For a Long time the human mind is a model for the development of intelligent machines. The greatest successes, the researchers have celebrated so far by Simulation: Artificial neural networks, the function form of operation of our brain in Software. This is the Royal road for Artificial intelligence, but he is still rocky, because of the complex calculations require crazy Hardware.

Therefore, the chip developer to simulate put on a different strategy for the near future: Why neurons and synapses using the Software, if you can replicate the human technology in the Form of neuromorphic Hardware? Why not an – admittedly still small – the brain is made of silicon build?

Tim Herman/Intel Corporation

This Board works with the neuromorphic Loihi Chips from Intel (on the right).

In comparison to typical artificial neural networks, which, for example, a graphics cluster is used as a hardware accelerator, is the human model a model of compact complexity. Around 100 billion neurons work in the brain, each of these neurons is via synapses with 1,000, in some cases, even up to a Million other neurons connected to it. If this network is running at full speed, can keep up, no Computer in the world. Although the neurons are expected to be several orders of magnitude slower than a CPU Transistor, the difference, however, makes the massive parallelism: Almost all the neurons of a brain can be simultaneously activated.

performance of the brain is still unattainable

overwhelmed even supercomputers, in spite of the superior computing potential, not a measure of your internal wiring with the billions-of-fire biological synapses. Because for the past 75 years, established Von Neumann architecture of a typical computer allows for calculations sequentially: the CPU and the memory are separated by a Bus, all commands are processed one after the other via this Bus – the is Of the-Neumann bottle neck.

What makes the whole thing easier is the plasticity of the brain. The networking is constantly changing, day-to-day to be replaced about ten percent of all neural Connections with new ones. Others will be weakened or strengthened. Each Neuron sends out when it is suggested that a short voltage pulse. The synapses transmit this Pulse to other neurons and depending on the connection strength of the actions. Neuromorphic Chips to form the wiring of neurons and synapses.

How the brain has the neural to set the semiconductor on parallel processing. There is no separation of memory and computing unit, these blocks are distributed over all the neurons. The nerve cells transmit information in the Form of electrical signals that are called action potentials or Spikes. The content of the Information, the spatial and temporal pattern of the signal is defined. Also, the error tolerance of the human thinking apparatus is imitated: a Neuron defect, take over neighboring cells.

The human brain is an energy-saving miracle

Today, all the CPU-manufacturer of neuro-build morph Designs. Under the code name Pohoiki Beach Intel has introduced a System that research institutions and industry partners. Pohoiki Beach has 64 so-called Loihi-Chips with eight million neurons. These Chips AI-tasks, creating up to 1,000 times faster than classic CPUs. Soon 100 million neurons will be possible.

the Siliziumhirne lag behind the biological model in all areas. Also in the case of power consumption: the brain consumes only 20 watts for its spectacular computing power. All the same, In a Deep-Learning Benchmark, the Intel system provides the power consumption of a graphics card cluster by more than a factor of 100.

Four million neurons and a billion synaptic Connections

In the framework of the EU-funded Human Brain Project (HBP) is created at the Kirchhoff Institute of the University of Heidelberg's brain scales. This neuromorphic Computer that mimics the cells, Connections, and communication of the brain by analog and digital Circuits. Brain scales consists of 20 silicon wafers with four million neurons and a billion synaptic Connections.

At the University of Manchester is working SpiNNaker, a complementary System of similar size. Together, the two silicon form brains, the Neuromorphic Computing Platform the HBP. SpiNNaker is considered to be the world's largest neuromorphic Supercomputer, he is a miracle of parallel processing: It can perform more than 200 millions of actions per second, and billions of small packets of information simultaneously, process.

Many play possibilities for the new Chips

your advantages of neuromorphic systems, especially in machine learning, where the time, profits and energy savings exorbitant. But also end users benefit from biologically inspired chip technology. Especially in the case of mobile devices, bionic IT is valuable – for example, when a networked Drive, or in the optimization of photos. Long smartphones, so – called NPUs-neural processing units, the cope to pattern recognition-based tasks autonomously without connection to the server cloud, work in Lace. These so-called Edge AI Chips can offline think.

FOCUS Online / Shuang Liu DLDaily

digitalization has changed Germany for the better, if we get it right. At the digital conference DLD, the as FOCUS Online, Hubert Burda Media, discuss several times in the year leading experts about how digitisation can succeed. For DLDaily FOCUS Online have the whole year with people, the concepts for the digital future: with innovative politicians, visionary thinkers, creative entrepreneurs. We explain the new technologies and show how they can be in their professional and private lives. All DLDaily article, you can find here.

the Internet of things takes this little thinking machines. The consulting firm Deloitte predicts that this year more than 750 million Edge AI Chips will be sold. The turnover of the neuromorphic computing units will grow significantly faster than the General processor market: "We expect that the sales of AI-will double Chips until 2024."

strong hail and Thunderstorm rain, Where on Tuesday dicey PCP will rain heavily, hail and Thunderstorms: Where is it on Tuesday tough that is

Date Of Update: 13 August 2020, 04:26