Confidentiality Preserving Integer Programming for global routing

Cloud computing for EDA requires a client to send problem instances containing confidential design information to an untrusted distributed network. To preserve the design information in such a framework, this work focuses on obfuscating the global routing problem modeled as an Integer Linear Program (ILP) for large industry benchmarks. Multiple transformations are introduced in a proposed framework in which the client masks the ILP instance before it is sent to the cloud. The cloud solves the masked instance and the client unmasks the generated solution. No approximations are involved in this process. The masked instance is shown to be substantially more immune to various introduced attacks. Otherwise layout statistics and even detailed connectivity information can easily be deciphered. When applying the transformation, the increase in immunity can be traded off with the induced runtime overhead.