Researchers from the University of Electronic Science and Technology of China and Nanyang Technological University in Singapore have created a transistor they call a “neuron transistor”. The neuron transistor operates much like a living biological neuron and could play an integral role the future of neuromorphic computing
The paper demonstrates that their neuron transistor is capable of weighted summation and threshold functions, just like a biological neuron. Weighted summation and threshold functions are the biological neurons’ ability to sum the input values of their connected neurons and whether or not to fire based on their threshold value.
Biological neurons may be connected by to up to 10,000 other neurons and transmit signals to each other via as many as 1,000 trillion synaptic connections. The estimated 86 billion neurons in the human brain regulate all of our cognition by continually carrying out weighted summation and threshold functions.
In the paper, the researchers show how a single neuron transistor can act like a single neuron, with the ability to preform weighted summation and threshold functions. The neuron transistor is constructed out of two-dimensional flake of molybdenum disulfide, rather than the typical silicon of conventional transistors. Molybdenum disulfidenew is a new class of of semiconductor called transition metal dichalcogenides. Transition metal dichalcogenide are atomically thin semiconductors that are part of the growing class of 2D materials.
The researchers demonstrated that their neuron transistor could match the function of a biological neuron by showing that it could be controlled by both one or two gates at the same time. When the neuron transistor is controlled by two gates it carries out a summation function. The researchers showed that this function could execute counting tasks similar to sliding the beads on a two-bead abacus.
A major advantage of the team’s new neuron transistor is its vastly improved operating speed. The previous neuron transistors that have been built have usually achieved an operating frequency of 0.05 Hz, far lower than 5 Hz firing frequency of a biological neuron. The new neuron transistor is capable of operating at a much wider range of frequencies, from 0.01 to 15 Hz, which the team believes will give it significant advantages for creating neuromorphic chips.
The next step for the researchers is to build neuron transistors that contain more control gates which will allow them to even more closely match a biological neuron with its legion of inputs. The researchers are also looking to incorporate their neuron transistors with memristors to simulate how a brain functions, they consider memristors to be the best hardware for simulating how synapses function in the brain.
The paper by, S. G. Hu et al., was published in the journal Nanotechnology.