In humans or animals, nerve cells - or neurons - exchange information via connections called synapses, resulting in actions and thoughts. In a biological brain, electrical signals activate these neurons, whilst in a ‘fibre brain’ something similar can be achieved with optical signals. This technology could give rise to computers that think and learn just like humans - and which might maybe even surpass us.
In a ‘traditional’ neural network many different approaches might need to be combined to mimic brain performance. However, if this can be achieved with a single fibre, it becomes easy to scale up. Today’s computers are designed in way that will never allow them to process the same amount of information that a human brain can, whilst using very little power – a mere 20W, in fact. One possible solution is offered by Neuromorphic computing, using electronics to imitate brain biology. Earlier, fully electronic neuromorphic systems couldn’t provide the required bandwidth, but now a team of researchers at NTU has found out that low power, high-bandwidth neuromorphic chips can be based on GLSO (Gallium lanthanum oxysulphide) fibres. These neuromorphic chips, which only consume energy when a signal is fired, make it possible to build a highly dense web of connections.
Different polarization, intensities and wavelengths of light change the properties of Chalcogenide materials, which can mimic the firing of nerve cells. Light absorption is controlled by exposure to different wavelengths, an effect known as ‘photodarkening’, enabling control of the efficiency of light propagation through the fibre.
Real Artificial Intelligence
The human brain is estimated to have some 86 billion neurons. An artificial brain with the same amount of fibre-neurons might be much faster, as light travels faster than the chemicals in a human brain.
Researchers at the University of Southampton and the Nanyang Technological University (NTU) in Singapore believe that the artificial brain may even display certain traits of the biological brain, such as its ability to learn by association or through experience and make autonomous decisions. The breakthrough may also have far-reaching consequences for computing. If we can exchange today’s binary/digital systems for neural communication protocols, computers may become orders of magnitude faster. They might possibly become a lot smarter than us, too…
A biological neuron transmits information by electrical and chemical signals (see figure ‘A’). Nerve cells - or neurons - communicate by propagating ‘action potentials’ through extensions called axons, and releasing neurotransmitters at chemical synapses. A photonic axon and synapse (see figure ‘B’) transmits information with photo-modulated optical pulses along the fibre.
The paper published in the journal Advanced Optical Materials is entitled
“Amorphous Metal-Sulphide Microfibers Enable Photonic
Synapses for Brain-Like Computing” (Behrad Gholipour, Paul Bastock, Chris Craig, Khouler Khan, Dan Hewak and Cesare Soci) and can be found here.