Method and Apparatus for Performing a Neural Network Operation

An integrated circuit comprising:

• an input configured to receive weight and bias values of a deep neural network (DNN);
• a control module coupled to the input;
• a neural network engine coupled to the control module, the control module configured to send an array of feature values to the neural network engine and to send a bias value and set of weight values for a single layer of the DNN to the neural network engine, the neural network engine comprising:
• one or more multiply and accumulate circuit (MAC) coupled to the control module, the one or more MAC configured to receive the array of feature values, the bias value and the set of weight values for a single layer of the DNN, the one or more MAC configured to perform multiply-and-accumulate operations on the single layer of the DNN to obtain a sum corresponding to each neuron in the single layer of the DNN;
• a neural network engine controller coupled to the one or more MAC, the neural network engine controller configured to couple a layer output value corresponding to each neuron in the single layer of the DNN to a corresponding input of the one or more MAC, where each layer output value includes of one of the sums or a function of the one of the sums;
• and
• one or more activation function module coupled to the one or more MAC,
• wherein the control module is configured to repeat the coupling the bias value and the set of weight values for the single layer of the DNN to the neural network engine, the one or more MAC are configured to repeat the performing multiply-and-accumulate operations on the single layer of the DNN and the neural network engine controller is configured to repeat the coupling the layer output value to the corresponding input of the one or more MAC until multiply-and-accumulate operations have been performed on all layers of the DNN to generate an output-layer-sum corresponding to each output-layer neuron, and
• wherein the activation function module is configured to perform an activation function on each output-layer-sum to generate one or more DNN output values representing an output layer of the DNN.

A flash controller including a read module, a status module coupled to the read module, a decode module coupled to the read module and a write module, the flash controller comprising:

• an input configured to receive weight and bias values of a deep neural network (DNN);
• a control module coupled to the input;
• a neural network engine coupled to the control module, the control module configured to send an array of feature values to the neural network engine and to send a bias value and set of weight values for a single layer of the DNN to the neural network engine, the neural network engine comprising:
• one or more multiply and accumulate circuit (MAC) coupled to the control module, the one or more MAC configured to receive the array of feature values, the bias value and the set of weight values for a single layer of the DNN, the one or more MAC configured to perform multiply-and-accumulate operations on the single layer of the DNN to obtain a sum corresponding to each neuron in the single layer of the DNN;
• a neural network engine controller coupled to the one or more MAC, the neural network engine controller configured to couple a layer output value corresponding to each neuron in the single layer of the DNN to a corresponding input of the one or more MAC, each layer output value includes of one of the sums or a function of the one of the sums;
• and
• one or more activation function module coupled to the one or more MAC,
• wherein the control module is configured to repeat the coupling a bias value and a set of weight values for a single layer of the DNN to the neural network engine, the one or more MAC are configured to repeat the performing multiply-and-accumulate operations on the single layer of the DNN and the neural network engine controller is configured to repeat the coupling the layer output value to a corresponding input of the one or more MAC until multiply-and-accumulate operations have been performed on all layers of the DNN to generate an output-layer-sum corresponding to each output-layer neuron, and
• wherein the activation function module is configured to perform an activation function on each output-layer-sum to generate one or more DNN output values representing an output layer of the DNN;
• wherein one or more of the read module, the status module, the decode module and the write module are configured to perform a function using the DNN output values.