After years of research and development, Uniquify, a Silicon Valley neural network and AI edge computing company, is ready to unveil neural network 2.0 technology at the CES 2022 event.
At the conference, Uniquify will showcase transformative neural network fabric technology. Currently, neural network technology is used in creating visual, audio, data, and natural language processing (NLP) models with the multiply-accumulate (MAC)-based operations. But with Uniquify’s second-generation neural network 2.0 technology, neural networks shrink neurons by using proprietary AI processing elements (AIPEs) in place of MAC operations.
Here’s how it works
AIPE technology shrinks the neurons in neural networks to enable the creation of the most advanced and complex AI visual, audio, and NLP models. In the past, MAC hardware was used to implement advanced but bulky neural network models, which severely hindered the possibilities of edge computing. Now the game-changer in AI edge computing is neural network 2.0 technology, which can reduce the cost and area required for AI edge computing systems. AIPE technology allows edge computing to be deployed anywhere and everywhere, including consumer markets, data centres, enterprises, manufacturing, health and medicine, and government applications.
The majority of neural network operations—such as CNN, RNN, FNN, AE, Batch Norm, and activation functions—comprise multiplication and addition operations conducted in inference mode (prediction mode). In hardware, MAC blocks can be used to implement these operations. However, Uniquify’s AIPE block is 20 times smaller, cheaper, and more power-efficient than the conventional MAC block. Plus, it can be stacked up to achieve 70 Tera Ops per Watt on a 7nm chip.
As a result of such efficient silicon real estate use, neural network 2.0 has a wide range of applications, including consumer appliances, the artificial intelligence of things (AIoT), surveillance, vision inspection, diagnosis, and autonomous driving.
The core business of Uniquify includes neural network platform and modelling, AI edge computing, consumer application SoC development, and semiconductor IP licensing.
Experts predict that AI models will become increasingly large and complex in the near future. That makes neural network 2.0 the best technology to support the virtually limitless market segments and industries that are ready to adopt neural networks and AI.