| Title |
Design and Verification of a Common Interface for Proprietary NPU Code Generation in General-purpose AI Compilers |
| Authors |
이제민(Jemin Lee) ; 권용인(Yongin Kwon) |
| DOI |
https://doi.org/10.5573/ieie.2023.60.10.29 |
| Keywords |
Nueral processing unit; Compiler; Deep learning |
| Abstract |
This paper proposes a common interface for connecting proprietary back-end compilers of NPU (Neural Processing Units) manufacturers with general-purpose AI compilers, enabling support for various AI models while protecting corporate assets. The proposed interface adheres to the ONNX standard and integrates ETRI's NEST-C compiler with OPENEDGES's ENLIGHT compiler through this interface. Experimental results showed that the integrated compiler achieved 100% result consistency for two types of models, ResNet50 and MobileNetV2, compared to using ENLIGHT alone. Therefore, the proposed interface offers a valuable solution for enhancing the versatility of NPU back-end compilers while maintaining their proprietary nature. |