pathway with a biomimetic implant integrated with the brain should recover the
lost function (i.e., the learning response).
To achieve the goal of the project, however, we need to make progress on related
component technologies:
■ development of effective physiological
recording strategies
■ robust detection of the stimulation signals from the data
■ delivery of input to a neuromorphic chip
that includes the cerebellar model.
We must develop a cerebellar model that
can mimic the damaged microcircuit and
learn from the stimuli the appropriate conditioned response output that will interface
with the biology and bring about the motor
eye-blink response. The current status of
each of these challenges is discussed below.
Clinical understanding
The success of our technological solution is
dependent upon our understanding of the
underlying biological system. Much effort
has been spent describing the cerebellar
microcircuit and its modulation by other
regions of the brain.
Tools have been developed to optimise
the quality of the recorded physiological
data. Silicon microelectrode arrays have
been developed with precise positioning
of the sensing pads that correspond to the
specific brain anatomy with which we are
concerned. Algorithmic methods are in
place to select the appropriate timing of the
stimulus delivery to facilitate learning and
FIGURE 2: Aggregation boards for interfacing the
electrodes to the physiological recording system
and the device that provides the stimulus to
action the eye-blink response are pictured.
These activities have supported development of the experimental paradigms
required for the practical implementation
of the biohybrid, such as simultaneous
recordings from multiple sites within the
brain.
Signal detection
The interface between the biology and
technology takes place in signal detection.
The detection of the onset and timing of
the CS and US must be extracted from
the physiological data in real time and be
provided to the biomimetic chip. The vast
amount of data present in the brain makes
this challenging. We are developing two
approaches to this problem: determining
the upper limit of information that can be
extracted from the data, and developing a
model-based signal detection methodology
that complies with the constraints of the
model and its hardware implementation.
The cerebellar model and hardware
implementation
The output of successful signal processing is the input to the synthetic cerebellar
model. The details of cerebellar anatomy
allow reliable bottom-up modelling of
cerebellar architecture. By contrast, the
physiology of the cerebellum allows top-down modelling of learning. A biologically constrained model was previously
described1 and has been further refined in
this project.
The CS and US identified from the
physiological data are fed into the synthetic
model, which must respond with a learning
of the conditioned response and provide
an output leading to stimulation of the
biohybrid’s facial nerve to bring about an
eye-blink response. For the first time, we
have demonstrated the practical real-time
bidirectional coupling between the physiological data and the synthetic system. The
cerebellar model has been shown to fully
substitute the function of the biological
cerebellar microcircuit in eye-blink conditioning and acquire the learning response.
Our task now is to implement signal
detection and the cerebellar model in
hardware form. An abstract version of the
model in field programmable gate array
(FPGA) chip form has demonstrated posi-
tive results. Current activity is focused on
implementation in an aVLSI chip, compat-
ible with physiological recording methods.
The way forward
The next stage for the project is critical.
The first series of integration experiments
interfacing multiple components of the system has been carried out. Promising results
have informed the planning of a large-scale
demonstration test in which a “
closed-loop” experiment will be conducted. This
integration of the complete system in a
biohybrid is the ultimate goal of the project
as it moves into its final year. These results
will put the project firmly on the path to
demonstrating the ReNaChip concept and
bringing the prospect of clinical therapy
one step closer. 1
References
1. C. Hofstötter et al. “The Cerebellum Chip: an
Analog VLSI Implementation of a Cerebellar
Model of Classical Conditioning.” Advances
in Neural Information Processing Systems,
577–584 (2005).
Angela Silmon, PhD
is ReNaChip Project Coordinator, Newcastle
University, INEX, Herschel Building, Newcastle-
Upon-Tyne NE1 7RU, UK
tel. + 44 1912 223 500.
For more information about the ReNa-
Chip project, visit www.renachip.org
or contact enquiries@renachip.org.
For more information on
design-related topics, go to
emdt.co.uk/categories/design
