Integer-based fused convolutional layer in a convolutional neural network

1. A fused convolutional layer, comprising:

a comparator configured to receive a first zero point and a multiply-accumulation result;

a first multiplexer coupled to the comparator, wherein the first multiplexer receives a plurality of power-of-two exponent values;

a shift normalizer, coupled to the first multiplexer, wherein the shift normalizer is configured to receive the multiply-accumulation result and the plurality of power-of-two exponent values, wherein the shift normalizer limits a quantization of the multiply-accumulation result to a power-of-two scale; and

a second multiplexer coupled to an output of the shift normalizer, wherein the first multiplexer receives a second zero point and outputs an activation.

2. The fused convolutional layer of claim 1 , wherein the multiply-accumulation result is an integer variable.

3. The fused convolutional layer of claim 1 , wherein the first zero point is fused.

4. The fused convolutional layer of claim 1 , wherein the first zero point is an approximated integer variable.

5. The fused convolutional layer of claim 1 , wherein the activation is an integer variable.

6. The fused convolutional layer of claim 1 , wherein the quantization is one of symmetric, asymmetric, layer-wise, or channel-wise.

7. The fused convolutional layer of claim 1 , wherein the activation is one of ReLU, pReLU, ReLUx, or hard-sigmoid.