IP개요 |
In this paper, we present a low cost and low power neuromorphic core for embedded AI system. The neuromorphic core is based on k-Nearest Neighbor (k-NN) and Radial Basis Function (RBF) algorithms. In order to reduce the operation workload in calculating distance, it adopts Manhattan distance (L1) instead of Euclidean distance (L2). Also, the neuromorphic controller adopts more efficient finite state machine (FSM) optimized with instructions to generate the concise instructions. Therefore, the neuromorphic core trains dataset in N-Cells rapidly and classifies test data with low power consumption. We will fabricate the low cost and low power neuromorphic core for embedded AI system with Samsung 65 um RFCMOS technology. The neuromorphic core operates in 1.2V, 50MHz frequency and the circuit is digital type. |