- Researchers demonstrated technology for controlling electronic device memory.
- Neuronal transistor mimics biological neurons' electrical switching behavior.
- Repeated electrical stimulation can drive material into a predictable jammed state.
- This discovery is crucial for the development of neuromorphic computing.
Researchers at UC San Diego have demonstrated a method for controlling how an experimental electronic device forms and erases memory, using electrical current to alter the behavior of a material called titanium dioxide. The device, called a neuronal transistor, mimics certain electrical switching behaviors of biological neurons. The researchers found they could influence its switching to become more predictable or return to a less predictable state. This discovery could help scientists design brain-inspired computing hardware that stores information through changes in physical structure. Such systems aim to perform certain computing and memory functions more effectively than conventional methods, which require moving data between separate processors and memory units.
A UC San Diego team led by physics professor Oleg Shpyrko used coherent X-ray diffraction to observe nanoscale changes inside the material as electrical current passed through the device. This allowed them to study how its internal structure influences switching behavior. Titanium dioxide can transition between electrically insulating and conducting states. This change is accompanied by structural changes in the material, making it useful for studying electronic devices that mimic certain features of neurons. In the experiments, the researchers examined how repeated electrical stimulation affects the arrangement and movement of tiny regions inside the material.
Research Methods and Material Properties
These regions, called domains, can interact when the device switches between states. The team identified a process called jamming, in which the movement of these domains becomes increasingly constrained. Their interactions begin to limit the material's changes, rather than acting independently. Repeated electrical stimulation beyond a certain threshold pushes the material into this jammed state. As a result, the device's subsequent switching becomes more predictable, showing that its early electrical activity influenced its future behavior. The researchers also found that stimulation with lower current can reverse this effect, restoring a less predictable switching state, providing a way to control the device's memory-like behavior through electrical input.
These experiments tracked structural changes over thousands of seconds, enabling the researchers to link the device's electrical history to changes inside the material. The findings provide a physical explanation for how neuronal transistors retain information about prior electrical activity without separate conventional memory components. In this context, learning does not mean the device understands information or solves problems independently, but refers to measurable changes in switching behavior caused by prior stimulation. Forgetting describes the reversal of this change.
Potential for Neuromorphic Computing
This distinction is crucial for neuromorphic computing, which aims to build electronic systems inspired by how neurons and synapses process and retain information. A device that can adjust its response based on prior electrical signals could eventually help create circuits that combine memory and computing functions. However, the researchers demonstrated this mechanism in an experimental material system, not a complete brain-inspired processor. Further work is needed to determine how to reliably control behavior across multiple devices, the speed at which memory states change, and whether this approach can yield measurable energy savings in practical circuits.
The study also provides a method for studying how nanoscale structural changes affect electronic switching in materials with coupled structural and electrical transitions. The findings have been published in Nature Communications.
Potential Applications and Challenges of Controlling Memory
The results of this research show that by controlling the memory behavior of electronic devices, future computing hardware may more closely resemble the operation of biological neural systems. This would not only improve computing efficiency but also reduce data movement between processors and memory units, thereby saving energy. However, the experiments remain at an early stage, and researchers need to further investigate how to reliably control behavior across multiple devices and ensure these technologies can achieve their intended effects in practical applications.