The architecture diagram in the article resembles the approach Apple took in the design of their neural engine. https://www.patentlyapple.com/2021/04/apple-rev

The architecture diagram in the article resembles the approach Apple took in the design of their neural engine. https://www.patentlyapple.com/2021/04/apple-reveals-a-multi-mode-planar-engine-for-a-neural-processor-that-could-be-used-in-a-series-screa.html https://www.patentlyapple.com/2021/04/apple-reveals-a-multi-... Typically these architectures are great for compute. How will it do on scalar tasks with a lot of branching? I doubt well.