FIGURE Q$4$ shows a sequence of computations needed in convolutional neural network $($CNN$)$ implementation. The first four steps execute three times $($repeatedly$)$ before moving to the fifth computational step $($i$.$e$.$ Dropout $0\mathrm{.}1$ rate$)$ which will proceed sequentially to the final step $($Predictions$).$ The delay of each block is given next to the block. We plan to implement this as an optimal pipeline. Given the clock overhead is $0\mathrm{.}15$ ns and the average pipeline interruption rate is expected to be $0\mathrm{.}18.$

a$.$ Determine the optimum number of pipeline stages for this system.

b$.$ Based on part $($a$),$ design the optimal pipeline implementation and determine the time it will take to process one input image through the pipeline $($i$.$e$.$ pipeline latency$).$ Sketch and explain

Cesign Test

the pipeline architecture that best achieves this requirement.

c$.$ For a machine learning application, you are required to train the CNN network using $10,000$ images. What is the minimum time required to complete the training based on the pipeline in part $($b$).$