A new era in brain-inspired computing
Computing technologies have reached unprecedented speed and computational power that allows them to simulate parts of animals’ brains and behaviours. However, in order to truly emulate animal intelligence, the energy required by these systems grows exponentially. The brain contains an extremely high number of synapses while neurons are plastic and adaptive. Strikingly, the brain is structured as an evolving system where synapses emerge and die as well as strengthen and weaken. These essentially reconfigure neuronal connectivity and allow the neural network to adapt motor and behavioural responses to the ever-changing environmental inputs. For years, the implementation of artificial neural networks has been in the form of software run on conventional computers. Recently, neuromorphic computation schemes have emerged that utilise nanoscale electronic devices capable of simulating neuronal and synaptic properties. A bio-hybrid system between artificial and real neurons The EU-funded RAMP project proposed to develop a bio-hybrid system that interfaces artificial neural networks to biological ones. ‘Our rationale was to exploit the intrinsic properties of real neurons and synapses as well as their organisation into neural circuits,’ explains project coordinator Prof. Vassanelli. Artificial neurons were formulated as silicon microchips and physically interfaced to natural neurons through electrical transducers, forming a bio-hybrid neurochip. Since neurons are electrically active excitable cells, transducers can record their electrical activity or facilitate their stimulation. For this purpose, researchers exploited one of the most promising technologies, the memristive integrator sensor (MIS), which can operate as a smart neuronal sensor by simultaneously detecting and encoding electrical signals in a communication set-up with real neurons. To emulate natural synapse behaviour, MISs were generated that could encode and compress neuronal spiking activity. Signals recorded from neurons through conventional extracellular electrodes are fed to the MIS, which senses neuronal spikes encoding them as changes in internal resistance in a similar way to what happens in real synapses. ‘Both in the case of the MIS and the synapse such individual changes of resistance are integrated over time upon arrival of a series of spikes,’ continues Prof. Vassanelli. In a parallel effort to reproduce in the artificial part of the bio-hybrid what happens in real synapses – where connectivity depends on neuronal activity – scientists generated neurobiology-inspired algorithms. In essence, the number of connections between artificial neurons was a function of activity emulating brain plasticity. Future applications MISs comprise invaluable tools for exploring brain computation and for reading out brain activity. Monitoring neuronal cell activity is fundamental to neuroscience, but processing of neural data in real-time poses restrictive requirements in bandwidth, energy and computation capacity. The memristive devices consume little energy in the encoding process and thus not only address this bottleneck but will also lead to a better understanding of the biophysical basis of information processing in real neuronal circuits. Long term, these elements could be employed to create innovative neuroprostheses such as in brain implants, where artificial neuromorphic circuits replace or assist native brain networks. Such prosthetics could be used in patients with neurological disorders for treatment and rehabilitation purposes. Intriguingly, to speed up development RAMP partners engineered a way to let neurons and MISs communicate through the internet. This opens a revolutionary way to build neuroelectronics networks across Europe. Overall, the RAMP system provides a first proof-of-concept that neurons can mutually interact with nanoelectronic memristive devices sharing similar memory and plasticity rules. The patent applications and start-up companies that emerged during the project further attest to the RAMP innovation. RAMP was officially funded through the European Commission’s Future and Emerging Technologies (FET) programme.
Keywords
Neurons, brain plasticity, neural networks, RAMP, memristive