Test scheduling for core-based systems

Abstract

We present optimal solutions to the test scheduling problem for core-based systems. We show that test scheduling is equivalent to the m-processor open-shop scheduling problem and is therefore NP-complete. However; a commonly-encountered instance of this problem (m = 2) can be solved in polynomial time. For the general case (m > 2), we present a mixed-integer linear programming (MILP) model for optimal scheduling and apply it to a representative core-based system using an MILP solver: We also extend the MILP model to allow optimal test set selection from a set of alternatives. Finally, we present an efficient heuristic algorithm for handling larger systems for which the MILP model may be infeasible.

An Integrated Framework for the Design and Optimization of SOC Test Solutions

“…In this example, bw(r 1 )=6 and bw alloc (r i , τ 1 , τ 2 ) = 4 since wire w 2 and w 3 are used by t 3 and t 5 and w 5 and w 6 are used by t 2 .…”

“…The main advantage is that the same control structure can be used for all tests in the session, which minimizes the routing of control wires. In general, (6,2) t 4 (4,2) t 2 (3,5) t 3 (4,2) test (power,time) power limit = 10 systems are not tested with a BIST structure only for each testable unit and therefore a TAM is required [6,7,8,24]. Chakrabarty proposed an integer linear programming (ILP) for the allocation of TAM width [6,7,8] and the effect on test time for systems using various design styles for test access with the TestShell wrapper is analysed by Aertes and Marinissen [1].…”

“…However, such approach increases the test time of the system. For instance, consider System S [5] (Table 1) where we added the floor planning (x, y co-ordinates for each core). A minimal TAM would be a single wire starting at the TAP, connecting all cores and ending at the TAP.…”

“…Given a system as in Figure 5 where for each of the cores c 1 5 . The resource graph in Figure 1 and the test compatibility graph in Figure 2 model the system when r 3 (test source 1) is needed by t 1 and t 3 and r 4 (test sink 1) needed by t 3 and t 5 .…”

“…Several heuristics have been proposed [1,4,5,9,11,14,23,27]; however, they have been evaluated using rather small benchmarks. For such benchmarks, a technique based on Mixed-Integer Linear-Programming (MILP) can be used [4].…”