Why get error "SVD did not converge'' when adding a LSTM layer using Keras?

Question

I am using Keras for end-to-end learning (tensorflow as backend). The model is very simple:

model = Sequential()
model.add(LSTM(units=neurons,input_shape=(x_train