IBM Research has developed software for experimental crystals that mimic the workings of the human brain. California researchers call their development a "revolutionary software ecosystem."
The technology allows programming of chips whose architecture is borrowed from nature. Microcircuits operate at reduced power, they are able to process sensory information, act and learn the world in the same way as the human brain does.
New software is fundamentally different from the traditional software. IBM claims that the programming model created by it breaks down the traditional model with sequential execution of operations, which dates back to the 1940s to the architecture of John von Neumann. In one clock cycle, the IBM model performs not one, but many commands simultaneously. The new model is designed for distributed, closely interconnected, asynchronous, parallel and large-scale cognitive computing architectures.
Architectures and programs are closely interrelated, so the new architecture requires a new programming paradigm. We are working to create a Fortran programming language for cognitive computer chips. In addition to the current computers, this will provide fundamentally new technological opportunities in the light of programming and introducing developing cognitive systems, says Dharmendra Modha, principal researcher and senior manager of IBM Research.
IBM announced its new chips based on the human brain back in 2011. However, the use of these chips required the creation of special programming tools.
According to IBM, brain-like computers can have extremely high performance and low power consumption. The project is funded by the Defense Advanced Research Project (DARPA). The agency has already invested in development more than 53 million dollars.
IBM researchers have taken important steps to build computers that function on the principle of the brain. They have developed a simulator of multi-threaded, massively parallel and well-scalable software using a massive network of neuro-synaptic cores, also known as Synapse processors.
IBM set the long-term goal of creating a computing system the size of a two-liter jar with ten billion neurons and hundreds of trillion virtual synapses inside that would consume no more than 1 kW of energy.
The programming model follows the principle of the neural model. It is composed of reusable building blocks, the so-called "corlet" ("Corelet"). Each corelet is a network of neurons. The internal operations of the neuron are hidden from the eyes of the programmer, who sees only the external inputs and outputs. Thus, the programmer can focus on what the neuron does, and not on how it does it. Corlets are hierarchical structures. They can be used to create new, more complex corlets.
Already more than 150 corlets became part of the library for programmers. IBM presents a new programming model at the International Joint Conference on Neural Networks (ICANN), which is being held these days in the US city of Dallas, Texas.
Earlier, scientists from Dalhous University announced the creation of the first practical high-level programming language for quantum computers.
The article is based on materials .
Comments
Post a Comment