The US company IBM announced that the company's researchers have achieved a major breakthrough in memory computing technology, inventing an unsupervised machine learning algorithm that can run on 1 million phase change memory (PCM), which is expected to be Traditional computers are up to 200 times faster in terms of computational speed and energy efficiency, making them ideal for high-density, low-power, large-scale parallel computing systems in artificial intelligence applications. This achievement was published in the journal Nature News.
Memory computing or computable storage is a new concept in recent years. Its principle is to use the physical characteristics of memory devices to store and process information at the same time, instead of using the traditional built-in cache, and connecting the external memory through the bus. Architecture, but omitting the bus design, directly put the memory and CPU core together, to maximize the possibility of eliminating the delay caused by data migration during the calculation process, significantly improving the computational efficiency.
The phase change memory device used in IBM's memory computing architecture consists of two layers of iridium composites sandwiched between two electrodes. A weak current can heat it, and the internal state of the composite changes with temperature. From the amorphous form to the crystalline state, the operation is performed using the crystallization dynamics principle. This works in the same way as the artificial neuron released by IBM in August 2016.
In the future, IBM may continue to combine artificial neurons with in-memory computing architectures, like the brain, to make logic generation and information storage happen in the same place, which is very suitable for artificial intelligence computing. This achievement is another recent development of IBM's creation of an "artificial brain."
Vape Pen,OEM/ODM Vape Pen,Filling Portable Disposable Vape Stick,Pen Style Vape Pen
tsvape , https://www.tsecigarette.com