Researchers at TU Dresden and HZDR Develop World’s First Neurotransistor
Using semiconductor materials, a joint research team claims to have successfully imitated the functioning of neurons—combining data storage and processing on a single component.
When design engineers want to enhance performance, they typically look for ways to reduce component size, especially the size of individual transistors. But this shrinking can’t go on indefinitely and it is new R&D approaches, like combining data processing and storage on a single chip, that innovate both performance and size.
One of the newest types of transistor, developed by scientists at TU Dresden and Helmholtz-Zentrum Dresden-Rossendorf (HZDR), is a so-called neurotransistor—the first of its kind. When developing the transistor, the joint research team looked to the workings of the human brain, particularly in how neurons function, for inspiration.
The research is published in Nature Electronics.
The Architecture of the Neurotransistor
To create the neurotransistor, the researchers mimicked the properties of neurons by looking to biosensors and a modified field-effect transistor (FET), according to HZDR physicist Larysa Barbaran.
Researchers manufactured the neurotransistor using a standard CMOS process on an 8-inch silicon-on-insulator wafer. To generate memory and promote plasticity, a silicate film coats the transistor, which includes mobile ions acting as a type of gate.
Rendering of a neurotransistor. Image used courtesy of E. Baek and TU Dresden
A sigmoidal transformation processes pulsed input signals non-linearly into the output current, which is said to mirror how neuronal membranes function. The input signal history, stored as ionice states in the silicate film, then dictates output response. The result is what we might interpret as the neurotransistor's "learning ability."
The advantage of this architecture is that information can be stored and processed simultaneously in a single component. In contrast, conventional transistors separate storage and processing, slowing down processing time and limiting performance.
A Chip Capable of Learning
The researchers of this neurotransistor project readily admit that modeling computers based on the human brain is far from a novel idea. One of the earliest attempts, according to the press release, involved hooking up nerve cells to electronics in Petri dishes.
“But a wet computer chip that has to be fed all the time is of no use to anybody,” says Professor Gianaurelio Cuniberti of TU Dresden.
Now, Cuniberti and colleagues have been able to re-apply these early principles of the brain-inspired electronics by using a viscous substance called "solgel" to a conventional silicon wafer with circuits. The solgel polymer then hardens and becomes a porous ceramic, allowing ions to move between the holes. These ions are much heavier than electronics and thus are slower to return to their position after excitation.
This delay, known as "hysteresis," is what causes the storage effect and is a core factor in how the neurotransistor functions. “The more an individual transistor is excited, the sooner it will open and let the current flow. This strengthens the connection. The system is learning,” explains Cuniberti.
A New Learning Tool for Robotics?
Cuniberti notes that his team isn’t focusing on conventional issues in this field of research, like the plasticity of artificial synapses. Instead, he and his team are looking toward the applications of this device—one being more intelligent neuromorphic machines that would be suited for advanced applications in robotics. Cuniberti specifically envisions robots that can learn to walk or grasp objects, all without having to rely on complicated software.
A robot automated to grasp.
Researchers also explain that because of the plasticity of neuromorphic computers, a robot with this chip would be able to adapt to new situations, even those they were not originally programmed for.