A new computer architecture inspired by Human Brain Will Help to Advance AI Tech

As Intelligent As The Human Brain

In the latest design, researchers merged the memory and processing components of a computer. Computers that are typical have central processors which is responsible for executing logic as well as computing tasks including memory units with mass storage, as well as out and input devices.

The human brain does not require all the same components. The synapses are used as processing and storage units. Combining the processor and memory unit is much more efficient, and could reduce power consumption.

Visit Computer Repairs Geebung offers nothing but the best desktop computer repairs in the whole area of Geebung and nearby Brisbane suburbs. We take in all brands and models of desktop computers both old and new machines.

“If you consider human beings, they perform computation using between 20 and 30 Watts of power, but AI today is built on supercomputers that run megawatts or kilowatts of power,” explained Abu Sebastian One of the researchers behind the study. “In neurons, synapses play a role in processing and storing data. In a brand new system that goes over von Neumann, memory has been able to play a greater part in the process of computing.”

Researchers also studied the brain’s synaptic structure to design arrays of phase-change memory devices that “accelerate the process of training deep neural networks.” Through electrical pulses, researchers were able to construct storage that can hold more data in one nanoscale device.

A New Game

Sebastian and his team at IBM have already conducted tests of their idea using an unsupervised machine-learning algorithm earlier this year. For comparison, they performed the same test using a traditional computer.

“We had always believed that the systems we tested to perform superior to traditional computing systems for certain tasks, however we were pleasantly surprised by that how much more efficient these methods are,” he stated.

The prototype for the new computing system IBM researchers developed has achieved 200 times more performance than the standard von Neumann computing system.