Researchers from the Massachusetts Institute of Technology have developed programmable transistors that operate 10,000 times faster than brain synapses. The technology used for analog machine learning not only ensures high processing speed but also good energy efficiency.
The working mechanism of the device is the electrochemical introduction of the smallest ion, proton, into the insulating oxide for the modularization of its electronic conductivity, explained by the authors of the work. Scientists used a powerful electric field to accelerate protons and convert ion transistors to nanosecond mode.
Researchers note that the secret of new devices in the use of inorganic phosphophosilicate glass is that it allows protons to move super fast because it contains many nanometre-sized pores whose surfaces provide ways to diffuse elemental particles; it can also withstand very strong impulse electrical fields.
The potential for action in biological cells increases and drops with a time scale in milliseconds, because the potential difference of about 0.1 volts is limited to the stability of water. In our work, we use up to 10 volts through a special hard-glass film of nano-gauge thickness that conducts protons without damaging it. And the stronger the field, the faster the ion devices work.
In the human brain, learning takes place by strengthening and weakening the links between synapses. Deep neural networks use a similar strategy when the weights of nodes are programmed using learning algorithms. When using processors, increasing and reducing the electrical conductivity of proton resistors provides analog machine learning.
The conductivity is controlled by proton movement. To increase conductivity, more protons are pushed into the canal in the resistor, and the protons are released to reduce conductivity. This is achieved by means of electrolyte that the protons do but block the electrons. Increased proton speed significantly accelerates the machine learning process.