IBM Research has tested a power-efficient analog-to-digital chip for performing deep neural network inference tasks. This was reported in the company’s blog.
The chip was developed at the Nanotechnology Facility in Albany, New York. It consists of 64 analog computing cores (tiles) in-memoryeach containing a cross-sectional array of synaptic cells measuring 256 by 256. According to the engineers, this is the first development of its kind, which showed the same results as digital counterparts.
“The chip demonstrated 92.81% accuracy on the CIFAR-10 image dataset. We believe that this is the highest accuracy among all currently presented chips using similar technology, ”the company said in a statement.
IBM noted that one of the advantages of the chip is maximum energy efficiency. The technology, in particular, significantly reduces the amount of information exchanged between the memory and the processor.
In a commentary for the BBC, Thanos Vasilopoulos of IBM’s Swiss lab explained that the new development will be able to “perform increasingly complex tasks in low power conditions.”
“In addition, cloud providers will be able to use these chips to reduce electricity costs and reduce carbon emissions,” the company representative concluded.
Earlier, IBM announced the launch of watsonx, a business-oriented artificial intelligence and data management platform. As the developers expect, it will greatly facilitate the integration of AI into commercial projects.
Recall that in May 2023, the American tech giant announced the reduction of about 7,800 employees. The company explained that the reorganization will affect those jobs that artificial intelligence can handle.
Found a mistake in the text? Select it and press CTRL+ENTER
ForkLog Newsletters: Keep your finger on the pulse of the bitcoin industry!