Scientists from Jülich have developed new memristors. Memristors are tiny devices that mix memory and resistors. Resistors control electric flow, and memory stores information. These new memristors use very little power and act like brain cells. They are stronger, work with more voltages, and switch between digital and analog modes. These traits could stop "catastrophic forgetting" in artificial neural networks. That’s when networks suddenly forget old lessons while learning new ones.
The brain avoids this forgetting problem. It adjusts how much connections change, a trick called metaplasticity. Metaplasticity helps the brain learn new things without losing old knowledge. The new memristors copy this idea. They change their resistance based on voltage. Unlike old memristors, they keep their changes even without power. This happens because tiny particles shift inside them.
Ideal candidates for neuro-inspired devices
Memristors could improve computer chips. Old memristors often fail, wear out fast, or break from heat and stress. In a press release issued by Forschungszentrum Jülich, research leader Ilia Valov says basic research fixes these issues. His team found a new way memristors work. It uses metal oxides instead of pure metal. This makes a stable filament, a tiny bridge for electricity. The filament adjusts but never fully breaks. Scientists call this the filament conductivity modification mechanism, or FCM.
The scientists have described the new memristors in a paper published in Nature Communications.
FCM memristors handle heat better, use less voltage, and last longer. Fewer break during making, too. They also mix digital and analog signals. This mix helps neural networks. Networks with these memristors learn new tasks without forgetting old ones. Valov’s team tested them in simulations with images. The system recognized patterns well.
In the future, the team wants better materials for memristors. They aim to boost electronics for "computation-in-memory," where chips think and remember at once. This could make computers faster and smarter, like human brains.