MIT new chip to enhance AI neural network instruction cycle seven times less energy consumption of 95%

AI development potential is optimistic about all parties, but AI operations need to consume a lot of energy, but it is behind the development of the hidden danger.MIT recently announced a new chip, seven times faster than the traditional AI chip neural network instruction cycle, reducing energy consumption by 95% In the future, if the popularization of AI can withstand the general consumer devices such as mobile phones installed above.

Face recognition, speech recognition, which we used to use, are behind the results of the neural network operations.Generally speaking, the neural network is very large, so to put the phone often only has a simplified architecture, can not play the nerve network's strong ability, or The new MIT chips reduce the amount of data passed to the memory and computing modules of the chip, increasing the computational efficiency by three to seven times, accelerating the speed and reducing the power consumption by 95%.

If MIT's new chips popularize, these new chips can drive local neural networks to home-based smart devices, and future human voice commands can be handled locally without having to be connected to the Web as now or Amazon Alexa Spread to the data center, and then back.

When Apple released the A11 bionic chip in 2017, it had the AI ​​capability to identify if it was the owner of a handset using Face ID. A similar application in the future will not be only on the iPhone, or Face ID, but more devices and applications .