In Deep Neural Networks (DNN), the depth-wise separable convolution has often replaced the standard 2D convolution having much fewer parameters and operations. Another common technique to squeeze DNNs is heterogeneous quantization, which uses a different bitwidth for each layer. In this context we propose for the first time a novel Reconfigurable Depth-wise convolution Module (RDM), which uses multipliers that can be reconfigured to support 1, 2 or 4 operations at the same time at increasingly lower precision of the operands. We leveraged High Level Synthesis to produce five RDM variants with different channels parallelism to cover a wide range of DNNs. The comparisons with a non-configurable Standard Depth-wise convolution module (SDM) on a CMOS FDSOI 28-nm technology show a significant latency reduction for a given silicon area for the low-precision configurations.

A Reconfigurable Depth-Wise Convolution Module for Heterogeneously Quantized DNNs / Urbinati, Luca; Casu, Mario R.. - ELETTRONICO. - (2022), pp. 128-132. ((Intervento presentato al convegno 2022 IEEE International Symposium on Circuits and Systems (ISCAS) tenutosi a Austin, Texas, USA nel 27 May 2022 - 01 June 2022 [10.1109/ISCAS48785.2022.9937753].

A Reconfigurable Depth-Wise Convolution Module for Heterogeneously Quantized DNNs

Urbinati, Luca;Casu, Mario R.
2022

Abstract

In Deep Neural Networks (DNN), the depth-wise separable convolution has often replaced the standard 2D convolution having much fewer parameters and operations. Another common technique to squeeze DNNs is heterogeneous quantization, which uses a different bitwidth for each layer. In this context we propose for the first time a novel Reconfigurable Depth-wise convolution Module (RDM), which uses multipliers that can be reconfigured to support 1, 2 or 4 operations at the same time at increasingly lower precision of the operands. We leveraged High Level Synthesis to produce five RDM variants with different channels parallelism to cover a wide range of DNNs. The comparisons with a non-configurable Standard Depth-wise convolution module (SDM) on a CMOS FDSOI 28-nm technology show a significant latency reduction for a given silicon area for the low-precision configurations.
978-1-6654-8485-5
File in questo prodotto:
File Dimensione Formato  
A_Reconfigurable_Depth-Wise_Convolution_Module_for_Heterogeneously_Quantized_DNNs.pdf

non disponibili

Tipologia: 2a Post-print versione editoriale / Version of Record
Licenza: Non Pubblico - Accesso privato/ristretto
Dimensione 727.83 kB
Formato Adobe PDF
727.83 kB Adobe PDF   Visualizza/Apri   Richiedi una copia
ISCAS_2022_Luca.pdf

accesso aperto

Tipologia: 2. Post-print / Author's Accepted Manuscript
Licenza: PUBBLICO - Tutti i diritti riservati
Dimensione 673.11 kB
Formato Adobe PDF
673.11 kB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2973053