The Loihi chip from Intel is based on the human brain. newsroom.intel
Intel has turned to the human brain to inspire the design for its latest chip. The experimental chip called "Loihi” uses an unproven type of technology called "neuromorphic". The new chip ignores the traditional design of logic gates and instead uses "spiking neurons" as its fundamental computing unit. These computer neurons can move signals along in a similar way to our own brains neurons. The Loihi chip contains around 130,000 simulated neurons. To put that into perspective, a lobster has about 100, 000 neurons and the human brain has about 80 billion neurons.
Design based on brain increase workload but uses less power
The theory of turning to the human brain as a base model for processing instead of other ways is that using this method may allow for an increase in the speed of machine learning without the massive power requirements. Human brains work by communicating information with pulses or spikes. Brain cells don’t work alone; the activity of one neuron affects another as the information is passed in a relayed fashion. This means groups of cells working together are what increase intelligence and learning. If we think about this in terms of the chip, what we can see if that the energy usage is shared among groups rather being ‘on’ all the time. Human brains work by relaying information with pulses or spikes, strengthening frequent connections and storing the changes locally at synapse interconnections. As such, brain cells don't function alone because the activity of one neuron directly affects others.
Intel says new chip has potential for use in personal robots
These types of chips could lead to the perfect kind of AI -- ones that can learn as they go, rather than being taught a long list of tasks prior to beginning their designated job. Intel already has their eyes on the applications of this chip saying, "The test chip [has] enormous potential to improve automotive and industrial applications as well as personal robots.” While neuromorphic sound good in theory they have yet to show their true potential up against the current deep-learning technology. But that isn’t deterring the big players to try and keep experimenting with this new technology. IBM built a neuromorphic chip called "TrueNorth" that has 4,096 processors simulating around 256 million synapses. But even with the potential power Facebook's deep learning specialist Yann LeCun has said that they are convinced the chip could stand up to complex tasks like image recognition.
Chip to be shared with leading AI researchers
Intel is determined to push Loihi as far as it can and will look to “"leading university and research institutions" as a testing ground for the chips. They plan to let these institutions who have a serious Ai agenda have free range with the chips in an effort to push their development and potential. The chips will be created using Intel's 14-nanometer process technology and be distributed in November this year.