Typical computers have logic and memory centers that are segregated and connected to only a handful of neighbors. They operate in a linear fashion as a result. While they excel at linear code, and computations involving limited variables, they lag in other areas. Building computers that operate like brains, in parallel, will enable sophisticated processes to occur in much less time. By connecting traditional circuits with memristors, synapse-inspired transistors, researchers are moving closer to biological styles of computing. The next step is to scale up into a larger system that will hopefully replicate supercomputer capacity in something the size of a 2L beverage container. Their DARPA funding will go a long way to ensure that this transformative research will continue to advance.
Systems using cat-inspired algorithms and program architectures could be more capable of learning, recognition, complex decision making, and simultaneous task performance. Current supercomputers can rival cats in some arenas, but contain 140,000+ CPUs and a dedicated power supply. They are still 83 times slower than a cat’s brain.
The goal is to build a computer the way nature builds a brain. Researchers chose a cat brain because it is more feasible than a human brain, but is still very difficult to emulate the complexity and efficiency. The researchers build a memristor, a device replicating a biological synapse, and can replace traditional transistors. Now, they demonstrate that memristors can connect conventional circuits and support higher level computations, much closer to the types performed by organisms’ brains. Biological synapses link thousands of neurons in complex pathways that remember past interactions based upon strength and timing of signals. This “spike timing dependent plasticity” has been demonstrated and marks a major advancement in the field.
Supercomputer performance; sophisticated AI decision making, memory, recognition, and parallel processing.Edit Summary