Nothing computes more efficiently than a brain, which is why scientists are working hard to create artificial neural networks that mimic the organ as closely as possible. Conventional approaches use artificial neurons that work together to learn different tasks and analyze data; however, these artificial neurons do not have the ability to actually “fire” like real neurons, releasing bursts of electricity that connect them to other neurons in the network. The third generation of this computing tech aims to capture this real-life process more accurately – but achieving such a feat is hard to do efficiently.
In a study published in IEEE Transactions on Electron Devices, a group of researchers in India propose a novel approach that allows these artificial neurons to fire in a much more efficient manner, allowing more “neurons” to be packed onto a computer chip. The advancement takes us one step closer to achieving more practical spiking neural networks (SNNs). This type of network could help us better understand the very organ that it’s inspired by, and thus better understand human disease, thought processes, and other mysteries of the brain.
Neurons in the brain “communicate” with each other by transmitting electrical spikes between one another. Networks of artificial spiking neurons can imitate this phenomenon by using leaky capacitors. Once a capacitor reaches a given threshold of electric charge, the voltage or current shoots out and affects the neighboring capacitor (analogous to another neuron).
A group of researchers at the Indian Institute of Technology Bombay designed the new SNN hardware. The design includes silicon-based electrical switches, called Metal-Oxide Semiconductor Field-Effect Transistors (MOSFETs), which are built on an insulating substrate.
To help the “neuron” fire and activate the other capacitors, the team added positively charged holes to the MOSFETs. Based on the nature of the MOSFETs, these holes allow for the quantum mechanical tunneling of electrons out of the capacitor. “The use of quantum mechanical tunneling provides incredible control, which is a huge advantage,” says Tanmay Chavan, a member of the research team that developed the SNN.
What’s more, this design can work in an off-current mode, which allows the capacitors to be 10,000 times smaller than if they required the current to be on. “In fact, the body capacitance is used to integrate the current within the transistor, leading us to utilize such tiny currents accurately without external loss,” explains Udayan Ganguly, another researcher at IIT Bombay involved in the study. “This… leads to extreme energy efficiency that we refer to as ‘computing at the current floor.’ Thus, fantastic energy and density is achieved.”
The researchers are interested in commercializing this design and are currently looking into forming new partnerships. “Given the fantastic performance at a unit neuron level, we plan to demonstrate networks of such neurons to understand how models of networks of neurons behave on silicon. This will enable us to understand the robustness and systems-level efficiency of the technology,” Chavan says.