According to foreign media reports, engineers at the Massachusetts Institute of Technology (MIT) have recently developed a new type of neural network chip that can reduce the power consumption of neural network information transmission by 95%. This will allow battery-powered mobile devices to run neural network programs with minimal power consumption.
It is reported that the MIT engineer developed this neural network chip, which can greatly reduce the need to transfer data between the chip memory and the processor to achieve a 95% reduction in power consumption. Chips designed by MIT engineers will be used in battery-powered mobile devices, such as smart phones for digital assistants, real-time translation and other artificial intelligence services, which require huge energy consumption problems caused by the use of the cloud to transmit data.
In general, a neural network consists of thousands of artificial neurons interconnected by layers. A single neuron receives input from the next layer of neurons. Once the combined input exceeds the threshold set during training, it is output. Go to multiple neurons on the upper layer. This means that a single neuron, chip retrieves the connection weights of a particular connection input data and memory, multiplies the stored results, and then repeats the process each time it is entered. This requires a lot of mobile data and energy consumption.
MIT engineers use an analog circuit to compute all inputs in parallel in memory. This not only reduces the amount of data that needs to be pushed, but also saves a lot of energy. This is the first powerful convolutional neural network to use this method to run image-based artificial intelligence applications.
Of course, this is not the first time developers have used memory to create processing data to reduce neural network power consumption. Dario Gill, IBM's vice president of artificial intelligence, previously said, "The results show that its performance is impressive when using memory arrays for convolution. It will certainly provide future image and video classification for the Internet of Things. More complex convolutional neural networks."
The development of memory chips that can process data will help to equip smart phones, home appliances, and various IoT devices with artificial intelligence. At the same time, it also further stimulated the investment and development of Silicon Valley giants in the field of low-power artificial intelligence chips.
Hd Screen Protector,High-Definition Screen Protector,Hd Clear Screen Protector,Transparent Screen Film
Shenzhen TUOLI Electronic Technology Co., Ltd. , https://www.hydrogelprotectors.com