This deep neural networks addressed the Amharic Named Entity Recognition (ANER) problem by employing a semi-supervised learning approach based on neural networks. Potential feature information represented as word vectors are generated by neural network from unlabeled Amharic text files. These generated features are used as features for classification. This approach aims at automating manual feature design and avoiding dependency on other natural language processing tasks. Word2vec tool with skip-gram model is used for generating word feature vectors in our experiments. BLSTM(bi-directional long short-term memory), LSTM(long short-term memory) and MLP(multi layer perceptron) deep neural networks are tested. From the experiments 92.6%, 90.4%, 90.6% F-Score was achived using BLSTM, LSTM, MLP Neural networks respectively. This shows that, automatic feature extraction has given better performance while reducing the effort in manual feature design.