Research and development of a computer, let it like human decision making, with human intelligence, it is the business of IBM is considering, now it's a big step toward goals.

IBM has developed a chip named TrueNorth, it can simulate human brain function. IBM is testing chip, to prove it, compared with the existing computer how fast and how much energy saving.

Positive results were impressive. IBM said the TrueNorth can call deep learning technology, through correlation analysis, and judgment possibility decision-making, like the human brain. Compared with other computer with the same purpose, new computer consumes less energy.

IBM said in the blog, TrueNorth chips have learning and computing power, it creates a kind of may, we can take all intelligence embedded computing devices, including the Internet, smart phones, robot, car, cloud computing, super computing.

Earlier this year, IBM has been display chip, the chip is loaded into a named NS16e computer, it is an imitation of human brain development. Computer can recognize pictures, voice, and mode, these features was conducted by neural network processing unit.

The human brain has 100 billion neurons, they can by trillions of neural nodes (synapses) communicate with each other. Among them, the visual cortex can identify the visual information, the other part is responsible for motor function.

And brain, NS16e has "number of neurons," number is less, including 16 TrueNorth chips in the system. Every TrueNorth chip with 1 million neurons, 256 million synapses, they contact each other through the circuit. NS16e installed the redesign of memory, computation and communication subsystem, data processing, energy consumption is much lower.

IBM has said, TrueNorth processor can according to the speed of 1200-2600 frames per second to image data classification, the consumption of energy is only 25-275 milliwatts. Processor can also identify model images, these images are generated by 50-100 camera, 24 frames per second. If install the chip to the handset, it can also have this ability, a few days without charge.

Compared to today's server, new chips on the energy consumption is much lower, today's dependence on traditional server chips, such as the GPU, CPU and FPGA, companies use these components to identify the image and voice. Facebook, Google, Microsoft, baidu with deep learning technology to study, through analyzing related to pictures, voice recognition, answer to learn. These deep learning system is usually driven by the GPU, more than 150 watts.

TrueNorth has algorithm and deep learning models, including identification model, can be associated with existing data in the past. IBM's deep learning model is developed for different corresponding algorithms, but the chips can be applied to existing systems, such as MatConvNet. In essence, a developer can use MatConvNet development learning model, TrueNorth responsible for background processing. There is no need for direct contact TrueNorth developers.

IBM's strategy and early game development is similar to that at the beginning of the most programmers don't contact the GPU, because most people don't know how to use the function of the chip. Recently, replacing OpenGLAPI Vulkan, showed the GPU functions directly to the programmers, so that they can make better use of the function of the chip.

Deep learning potential has been reflected in the driverless cars, powerful computer boot auto safety, correct identification signal, lanes and other objects. Like cars, the server chips, TrueNorth will be conducted on each neuron low-level processing, then the results summary, so as to identify the object in the image, or identify a voice. Nvidia and Intel chips also USES the same technology, but their energy consumption of the chip is higher than TrueNorth a lot.

The TrueNorth chip is still in early stage of development. IBM plans to develop a computer, which is installed TrueNorth chips, specifications with the human brain, but IBM face some challenges, mainly because of this computer is too large, difficult to develop algorithms and applications.

Starting in 2004, IBM is in research and development of the brain type chip, in 2009, computers have can simulate a cat's brain. In 2011, IBM has developed a prototype chip, it has 256 digital neurons, has the ability of pattern recognition. If you want to develop a real computer, let it simulate the human brain, have to wait a long time.

IBM is also in the development of quantum computer, it can replace the existing PC and server, existing equipment adopts the design of the already very old, used for decades. Some enterprises and institutions are also developing chips can simulate the human brain, such as hewlett-packard company (HewlettPackardEnterprise), Stanford university, university of Heidelberg, Germany, the university of Manchester.