Many computer functions we refer to as artificial intelligence or machine learning aren’t really “learning” as we refer to human intelligence-building processes. Human brains tend to adapt when they learn something new. But computers tend to remain static and forget what they learned through artificial intelligence.
A study by a research team including scientists from Purdue University, the University of Illinois at Chicago, and Argonne National Laboratory shows how computer chips could rewire themselves on demand to take in new data like a human brain does.
“The brains of living beings can continuously learn throughout their lifespan. We have now created an artificial platform for machines to learn throughout their lifespan,” Shriram Ramanathan, a Purdue professor, said in a news release.
Human brains constantly form new connections between neurons to allow for learning and information retention. But the circuits on a computer chip don’t change
Artificial intelligence operations typically run through a device’s software instead of being embedded directly into its hardware. The researchers determined that ingraining the AI functionality into the hardware would help machines operate more efficiently and effectively.
They built a new kind of hardware that can be reprogrammed on demand via electrical pulses. This is essentially a system of reconfigurable artificial neurons and synapses that operates similarly to a human brain.
The small, rectangular chip is made from perovskite nickelate. Perovskite is a crystalline material that shows promise as an energy-efficient semiconductor. Its use is especially being explored in next-gen solar technologies.
Perovskite nickelate is sensitive to hydrogen, and adding electrical pulses at different voltages quickly rearranges hydrogen ions. Different concentrations of hydrogen caused the device to act like different brain functions: More hydrogen near the center replicates a neuron, and less hydrogen near the center replicates a synapse that connects neurons.
The team demonstrated that their device provides a dynamic structure that computes more efficiently than static structures. As new problems enter the system, the dynamic network decides which circuits are best to handle the problem.
The scientists believe the semiconductor industry could easily adopt this technology because the study used standard fabrication techniques and the system runs at room temperature.
“We demonstrated that this device is very robust,” said Michael Park, a Purdue Ph.D. student in materials engineering. “After programming the device over a million cycles, the reconfiguration of all functions is remarkably reproducible.”
The team’s chips are the building blocks of computers. They’re working to demonstrate the concepts from their study on large-scale test chips and to build a full computer inspired by the human brain.