Compact cam array for decision tree inference (tree-cam)

Abstract

'Tree-CAMs' constructed to implement decision trees more efficiently are provided—namely with less hardware (i.e., fewer memristors and transistors), less power consumption, and less memristor programming time than existing aCAMs used to implement decision trees. A tree-CAM realizes these optimizations by leveraging a shared comparison sub-circuit that stores a threshold shared among evaluable conditions for multiple root-to-leaf-paths of a decision tree. The threshold may be evaluated against a common feature of a feature vector. The shared comparison sub-circuit may comprise just a single memristor programmed to store the threshold. Accordingly, the tree-CAM can represent/implement evaluable conditions (sharing a common threshold and evaluated against a common feature) across multiple root-to-leaf paths of a decision tree using just a single memristor—thus reducing the number of memristors required to implement the decision tree.