Skip to main content

A SYnaptically connected brain-silicon Neural Closed-loop Hybrid system

Periodic Reporting for period 2 - SYNCH (A SYnaptically connected brain-silicon Neural Closed-loop Hybrid system)

Reporting period: 2020-01-01 to 2021-12-31

The SYNCH project aims at introducing a novel concept and technology of brain-computer interfacing to help treating neurological disorders. The core idea is to connect an artificial network of electronic neurons realized on a silicon chip to brain neurons. The connection between silicon and biological neurons is established through memristors, electrical nanodevices that emulate the function of brain synapses. Artificial neurons and brain neurons are reciprocally connected, such as biological neurons provide inputs to the artificial neurons and, along the return pathway, artificial neurons modulate the activity of brain neurons. Brain neurons are physically connected to memristors and to their electronic counterpart through a neural interface implanted in the brain, which enables the recording and adaptive electrical stimulation of neurons at the site of implantation. As such, the system constitutes a brain-computer interface, where the PC is replaced by a brain-inspired architecture able to 'speak' the spikes-based language of real neurons and where the connection strength between artificial neurons and biological neurons varies according to synaptic plasticity rules.
We envisage that this technology will be the basis for implantable devices where artificial neurons on-chip can rescue dysfunctions of brain networks that are focally damaged by injury – e.g. as a consequence of stroke – or by neurological diseases as in Parkinson’s and dementia. The approach is validated in a rat model of neurological disease, where a deficit of dopamine in deep brain structures causes neurological deficits, including learning impairment.
As such, the SYNCH project promises to open new avenues for the treatment of neurological disorders, by improving and expanding the use of brain nerumodulation to treat a variety of diseases that are relevant for the European citizens and that include Parkinson’s, epilepsy and stroke.
During these first 3 years of the 4.5 years of the project, we have set the foundations of the new technology, working on the components of the system and implementing and testing the first intercommunication infrastructure and protocols.
Regarding the components, a major activity was dedicated to the development and testing of the control unit, that is the pivotal component of the system that is responsible for routing the signals between brain neurons and artificial neurons on chip and for managing the synapse-inspired link through the memristors. A first control unit prototype was developed and is now available based on an FPGA. Manufacturing of memristors has been improved and a new control board for enhanced management of the memristors arrays has been designed. Interfaces for connecting the silicon neurons and the implanted neural probe to the control unit have been updated.
A major effort was dedicated to establishing the components and intercommunication protocols of the SYNCH platform, performing partial integration and validation trials in preparation of full integration in the last period, assessing applications of neuromorphic processing of brain activity – spikes and local field potentials – with memristors and artificial network of spiking neurons.
From the theoretical point of view, adaptation algorithms have been developed and preliminary tested in the artificial neuronal network of spiking neurons to recognize brain signals as recorded in vivo from the rat somatosensory cortex. Known sensory inputs were generated by deflecting a single whisker of the rat by different amplitudes, and the resulting neuronal activity recorded in the brain by the implanted neural probe. The artificial neuronal network, once it was trained to classify signals generated by whisker deflection, successfully recognized whisker deflections of different amplitudes, which represents a good starting point toward the final project goal. We also identified neural patterns in the midbrain that are candidates for the SYNCH system (‘brain-to-artificial network’ branch of the loop) as detectable signatures of reward-based learning. In parallel, we continued investigating microstimulation protocols and strategies to drive activity of a target brain circuit, in the view of implementing adaptive stimulation of the midbrain ('artificial neural network-to-brain' branch of the loop).
We have investigated and characterized the activity of neurons in barrel cortex and in the basal ganglia (ventral striatum and ventral tegmental area) of the rat to assess whether relevant information about sensory inputs and reward cues can be retrieved from measured signals. We accumulated preliminary evidence that this is indeed possible, which represents an important result for the success of SYNCH. In this context, we have investigated how spontaneous brain activity induces variability in the response of brain cortical networks to sensory inputs and identified strategies to improve classification of sensory inputs even during epochs of high brain activity. We started ethics and social assessments with respect to the topic AI-Brain interaction.
With respect to components, the control unit is a new and highly versatile prototype for the management of event-based signals from and to the brain as well as from and to artificial neuronal networks of spiking neurons. It is the most advanced solution available that is capable to include both memristors arrays and neuromorphic neurons in the processing chain. Memristors arrays have been improved and the new control board that was developed within the project, and that will be manufactured early next year, outperforms any existing and commercially available equipment. We have also developed an advanced finite element model to simulate the effect of electrical stimulation with microelectrodes in the brain. To our knowledge, this represents a novelty, in particular with respect to capacitive microelectrodes and, after appropriate consolidation and refinement in the next year of the project, it may become a useful tool to improve the application of neurostimulation techniques in clinics. We have explored new theoretical avenues to take advantage of spiking neuronal networks to process analog signals in an adaptive manner, by implementing adaptation algorithms and exploring reward-based approaches and backpropagation in recurrent neuronal networks of spiking neurons. These represent important results toward the future creation of neuromorphic prostheses that can take advantage from spiking neuronal networks to process brain signals and adapt stimulation in real-time.
We have performed electrophysiological studies in the barrel cortex and in the basal ganglia to enable the extraction on useful information about sensory inputs and reward cues. We have obtained novel results on the comparison between two types of signals generated by brain networks during activity (local field potentials and spikes) and on their respective information content. Furthermore, we have provided evidence for the first time that reward cues can be captured from recordings of the ventral striatum and ventral tegmental area, both belonging to the reward circuit in the rat brain. Taken together, these electrophysiology observations form a new and solid basis to attack the final project objective in the coming years.
kick-off meeting in Venice in February 2019
scheme of the project objectives
concept of the project