Designing Deep Learning Hardware Accelerator and Efficiency Evaluation

Table 1

AlexNet model parameter architecture [14].

Layer
CONV1	3	96	227 × 227	55 × 55	11 × 11	4
POOL1	96	96	55 × 55	27 × 27	3 × 3	2
CONV2	48	256	27 × 27	27 × 27	5 × 5	1
POOL2	256	256	27 × 27	13 × 13	3 × 3	2
CONV3	256	384	13 × 13	13 × 13	3 × 3	1
CONV4	192	384	13 × 13	13 × 13	3 × 3	1
CONV5	192	256	13 × 13	13 × 13	3 × 3	1
POOL5	256	256	13 × 13	6 × 6	3 × 3	2
FC6	9216	4096	1 × 1	1 × 1	—	—
FC7	4096	4096	1 × 1	1 × 1	—	—
FC8	4096	1000	1 × 1	1 × 1	—	—