Andy Thomas constructed a memristor that is capable of learning. Andy and Bielefeld University researchers are now using his memristors as key components in a blueprint for an artificial brain. Memristors are made of fine nanolayers and can be used to connect electric circuits. For several years now, the memristor has been considered to be the electronic equivalent of the synapse. Synapses are, so to speak, the bridges across which nerve cells (neurons) contact each other. Andy is first to summarize which principles taken from nature need to be transferred to technological systems if such a neuromorphic (nerve like) computer is to function. Such principles are that memristors, just like synapses, have to ‘note’ earlier impulses, and that neurons react to an impulse only when it passes a certain threshold.
Both a memristor and a bit work with electrical impulses. However, a bit does not allow any fine adjustment – it can only work with ‘on’ and ‘off’. In contrast, a memristor can raise or lower its resistance continuously. ‘This is how memristors deliver a basis for the gradual learning and forgetting of an artificial brain,’ explains Thomas.
Hewlett Packard should being selling chips with billions to trillions of memristors that compete with flash memory in density and scale in 2014. Memristors seem able to create simple, scalable and efficient devices for mimicking trillions of neurons and synapses.
Schematic representation of two interconnected neurons. The contact areas where the information is transmitted are called synapses. A signal from the presynaptic cell is transmitted through the synapses to the postsynaptic cell.
ABSTRACT – The synapse is a crucial element in biological neural networks, but a simple electronic equivalent has been absent. This complicates the development of hardware that imitates biological architectures in the nervous system. Now, the recent progress in the experimental realization of memristive devices has renewed interest in artificial neural networks. The resistance of a memristive system depends on its past states and exactly this functionality can be used to mimic the synaptic connections in a (human) brain. After a short introduction to memristors, we present and explain the relevant mechanisms in a biological neural network, such as long-term potentiation and spike time-dependent plasticity, and determine the minimal requirements for an artificial neural network. We review the implementations of these processes using basic electric circuits and more complex mechanisms that either imitate biological systems or could act as a model system for them.
Implementations using memristive systems
The question may be asked as to why we want to construct neuromorphic systems that emulate the presented biological mechanisms. Carver Mead pioneered a considerable amount of the research on this topic and explained the reasoning in the following way : For many problems, particularly those in which the input data are ill-conditioned and the computation can be specified in a relative manner, biological solutions are many orders of magnitude more effective than those we have been able to implement using digital methods. […] Large-scale adaptive analog systems are more robust to component degradation and failure than are more conventional systems, and they use far less power. The need for less power is particularly obvious if we compare the performance of the brain of even an invertebrate with a computer CPU and contrast the power consumption. However, there are some tasks that are difficult for a human and easy for a computer, such as multiplying two long numbers, and other problems that a human can easily solve but computers fail to solve.
Human Brain Project Using other Technology was funded for 1 billion Euro over ten years
There is a 108 page report on the Human Brain Project. The Human brain project would likely adopt memristors if that technology proves out.
One of the main applications of the Human Brain project if neurorobotics.
One of the most important potential applications for neuromorphic computing would be in neurorobots – robots whose controllers incorporate a model brain, implemented on a neuromorphic computing system. Neuromorphic controllers would benefit from many of the intrinsic advantages of neuromorphic technology, including the ability to acquire new capabilities without explicit programming, to process high-dimensional input streams, and to control robot bodies with many degrees of freedom. Physical implementations of such controllers could run up to ten thousand times faster than biological real time, allowing very rapid training. The Neurorobotics Platform would allow researchers and technology developers to transfer the resulting brain models to many-core devices, suitable for integration in physical robots, providing the know-how, hardware, and software they would need to explore these possibilities.
Some possible applications would be developed in pilot projects, others by groups from outside the HBP, others by industry partnerships. The compact size, high resilience and low power consumption of neuromorphic controllers would facilitate the development of a broad range of applications with a potentially very large social, industrial and economic impact. The HBP would help European industry to drive the technology instead of becoming a user and importer of technologies developed elsewhere.