The researchers from IBM managed to create both artificial neurons and synapses that simulate the learning capacity of the brain.
The science project involved the creation of artificial neurons that spike randomly, by using materials with changing phases that can store data and process it.
The device will further be used in cognitive computing, as the highly dense neuro network is of extreme value in processing information. The Internet of Things, social media, and stock market trades could all benefit from the invention.
The researchers worked for ten years on the artificial neurons, and there would be several years before the phase change memory chip would be available on the market.
To imitate microscopic biological systems had been a challenge for the science world, as the very low voltage and the high densities of neuron systems could hardly be replicated in the lab. However, the scientists used the random property of the neuron network to create the artificial synapse system.
Just like short voltages excite neurons through synapses, the researchers used nanosecond pulses to change the material. In computing, every neuron responds differently and creates new ways to compute and represent signals.
The scientists used the process of crystallization and amorphization to induce variation.
The artificial neurons have only 90 nanometers in size, and they can be reduced to even 14 nanometers. They can sustain billions of switching cycles and could substitute years of operations.
The electric pulse used was less than five picojoule, and the power was less than 120 microwatts when a lightbulb needs 60 million microwatts.
The phase-change random access memory is based on electrical charges afflicted on a glassy material with randomly distributed atoms. While the traditional type of equipment can register data in the form of zeros and ones, the artificial neurons can create data storage where the atoms can have multiple values.
The non-volatile memory had been researched by companies such as Everspin, Samsung and Micron, all trying benefit from its high processing performance.
The artificial neurons are built from phase-change materials such as antimony, germanium, and telluride, which are already used in the re-writable Blue-ray discs. The difference is that they do not store digital information as they are analog.
The demonstration included in the study shows how the electrical pulses applied to the artificial neurons can provoke progressive crystallization and cause the neurons to activate. The function is called the integrate-and-fire property of neurons, which is the foundation of event-based computation.
Image Source: Public Domain Pictures