The concept of an electro-optic wireless system for CL
optogenetics implementing the technique previously
described in Section III-A is presented in Fig. 3(a). As
can be seen, this system includes all the required building blocks to perform neural signal conditioning, AP
detection, compression, and sorting and to issue optical stimulation pulses using the detected, compressed
and sorted AP information. Two custom microelectronic
chips implemented in CMOS 0.13-Âµm can be used to efficiently implement such a platform: i) a mixed-signal
chip to perform multichannel neural recording and optical stimulation in parallel, and ii) a Neural DSP chip
to provide real-time neural data decoding to establish
a connection between the readout and the stimulator of
the mixed-signal chip.
The mixed-mode chip block diagram, which includes
key circuits for electrophysiological recording, digitization and optogenetic stimulation is presented in Fig. 3(b),
while the Neural DSP block diagram is depicted in Fig. 3(c).
The Neural DSP module has three main cores: the Detector core, the Compression core, and the Sorting core to detect, compress and perform automatic AP sorting of the
AP waveforms in the chip. In addition to the mixed-signal

IEEE CIRCUITS AND SYSTEMS MAGAZINE

FOURTH QUARTER 2020

IEEE Circuits and Systems Magazine - Q4 2020

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q4 2020