Stochastic logics for low cost and low power wearable applications

摘要

This paper gives a special insight standing on an unusual computing algorithm ??? stochastic computing ??? to discuss the strategy on designing low cost and low power smart wearable devices from the system architecture view. Some new stochastic logics are introduced to approximate complex mathematic functions like the Gaussian function with simple circuits, which are used to build the low cost and low power artificial neural network for data intensive applications for the smart wearable devices. The data process kernel is distributed close to the data source. An experiment on the facial recognition shows how the system complexity, speed, cost, power, and flexibility can be improved by the stochastic logic architecture.