Developing Logic Synthesis Flow for NVDLA IP
Lindqvist, Aarno (2022)
Lindqvist, Aarno
2022
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2022-05-19
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202205134836
https://urn.fi/URN:NBN:fi:tuni-202205134836
Tiivistelmä
Modern digital devices require high computing performance; thus, markets have a huge
demand for SoC. The most powerful SoC are implemented on ASIC chips since, it is the most cost-efficient technology when production volumes are high. An important step on ASIC design process is the logic synthesis. By utilizing dedicated software tool, it transfers RTL code into gate level netlist. The logic synthesis process is executed multiple times alongside the RTL code development to meet the desired specifications for the chip.
This thesis project used the NVDLA IP as a use case to execute logic synthesis. NVDLA is an open-source deep learning accelerator developed by NVIDIA. The design is able to execute CNNs making it efficient. Each component in the NVDLA can be configured independently, which make it flexible and cost effective. NVDLA software ecosystem has extensive cover of software features. NVDLA is divided into five partitions according to their functionality. Each partition is an individual top-level synthesis hierarchy.
The target of this thesis is to develop a logic synthesis flow for NVDLA in the company design environment. This was achieved by exploiting NVDLA design environment, company internal memory wrapper, and Synopsys Design Compiler and IC Compiler 2 tools to execute logic synthesis for TSMC 7 nm standard cell technology. All the used RTL codes and scripts were downloaded from NVDLA GitHub webpage. The memory wrapper was created by the company memory wrapper tool. It connects the NVDLA design and the RAM instances. The Design Compiler tool was used to generate the initial netlist for NVDLA partitions. The IC Compiler 2 tool was used to create individual floorplans for each partition. The generated DEF file was used for second pass synthesis to obtain the final logic synthesis results. The results demonstrate that the company design environment can be used to run synthesis for open-source IP blocks. Further, the developed flow provides a platform to exploit different kind of open-source IP’s on industrial development environment since, it can generate synthesis results for 7 nm standard cell technology quickly.
demand for SoC. The most powerful SoC are implemented on ASIC chips since, it is the most cost-efficient technology when production volumes are high. An important step on ASIC design process is the logic synthesis. By utilizing dedicated software tool, it transfers RTL code into gate level netlist. The logic synthesis process is executed multiple times alongside the RTL code development to meet the desired specifications for the chip.
This thesis project used the NVDLA IP as a use case to execute logic synthesis. NVDLA is an open-source deep learning accelerator developed by NVIDIA. The design is able to execute CNNs making it efficient. Each component in the NVDLA can be configured independently, which make it flexible and cost effective. NVDLA software ecosystem has extensive cover of software features. NVDLA is divided into five partitions according to their functionality. Each partition is an individual top-level synthesis hierarchy.
The target of this thesis is to develop a logic synthesis flow for NVDLA in the company design environment. This was achieved by exploiting NVDLA design environment, company internal memory wrapper, and Synopsys Design Compiler and IC Compiler 2 tools to execute logic synthesis for TSMC 7 nm standard cell technology. All the used RTL codes and scripts were downloaded from NVDLA GitHub webpage. The memory wrapper was created by the company memory wrapper tool. It connects the NVDLA design and the RAM instances. The Design Compiler tool was used to generate the initial netlist for NVDLA partitions. The IC Compiler 2 tool was used to create individual floorplans for each partition. The generated DEF file was used for second pass synthesis to obtain the final logic synthesis results. The results demonstrate that the company design environment can be used to run synthesis for open-source IP blocks. Further, the developed flow provides a platform to exploit different kind of open-source IP’s on industrial development environment since, it can generate synthesis results for 7 nm standard cell technology quickly.