Current hardware for artificial intelligence is inefficient. For today's supercomputer center, it takes a long time, huge datasets and the energy of a whole power plant to train a high-end AI model, resulting in high costs and unsustainability. The problem is rooted in the fundamental architecture of today’s digital hardware itself, since it has nothing in common with the way the human brain works. GEMESYS Technologies offers an analog chip design based on the same information-processing mechanisms as the human brain. This enables AI hardware vendors to distribute a novel chip, that trains neural networks 20,000 times more energy-efficient than current technology. It not only significantly reduces the cost, time and amount of data required to train a neural network, but also increases overall quality as well as performance. Its small size and high energy efficiency allows it to be embedded in nearly every device, enabling decentralized on the edge training, data processing and decision making. 



Semiconductor

Description

People (4)

Dr.-Ing. Dennis Michaelis

Daniel Krüger

Dr.-Ing. Enver Solan

Moritz Schmidt

Jobs