Solid-state neurons: are we getting closer to building artificial brains?
by Antonia Alalitei
Ever since neuroscience and technology substantially developed, scientists have slowly turned towards taking inspiration from biological neurons in creating new technological breakthroughs, as well as directly aiming to mimic the functionality of a neuron in a solid-state, analogue replica. It was not until the 1980s, however, that technology offered viable instruments that could sustain the recreation of a functional artificial neuron. Carver Mead was first to note the complementary metal-oxide-seminconductor (CMOS) transistors, digital circuits which can act as basic binary switch that turn on or off as the transistor gate voltage crosses some threshold, can express surprisingly similar current-voltage relationships as do neuronal ion channels, with little power expense. Thus, CMOS technology became the ideal tool for analogue neural function replication.
Not only was this a groundbreaking discovery, but it also inspired the creation of a dedicated field of neuromorphic engineering, describing the use of ver-large-scale integration (VLSI) systems containing analogue circuits to mimic biological architectures of the nervous system.
The scientific community has come a long way in 40 years, especially in taking inspiration from complex neural architectures to build digital systems that can act as artificial sensory transducers. However, in this article we will focus on the new developments related to solid-state neurons, precise replication of biological individual neurons and their functions.
This unique device physics led to the first descriptions of “neuromorphic” silicon neurons (SiNs) or solid-state neurons, which allow the action potential sequences of neurons to be directly emulated on analog electronic chips without the need for digital software simulations.
However, CMOS technology is facing many challenges when designing large-scale neural architectures, which would, in the end, be the actual main purpose of neuromorphic engineering: recreating an entire brain. Limitations such as the impossibility of adapting the characteristics of microchips after fabrication, a certain threshold voltage of operation, as well as the complexibility, adaptability and unpredictability of biological processes in the nervous system make the challenge still far from coming to reality.
Therefore, the need for further simulating and developing a robust model that would replicate the true physical behaviour of individual neurons is still the main milestone to overcome before moving on to developing fully analogue brains.
What’s the novelty?
Earlier this week, in a paper published in Nature Communications, a team of scientists from the University of Bath in the UK, in collaboration with scientists from the University of Bristol, the University of Zurich in Switzerland and the University of Auckland in New Zealand, have dived deeper into the creation of biophysical, analogue neuron models by creating an ultra low-power device that mimics a real neuron. The solid-state neurons “respond nearly identically to biological neurons under stimulation” of a computer program, by taking in a variety of current inputs, as well as outputting action potentials. The complete dynamics of hippocampal and respiratory neurons have been replicated in silico, which allows for hope of successful analogue implementations in the future.
By focusing on accurate simulations of individual neural models, the aim is to reach a state of artificial individual neural responses being indistinguishable from a natural neuron, thus possibly being used to replace parts of the brain that have been damaged by age and disease.
Stephan Furber, professor of computing at the University of Manchester, argues that currently, the level of detail of these neural models are impossible to use in complex neural architectures due to their laborious characteristics. However, as the paper also proposes, these devices can find applicability in simpler neural architectures of the nervous system, such as heart rate and breathing synchronization, where the adaptability of biological neurons to changing environment is vital to the overall health of the body. Thus, replacing a couple of neurons with their electronic chip artificial replicas could solve simple problems with complex long-term repercussions.
Next steps in developing this technology are testing the respective technology in its physical state, to account for real life conditions, as well as actually integrating it in a physical, living biological system. As individual neuronal models are starting to be produced with higher reliability, both in computational and physical models, more complex solid-state neuronal architectures could be developed, with the hope of a future where we could see successful bypasses of damaged brain regions with analogue electronic solid-state neuron microchips.