No computer chips today can compete with a functional human brain in the power of information gathering, intelligence and energy efficiency, yet with its moderate size. That is because a human brain functions drastically different from today’s computers. But what if future computers adopt the brain neuron structure and start processing information by logics closer and closer to the ways a human brain works? If with huge amount of processing power and memory storage available, would computers one day indeed surpass the brain power? The answer has become increasingly difficult to come by.
Scientists and engineers together have been trying to build a brain-like computer for decades. With the concepts of Artificial Neuron Network (See our earlier blog on Artificial Neuron Networks ) and advancement in computer engineering, a modern computer chip designed without a conventional powerful central processing unit (CPU), but with millions of parallel “neurons” and connecting “synapses” packed into a single unit to simulate one brain function, e.g., one cognitive ability, may well come closer to capture that specific brain function after repeated learning, storing and processing information. When a large number of these special units combined together in a coordinated fashion, a “machine-made brain” may just be born. Thus “neuromorphic computing” evolves from here. IBM, Qualcomm and several other chip designers and manufacturers have been experimenting with the ideas with great progress in recent years.
Besides enhanced “brain-like thinking” capabilities of such machines, another key benefit from neuromorphic chips is the energy conservation. Information storage and processing are now arranged inside the same interconnected neuron nodes. The cost of energy and heat from the switching, such as those in between memory and CPU in conventional chips, has been drastically reduced, resulting in better performance in general.
Due to its significant disruptive nature (to both future hardware and software) and high potential commercial clout, the World Economic Forum’s Meta-Council on Emerging Technologies ranked neuromorphic technology as one of the Top 10 Emerging Technologies of 2015. The interesting facts today are that although prototypes of the neuromorphic chips are available, great software to demonstrate their “brain” power are yet to come out.
What are the best ways to test the intelligence level of a machine-made brain? And where would we use it first, on a robot called Chappie?