New Electronic Switch Mimics Behavior of Biological Neuron
University of Michigan engineers are developing a new type of electronic switch that mimics the behavior of a biological neuron in the human brain, which is able to perform complex tasks much more efficiently than regular computers. Like brains, these computers would be able to reach beyond their local neighborhood to establish connections, and grow stronger connections the more they are used. Rather than being based on a separation of memory and CPU as traditional computers, this new computing architecture would allow computations to happen all over the design. These artificial synapses would be created through 'memristors' built through nanotechnology that can be used to connect a large number of processes, thereby creating a type of computer that can learn without being programmed.
Transcript
00:00:03 Our brains are very powerful computers. We can process very complex tasks much more efficiently than conventional computers. So if you've watched the movie Avatar, the home tree is a very good metaphor for neurons So if you think about it, you have one tree, one neuron, which has thousands of leaves, which are then connected to thousands of other neurons The connections between each neuron and its neighbors, is not static It actually can be adapted In fact, it's adapted constantly on the flat So that's why we can learn things, we can do tasks that were not told to us, where computers, we have to tell them what to do and train them for repetitive tasks Basically we are developing a new type of electronic switch
00:00:46 and we are primarily interested in this device because it mimics the behavior of a biological neuron Conventional computers are built out of transistors It's just a digital switch, but each individual device can only connect to a few other devices in its local neighborhood; whereas in a biological logical system, you have these billions of neurons and millions of synapses that allow them to reach beyond their local neighborhood So our devices exhibit some of the dynamics of a synapse in that they grow stronger connections the more they're used and they also can connect to many other devices, not just what's locally around them
00:01:32 So this way, we can design a new computing architecture that is not based on a separation between memory and the CPU, which is what you normally find Instead, we're looking at a distributed computing, where computations happen all over, all over the design and they are all happening in parallel So this gives us a very different type of computing To enable this kind of connection, we are using nanoscale devices We developed a new type of device structure
00:02:04 These devices are so-called two-terminal resistive switches, or some people call them memristors A memristor is basically a resistor with memory It's a very simple structure so that it can be used to connect a large number of processing units I think more you would expect to see in the end is computers that, if they are built with our devices, they'll be good at things that a human is good at so driving, or identifying objects,
00:02:32 identifying faces They probably won't be as good at calculating exact figures, but you will see them able to do the kind of thing where they can learn without being taught or without being explicitly programmed