Research Bits: March 29

Brain-like AI chip
Researchers from Purdue University, Santa Clara University, Portland State University, Pennsylvania State University, Argonne National Laboratory, University of Illinois Chicago, Brookhaven National Laboratory, and University of Georgia built a reprogrammable chip that could be used as the basis for brain-like AI hardware.

“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,” said Shriram Ramanathan, a professor in Purdue University’s School of Materials Engineering.

The device can be reprogrammed on demand through electrical pulses. “If we want to build a computer or a machine that is inspired by the brain, then correspondingly, we want to have the ability to continuously program, reprogram and change the chip,” Ramanathan said.

The device is made from perovskite nickelate, which is very sensitive to hydrogen. Applying electrical pulses at different voltages allows the device to shuffle a concentration of hydrogen ions in a matter of nanoseconds, creating states that the researchers found could be mapped out to corresponding functions in the brain. When the device has more hydrogen near its center, it can act as a neuron. With less hydrogen at that location, the device serves as a synapse, a connection between neurons, which is what the brain uses to store memory in complex neural circuits.

The internal physics of this device creates a dynamic structure for an artificial neural network that is able to more efficiently recognize electrocardiogram patterns and digits compared with static networks, the researchers said. This neural network uses reservoir computing, which explains how different parts of a brain communicate and transfer information.

As new problems are presented, a dynamic network can “pick and choose” which circuits are the best fit for addressing those problems.

“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.”

Since the team was able to build the device using standard semiconductor-compatible fabrication techniques and operate the device at room temperature, Ramanathan believes that this technique can be readily adopted by the semiconductor industry. The researchers are working to demonstrate large-scale test chips that would be used to build a brain-inspired computer.

Combining oscillators with memristors
Researchers from the University of Gothenburg and Tohoku University propose combining oscillator networks with memristors to create more brain-like hardware for AI.

“Finding new ways of performing calculations that resemble the brain’s energy-efficient processes has been a major goal of research for decades. Cognitive tasks, like image and voice recognition, require significant computer power, and mobile applications, in particular, like mobile phones, drones and satellites, require energy efficient solutions,” said Johan Åkerman, professor of applied spintronics at the University of Gothenburg.

Åkerman described oscillating circuits that can perform calculations as comparable to human nerve cells, while memristors that perform calculations and have integrated memory as comparable to memory cells. In the team’s device, memristors were used to control mutual spin Hall nano-oscillator synchronization.

“In its high resistance state, the memristor modulates the perpendicular magnetic anisotropy at the CoFeB/MgO interface by the applied electric field. In its low resistance state the memristor adds or subtracts current to the SHNO [spin Hall nano-oscillator] drive. Both electric field and current control affect the SHNO auto-oscillation mode and frequency, allowing us to reversibly turn on/off mutual synchronization in chains of four SHNOs. We also demonstrate that two individually controlled memristors can be used to tune a four-SHNO chain into differently synchronized states. Memristor gating is therefore an efficient approach to input, tune and store the state of SHNO arrays for non-conventional computing models,” the researchers wrote.

“This is an important breakthrough because we show that it is possible to combine a memory function with a calculating function in the same component. These components work more like the brain’s energy-efficient neural networks, allowing them to become important building blocks in future, more brain-like computers,” said Åkerman. “The more energy-efficiently that cognitive calculations can be performed, the more applications become possible.”

Insect-inspired memory
Researchers from CEA-Leti received a €3 million grant from the European Research Council (ERC) to build nanoscale memory devices inspired by insect nervous systems for such applications as consumer robotics, implantable medical diagnostic microchips, and wearable electronics.

“My project is to take inspiration from insects’ nervous systems to relax hardware requirements in terms of memory density and reliability, and to build the new nanosystems we need to enable learning from a very limited volume of noisy data,” said Elisa Vianello, CEA-Leti’s Edge AI Program Coordinator and recipient of the grant.

“Crickets make accurate decisions based on sluggish, imprecise, and unreliable neurons and synapses in order to escape their predators. Looking closely at their biology, we identified a diversity of memory-like functions at play in their sensory and nervous systems,” she said. “By combining these different functions, the cricket’s internal computing system achieves amazing performance and energy efficiency.

“Our ERC-funded project will use these novel nanoscale memory technologies to mimic the biological mechanisms observed in insects and create high-performance, energy-efficient, silicon-based nanosystems.

In particular, the researchers plan to build networks of physical nanoscale memory devices that translate insect biological principles into physical principles to enable learning from very limited volumes of noisy data, such as data measured in real time from different sensors in video cameras, radar sensors, ECG, EMG, bio-impedance streams, and potentially also brain signals through EEG sensors and neuro-probes. “Since the ideal memory does not exist today, the project aims at building a hybrid synapse that co-integrates different memory technologies,” Vianello said.