Hardware designed for efficient deep net algorithms

Optical computers, which transmit and store data using photons rather than electrons, and quantum computers, which represent data in multiple states simultaneously and have the potential to exponentially increase computing power, are among the alternatives being investigated as deep nets push classical computers to their limits.

Researchers have concentrated on adapting the current chip to the requirements of deep learning until a new paradigm is established.

The discovery that video game graphical chips, or GPUs, could accelerate deep-net training by performing massively parallelized matrix computations sparked the trend.One of the mainstays of modern AI are GPUs, which have led to new ideas for using specialized hardware to improve deep net efficiency.