Not that anyone doubted it, but IBM was serious when he revealed the plan to invest $ 3 billion in technologies to succeed Moore’s Law . One has recently been made: the TrueNorth a processor inspired by the brain as such, is able to simulate a network 1 million neurons.
The novelty does not (yet) have the same processing power of today’s chips (based on x86 and ARM, for example) and, of course, is far from what the human brain is able to do – in fact, there is not as compare, inside our heads there are tens of billions of neurons.
But at this stage, the chip consumes much less energy – around 70 milliwatts – and proves to be quite efficient in processing sensory information, such as images, sounds and movements.
At first glance, the TrueNorth refers to today’s processors. The current prototype, whose production was handled by Samsung, has 28 nanometer manufacturing process and consists of 5.4 billion silicon transistors.
The differences are evident in the form of interconnection. The transistors are connected in a similar way to neurons, ie, forming synapses. The result of this proposal is a network of 4,096 “neuro-synaptic nuclei” which, according to IBM, corresponding to 256 million synapses.
Very cool! But what need of this? Well, the current computation is based on the von Neumann architecture model that describes processing of data stored in memory from linear operations. The architecture inspired by the brain runs this concept.
At TrueNorth, each “neuro-synaptic core” works as if it had its own set of memory and processing, since synapses and neurons assume these functions. Thus, several groups of nuclei can be targeted to different tasks. Similarly, a large number of them can work together in processing a demanding activity.
The performance gain of this model tends to be huge. According to IBM researchers, the main problem of the current architecture is that access Serial forces the processor to send and seek information in memory constantly, creating performance bottlenecks and requiring more energy.
How are small and consume little electricity, chips “brain type” can make wearable devices finally deslancharem. The sensors could act as small processors, opening up a huge range of possibilities for these products.
The TrueNorth own, despite not pass prototype, shows already skilled at recognizing: to analyze a video, the chip was able to identify people, cars and even bicycles crossing a road.
The technology can also be directed to complex applications. IBM is already studying ways to interconnect thousands of chips TrueNorth to build a supercomputer.
If you assumed that it will take some time for this type of processor has space hit. Besides lapidary technology, IBM has the challenge of making the proposal be accepted: TrueNorth as chips represent a paradigm shift, however, may encounter resistance because of its totally new concepts of programming.
Even experienced developers consider the complicated idea, according to Dharmendra Modha, IBM’s chief scientist. But their lives should not be too difficult: the project team has been working to make the technology available with a library of pre-made rather generous codes.