Artificial neurons replicate biological function for improved computer chips

Researchers at the USC Viterbi School of Engineering and the School of Advanced Computing have developed artificial neurons that replicate the complex electrochemical behavior of biological brain cells.

The innovation, documented in Natural electronicsrepresents a step forward in neuromorphic computing technology. The innovation will reduce the size of the chip by several orders of magnitude, reduce its power consumption by several orders of magnitude and could advance artificial general intelligence.

Unlike conventional digital processors or existing neuromorphic chips based on silicon technology that simply simulate neuronal activity, these artificial neurons physically embody or mimic the analog dynamics of their biological counterparts. Just as neurochemicals trigger brain activity, chemicals can be used to initiate calculations in neuromorphic (brain-inspired) hardware devices. By being a physical replication of the biological process, they differ from earlier iterations of artificial neurons that were just mathematical equations.

The work, led by Joshua Yang, a professor of computer and electrical engineering at USC who also led the work in a seminal paper on artificial synapses more than a decade ago, introduces a new type of artificial neuron based on what is known as the diffusive memristor. THE Natural electronics The paper explores how these artificial neurons can enable a new class of chips that complement and augment current silicon-based technologies, which power almost all modern electronic devices and rely on the movement of electrons for calculations.

Instead, the diffusive device introduced by Yang and his colleagues to build the neurons would rely on the movement of atoms. Such neurons may enable the creation of newer chips that would function more similarly to how our brains work, be more energy efficient, and could lend themselves to the advent of what is known as artificial general intelligence (AGI).

How the device works

In the biological process, the brain uses electrical and chemical signals to drive the body’s action. Neurons or nerve cells start as electrical signals which, when they reach the gap or space at the end of the neuron (the synapse), are converted into chemical signals in order to transmit and process information. Once the information passes to the next neuron, some of these signals are converted again into electrical signals throughout the body of the neuron.

This is the physical process that Yang and his colleagues were able to mimic with great fidelity in several critical aspects. The big advantage: Their memristor-based diffusive artificial neuron requires only the space of a single transistor, rather than the dozens or hundreds used in conventional designs.

In particular, in the biological model, ions or charged particles help generate the electrical signals necessary for action within the neuron. In the human brain, such processes rely on chemicals (e.g. ions) like potassium, sodium or calcium to force this action.

In the current paper, Yang, director of the Center of Excellence in Neuromorphic Computing at USC, uses silver ions in the oxide to generate the electrical pulse and mimic the processes to perform calculations for activities such as movement, learning and planning.

“Even though they are not exactly the same ions in our artificial synapses and neurons, the physics governing the movement of ions and their dynamics are very similar,” he says. “Silver is easy to diffuse and gives us the dynamics we need to mimic the biosystem so we can achieve the function of neurons, with a very simple structure.”

The new device capable of creating a brain-like chip is called a diffusive memristor because of the movement of ions and dynamic diffusion that occurs with the use of silver.

He adds that the team chose to use ion dynamics to build artificial intelligent systems “because that’s what’s happening in the human brain, for good reason and since the human brain is the ‘evolutionary winner’ – the most efficient intelligent engine.”

“It’s more efficient,” he says, explaining: “It’s not that our chips or our computers aren’t powerful enough for everything they do. It’s that they’re not efficient enough. They consume too much power.”

This is particularly relevant, given the level of energy required to run large software models with a huge amount of data, such as machine learning for artificial intelligence.

Yang goes on to explain that unlike the brain, “our existing computer systems were never designed to process massive amounts of data or to learn on their own from just a few examples. One way to increase both the energy and efficiency of learning is to build artificial systems that operate according to the principles observed in the brain.”

If you’re looking for pure speed, the electrons that run modern computing would be best for fast operations. But, he explains, “ions are a better way than electrons to embody the principles of the brain. Because electrons are light and volatile, computing with them allows for software learning rather than hardware learning, which is fundamentally different from how the brain works.”

In contrast, he says, “the brain learns by moving ions across membranes, thereby achieving energy-efficient and adaptive learning directly in hardware, or more precisely in what people might call ‘wet software.’

For example, a young child can learn to recognize handwritten numbers after seeing only a few examples of each, whereas a computer typically needs several thousand to accomplish the same task. Yet the human brain accomplishes this remarkable learning while consuming only about 20 watts of energy, compared to the megawatts required by today’s supercomputers.

This new method is one step closer to imitating natural intelligence.

Yang noted that the silver used in the experiment is not easily compatible with conventional semiconductor manufacturing and that other ionic species will need to be studied for similar functionality.

The efficiency of these diffusive memristors includes not only energy, but also size. Normally, a smartphone has about 10 chips but billions of transistors or switches that control the on/off (of the 0s and 1s) that underpin the calculation.

“Instead (with this innovation), we are simply using a footprint of a transistor for each neuron. We are designing the building blocks that will ultimately allow us to reduce chip sizes by orders of magnitude and reduce power consumption by orders of magnitude, so that it can be sustainable to achieve AI in the future, with a similar level of intelligence without burning energy that we cannot maintain,” says Yang.

“Now that we have demonstrated capable, compact building blocks, artificial synapses and neurons, the next step is to integrate large numbers of them and test how well we can replicate the efficiency and capabilities of the brain.”

“Even more exciting,” Yang concludes, “is the prospect that such brain-fidelity systems could help us discover new insights into how the brain itself works.”