Upper-bound computation for optimal retargeting in IEEE1687 networks

IEEE 1687 enables flexible access to on-chip instruments via dynamically reconfigurable networks. Reconfiguration allows reducing instrument access time by keeping only those instruments on the scan-path which are required for each access. To perform reconfiguration and execute commands described in instrument access procedures, scan vectors are generated in a process called retargeting. These vectors are then applied through a number of capture-shift-update (CSU) operations. Generating the optimal set of vectors w.r.t. application time is modeled as an Integer Linear Optimization Problem, which is an NP-hard problem. In the modeling, an IEEE 1687 network is represented as a sequential problem unrolled over a number of time frames, each frame corresponding to a CSU operation. A key challenge is to find the number of required CSU operations, which should be sufficiently high so that the optimal solution is included in the search space but kept as low as possible to keep the model less complex and thus suitable for large IEEE 1687 networks. In this work, we propose a method to compute an upper-bound on the number of required CSU operations. Through experiments, we show that our method results in a tight upper-bound, is applicable to a large variety of IEEE 1687 network designs, and is able to handle large designs.