IBM unveils a ‘green’ chip for analog AI

IBM Research has tested an energy-efficient analog-digital chip designed to run inference on deep neural networks. The news is in the company’s blog.

The chip was developed at a nanotechnology campus in Albany, New York. It comprises 64 analog computing cores (tiles) in-memory, each containing a crossbar array of 256-by-256 synaptic cells. According to the engineers, this is the first development of its kind to deliver results on par with digital counterparts.

“The chip demonstrated an accuracy of 92.81% on the CIFAR-10 image dataset. We believe this is the highest accuracy among all chips currently presented that use similar technology,” the company said.

IBM noted that one of the chip’s advantages is maximum energy efficiency. The technology, in particular, substantially reduces the amount of data exchange between memory and the processor.

In a BBC interview, IBM’s Thanos Vasiliopulos explained that the new development will be able to “perform increasingly complex tasks under low power.”

“In addition, cloud providers could use these chips to cut electricity costs and reduce carbon dioxide emissions,” the company said.

announced the launch of watsonx, a business-focused AI and data-management platform. Developers expect it will significantly ease the integration of AI into commercial projects.

Earlier in May 2023, the American tech giant announced a headcount reduction of about 7,800 employees. The company said the restructuring would affect roles that AI can handle.