Memrestive components show an electrical behavior as synapses (blue dasgestellt), the neurons in the brain connected with each other.
Image: University of Massachusetts, Yao lab
The computer chips of the future are to work according to the nature of the human brain, only faster. In many small steps, researchers come closer to this goal. To do this, use the special components, so-called Memristoren.
Ob chess, Poker or the Board game of Go, losing, in the meantime, people in almost all strategy games, against self-learning computer systems. Nevertheless, the human thinking apparatus is brains continue to the Computer in a number of things to consider. While KI is specialized in the rule, you can do the brain is an incredibly wide range of tasks. And it is extremely efficient. A normal brain only uses as much energy as a 20 Watt light bulb. Around the globe, researchers are looking for Ways, which form the Hardware of the “biological computer” is technically track. You are working on so-called neuromorphic processors that work not only more efficient than traditional computer chips, but also similar to the brain – independently, and can forget to learn. Previous AI concepts to mimic the self-learning mechanisms in neural networks, not with the Hardware, but only by means of sophisticated Software.
Scientists from the University of Massachusetts in Amherst have developed a three-dimensional neuromorphic processor, the complexity of the natural model comes a step closer. How Qiangfei Xia and his colleagues in the journal “Nature Electronics” report, uses the Chip instead of the Transistors and other classic components, a so-called Memristoren. These switching elements have special electrical properties and have been in force since 2008, when researchers from Hewlett Packard (HP) presented for the first time, as a hot candidate for the construction of neuromorphic circuits.
3D Chip processes signals according to the model of the brain
The name Memristor is a portmanteau of “memory” (memory), and a “resistor” (resistance) – goes back to the American engineer, Leon Chua, who designed in 1971 the theoretical concept for this component. Unlike the conventional switching elements, the electrical resistance and thus the conductivity in dependence of the applied voltage changes in the case of a Memristor. After switching Off the voltage, the Memristor retains its last resistance value. The component can therefore process information at the same time and save. This behavior is based on structural changes of the Memristors – for example, by the voltage pulse between the two electrodes charged atoms – ing and fro-ing and accumulate, which lowers the resistance, or lifting.
In their electrical behavior Memristoren are in a way similar to biological synapses. Because learning and memory skills of the brain are essential due to the fact that the Connections between nerve cells strengthen, so to speak. Synapses pass signals to different degrees, if you are quickly excited in a row. In addition, the activity provides a long-term anatomical Change, which affects the strength of the Transmission even for life. Brain researchers speak of the synaptic plasticity. While a traditional semiconductor element is a rigid construct, are Memristoren – similar to how synapses – capable of change. And, if you will, even capable of “learning”. The Memristor, the Qiangfei Xia and his colleagues use, consists of a thin layer of hafnium oxide which is sandwiched between a titanium, and a tantalum electrode. The component changes quickly – within nanoseconds – its electrical properties when it creates a corresponding voltage pulse.