Much is said about Artificial Neural Networks today, due to the huge advance of the last few years. It is also known the enormous expenditure of time and energy to train these networks.
Neuromorphic computing is an alternative approach.
Today’s computers use Von Neumann’s architecture, for hardware organization. It has the CPU, peripherals, a clock to coordinate actions. In essence, it is centrally controlled serial processing.
Neuromorphic computing seeks to create hardware that mimics the human brain: elements that talk to each other, without a CPU.
It would be parallel processing in nature, much closer to what occurs in the brain.
In addition, current computers are digital (0 or 1), whereas neuromorphic ones can be analog.
Transistors x Memristors
Transistors are a basic element of computing today. The equivalent, in neuromorphic computing, would be the memristor — a junction of memory and transistor at the same time.
Currently there is a memory unit, and the data must be read for processing to be done by the transistors.
With the memristor, data and computing would be done together, just as the human brain works.
To be clear. The drastic change would be in terms of hardware, not just software (as it is today).
Deep learning failed to imitate the brain’s energy efficiency. Neuromorphic hardware can fix this.
The way forward is long and revolutionary. Let’s hope to suceed and continue to follow the evolution.