Implementing machine learning at the edge creates challenges related to power, performance, thermals, privacy, reliability, and cost. Today’s state of the art digital computation engines are efficiency limited for the heavy mathematical operations encountered in machine learning. One of the initiatives to improve AI efficiency in the industry is to use analog techniques that exploit transistor device physics significantly better than digital approaches ever can. Blumind has developed an analog compute technology that delivers microwatt power, microsecond latency and low cost while directly interfacing with analog sensors in our inherently analog world. This session will introduce Blumind’s analog compute architecture. We will also discuss new use cases that will be activated with this technology.