Scientists from Texas A&M University, Hewlett Packard Labs and Stanford University they described a new nanodevice that acts almost identically as a brain cell. Moreover, they showed that these synthetic brain cells can be joined together to form complex networks that can then solve problems in a brain-like way.
“This is the first study in which we have been able to emulate a single-nano-scale neuron that would otherwise need hundreds of transistors,” said Dr. R. Stanley Williams, lead author and professor in the Department of Electrical and Computer Engineering. “We have also been able to successfully use our artificial neural networks to solve toy versions of a real-world computing-intensive problem, even for the most sophisticated digital technologies.”
In particular, the researchers demonstrated the concept that their brain-inspired system can identify possible mutations in a virus, which is extremely relevant to ensuring the effectiveness of vaccines and drugs for strains with genetic diversity.
What does this new device help with
In recent decades, digital technologies have become smaller and faster, largely due to advances in transistor technology. However, these critical components of the circuit are rapidly approaching their limit of how small they can be built, initiating a global effort to find a new type of technology that will complement, if not replace, transistors.
In addition to this “reduction” problem, transistor-based digital technologies have other well-known challenges. For example, they struggle to find optimal solutions when presented with large data sets.
“Let’s take a familiar example, to find the shortest route from the office to your home. If you have to make a single stop, it is a fairly easy problem to solve. But if for some reason you have to make 15 stops between them, you have 43 billion routes to choose from, ”said Dr. Suhas Kumar, lead author of the study and researcher at Hewlett Packard Labs. “This is now an optimization issue, and today’s computers are pretty inept at solving it.”
Kumar added that another difficult task for digital cars is recognizing patterns, such as identifying a face, regardless of angle of view, or recognizing a familiar voice buried in a noise of sounds.
But the tasks that can send digital machines into a computational thesis are those in which the brain excels. In fact, brains are not only fast at recognizing and optimizing, but they also consume less energy than digital systems. Therefore, mimicking how the brain solves these types of tasks, Williams said brain-inspired or neuromorphic systems could overcome some of the computational hurdles facing current digital technologies.
How they mimicked the fundamental element of a neuron
Thus, to build the fundamental element of the brain or a neuron, the researchers assembled a synthetic nanoscale device, consisting of layers of different inorganic materials, each with a unique function. However, they said that the real magic happens in the thin layer of the niobium dioxide compound.
Therefore, when a small voltage is applied in this region, its temperature begins to rise. But when the temperature reaches a critical value, niobium dioxide undergoes a rapid change, transforming from an insulator to a conductor. But as it begins to conduct electricity, the temperature drops and niobium dioxide returns to being an insulator.
These back and forth transitions allow synthetic devices to generate an electrical current pulse that closely resembles the profile of electrical peaks or action potentials produced by biological neurons. Moreover, by changing the voltage on their synthetic neurons, the researchers reproduced a wide range of neural behaviors observed in the brain, such as sustained, explosive, and chaotic triggering of electrical peaks.
“Capturing the dynamic behavior of neurons is a key goal for brain-inspired computers,” Kumar said. “In total, we were able to recreate about 15 types of neural trigger profiles, all using a single electrical component and at much lower energies compared to transistor-based circuits.”
The researchers found that within a few microseconds, their network of artificial neurons stabilized to a state that indicated the genome of a mutant strain.
To assess whether their synthetic neurons can solve real-world problems, the researchers first connected 24 such nanoscale devices to a network inspired by connections between the cerebral cortex and the thalamus, a well-known neural pathway involved in pattern recognition. They then used this system to solve a toy version of the viral quasi-species reconstruction problem, where mutant variations of a virus are identified without a reference genome.
Using the data entered, the researchers introduced the network into fragments of short genes. Then, by programming the power of the connections between the artificial neurons in the network, they established basic rules about the association of these genetic fragments. The task of the network puzzle was to list mutations in the virus genome based on these short genetic segments.
Williams and Kumar noted that this result is a proof of principle that their neuromorphic systems can quickly perform tasks in an energy efficient way. The researchers said the next steps in their research will be to expand the repertoire of problems that their brain-like networks can solve by incorporating other shooting patterns and distinctive properties of the human brain, such as learning and memory. It also intends to address the hardware challenges of implementing their technology on a commercial scale.
“Calculating national debts or solving large-scale simulations is not the kind of task that the human brain is good at and that’s why we have digital computers. Alternatively, we can use our knowledge of neural connections to solve problems that the brain exceptionally has, ”Williams said. “We have shown that, depending on the type of problem, there are different and more efficient ways to do calculations, other than conventional methods that use digital computers with transistors.”