Software Defect Prediction for LSI Designs

While mining software repositories is a field which has greatly grown over the last ten years, Large Scale Integrated circuit (LSI) design repository mining has yet to reach the momentum of software´s. We felt that it represents untouched potential especially for defect prediction. In an LSI, referred to as hardware later on, verification has a high cost compared to design. After studying existing software defect prediction techniques based on repository mining, we decided to adapt some for hardware design repositories in the hope of saving precious resources by focusing design and verification effort on the most defect prone parts of the design. By focusing our resources on the previously mentioned parts, we hope to improve our designs quality. We discuss how we applied these prediction techniques to hardware and show our results are promising for the future of hardware repository mining. Our results allowed us to estimate a possible total verification time reduction of 12%.