A big part of TinyML is finding alternate paths for implementing AI in resource-constrained space. We already have SoC that is functionally not so much differe

A big part of TinyML is finding alternate paths for implementing AI in resource-constrained space. We already have SoC that is functionally not so much different from modern microcontrollers. Depending on economics of scale, it isn’t that big of a leap of imagination to see a microcontroller which includes 4-bit vector ops like a GPU.