TY - GEN
T1 - Redundant neurons and shared redundant synapses for robust memristor-based DNNs with reduced overhead
AU - Zhang, Baogang
AU - Uysal, Necati
AU - Fan, Deliang
AU - Ewetz, Rickard
N1 - Publisher Copyright:
© 2020 Association for Computing Machinery.
PY - 2020/9/7
Y1 - 2020/9/7
N2 - The dominating computational workload in the inference phase of deep neural networks (DNNs) is matrix-vector multiplication. An arising solution to accelerate the inference phase is to perform analog matrix-vector multiplication using memristor crossbar arrays (MCAs). A key challenge is that stuck-at-fault defects may degrade the classification accuracy of the memristor-based DNNs. A common technique to reduce the negative impact of stuck-at-faults is to utilize redundant synapses, i.e, each row in a weight matrix is realized using two (or r) parallel rows in an MCA. In this paper, we propose to handle stuck-at-faults by inserting redundant neurons and by sharing redundant synapses. The first technique is based on inserting redundant neurons to surgically repair neurons connected to rows and columns in the MCAs with many stuck-at-faults. The second technique is focused on sharing redundant synapses between different neurons to reduce the hardware overhead, which generalizes (1:r) synapse redundancy in previous studies to (q:r) synapse redundancy. The experimental results demonstrate new trade-offs between robustness and hardware overhead without requiring the neural networks to be retrained. Compared with state-of-the-art, the power and area overhead for a neural network can be reduced with up to 16% and 25%, respectively.
AB - The dominating computational workload in the inference phase of deep neural networks (DNNs) is matrix-vector multiplication. An arising solution to accelerate the inference phase is to perform analog matrix-vector multiplication using memristor crossbar arrays (MCAs). A key challenge is that stuck-at-fault defects may degrade the classification accuracy of the memristor-based DNNs. A common technique to reduce the negative impact of stuck-at-faults is to utilize redundant synapses, i.e, each row in a weight matrix is realized using two (or r) parallel rows in an MCA. In this paper, we propose to handle stuck-at-faults by inserting redundant neurons and by sharing redundant synapses. The first technique is based on inserting redundant neurons to surgically repair neurons connected to rows and columns in the MCAs with many stuck-at-faults. The second technique is focused on sharing redundant synapses between different neurons to reduce the hardware overhead, which generalizes (1:r) synapse redundancy in previous studies to (q:r) synapse redundancy. The experimental results demonstrate new trade-offs between robustness and hardware overhead without requiring the neural networks to be retrained. Compared with state-of-the-art, the power and area overhead for a neural network can be reduced with up to 16% and 25%, respectively.
UR - http://www.scopus.com/inward/record.url?scp=85091304373&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85091304373&partnerID=8YFLogxK
U2 - 10.1145/3386263.3406910
DO - 10.1145/3386263.3406910
M3 - Conference contribution
AN - SCOPUS:85091304373
T3 - Proceedings of the ACM Great Lakes Symposium on VLSI, GLSVLSI
SP - 339
EP - 344
BT - GLSVLSI 2020 - Proceedings of the 2020 Great Lakes Symposium on VLSI
PB - Association for Computing Machinery
T2 - 30th Great Lakes Symposium on VLSI, GLSVLSI 2020
Y2 - 7 September 2020 through 9 September 2020
ER -