DAK'93:

Partial and dynamically reconfigurable SRAM-based FPGAs (Field Programmable Gate Arrays) enable the implementation of reconfigurable systems hosting several applications simultaneously, which share the available resources according to the functional requirements that are present at any given moment. Time and space sharing strategies enabled the concept of virtual hardware, supporting the concurrent implementation of applications which would otherwise require far more complex resources. However, the performance of these applications (e.g. execution speed and reliability, activation delay) is directly influenced by the efficiency of the management strategies that allocate the logic space to the various functions that are waiting to be activated (each function requiring a specific amount of logic resources). Because the activation requests are in most cases not predictable, all resource allocation decisions have to be made online. The consequences of such working contexts are twofold: All FPGA resources must be tested regularly, to exclude malfunctioning due to the allocation of faulty elements. Since the process of launching / halting active functions takes place asynchronously at any given moment, an online concurrent test scheme is the only way of ensuring reliable system operation and predictable fault detection latency;  As the resources are allocated to functions and later released, many small “islands” of free resources are created. If these areas become too small, they will be left unused due to routing restrictions. The defragmentation of the FPGA logic space must therefore be carried out regularly, to avoid the wasting of logic resources. This paper presents a non-intrusive solution for the concurrent replication of active logic blocks (i.e. logic blocks that are being used to implement part of an active function), transferring their functionality to fault-free resources that are available in the FPGA logic space. This replication scheme is then used as the core of an online concurrent test strategy that scans the complete FPGA, reusing the available 1149.1 test infrastructure to carry out a structural test of each logic block that has just been released. The overhead of the proposed solution, in terms of the number of configurable logic resources required for its implementation, as well as its performance (e.g. the resulting fault detection latency), are quantified. Further to the test aspects, an online concurrent defragmentation strategy based on the same replication scheme is also proposed. A rearrangement of the available logic space is carried out by selectively releasing active logic blocks, with the objective of enforcing the adjacency of those blocks that share the implementation of a common function, and the creation of wider pools of logic resources that may be used to implement new functions.