IBM developed back in 2014 the chip SyNapse (on DARPA funding for the project named -guess what- SyNAPSE) mimicking neurones and their interconnections in a brain. It had 1 million neurones equivalent capacity with 256 million synapses (using 5.6 billion transistors). In 2015 they designed TrueNorth, an evolution of Synapse, with an amazingly low power consumption, 70mW -1,000 times less than an equivalent computer-. TrueNorth has been designed as a module that can be stacked to create more powerful "brains".
Now IBM has delivered a supercomputing platform based on 16 TrueNorth chips to the Lawrence Livermore National Laboratory that plans to use it for deep learning applications. It provides an equivalent of 16 million neurones and 4 billion synapses. To give you a feeling this is the number of neurones in a frog brain. Not a particulalry smart creature by our standards but nevertheless pretty good (better than you and me) in recognising bugs and catching them in a blink of an eye… A mouse has some 70+ million neurones and 100 billion synapses. Our brain goes short of a 100 billion neurones (86 billions in a mature male according to some scientists, but the number changes -decreases- over time) and some million billion synapses.
LLNL is planning to use the chip to evaluate and increase security of the US nuclear stockpile. The brain like supercomputer is pretty good at analysing patterns and finding "clues", just like our brain does (look at the sky and don’t you often see clouds in shape of animals, faces… that is pattern recognition).
The supercomputer doesn’t come "cheap" with a price tag of 1 million $ but still it is cheaper in terms of processing power in its own domain than a classical supercomputer. In addition it consumes "nothing", just 2.5 W, less than your bedpost LED lamp!
Consider that the price does not correspond to the production cost, it just reflect the cost of investment required. In ten years time I am pretty sure that this kind of "brains" could be available at a few dollars a dozen and will find their way into many objects. This can be a game changer, since it will provide some kind of awareness to things we interact with everyday. Now we are interacting with atoms (and often need a manual to learn how to interact), in the next decade we will like interact with objects as we interact among ourselves.
This will also give a boost to robots and their capabilities.