Chipmaker Intel announces Loihi 2, the second generation of so-called neuromorphic research chips. The chips, and the associated software, are based on neurons in the brain.
The idea behind neuromorphic computers is that they imitate the behaviour of individual neurons in the brain. So calculations are done by a whole series of smaller units that communicate with each other via electrical peaks (‘spikes’), with increases and decreases in voltage. The terminology is similar to how we also talk about brains, where tiny neurons communicate via pulses of electricity. This should make the chips better suited for artificial intelligence, for example.
Intel is, therefore, already working on the second generation of these neuromorphic chips: Loihi 2. The processor will be upgraded with the new release, but the hardware will also be heavily modified. That should make it possible to run a whole host of new algorithms and discover patterns in data in more efficient ways. Their main advantage is their energy consumption, which should be considerably lower for complex tasks than with more traditional chips.
The chip consists of 128 cores, each made up of 8,192 components. They send electrical pulses to each other to form a kind of neural network that tackles a particular problem. Information between those ‘neurons’ is sent in traditional binary spikes (zeros and ones) or via graduated spikes (32bit) that imitate the pulses in our brains. Intel writes this in its extensive technical explanation, which you can read here (pdf). The company also says that Loihi 2 is ten times faster than its predecessor from four years ago.
The chip is primarily intended for research, not yet for use in the workplace. However, researchers can test it via the Intel Neuromorphic Research Cloud and the open-source framework Lava, which Intel has written for these chips.
