Fault detection and diagnosis Sample Clauses
Fault detection and diagnosis. The core of the diagnosis unit is given by comparators which can be implemented in two different ways, by: - using a level of XORs and an OR gate to provide a single output encoding of the equality test; - using a two-rail checker TRC (with the second word which is negated); We opted for the TRC approach, which achieves the self-testing and fault-secure properties [26] although leading to a more complex circuit. In the diagnosis unit we use 10 different comparators to compare data from all the possible pairs of switch input ports. A smaller number of comparators could be used. Unless time multi- plexing is exploited, this would trade cost for diagnosis capability. The maximum number of usable comparators also depends on the number of switch I/O ports. In what follows, we will focus on the internal switches of a 2D mesh for the sake of simplicity (featuring 5 I/O ports, including the local connection to the network interface), however all irregular topologies supported by LBDR and making use of switches with at least 3 I/O ports are suitable for our methodology. Obviously, the lower the number of ports, the lower the diagnosis capability. If we denote two faults in different ports under comparison as equivalent if they produce the same output sequence in response to the same input stimuli, then our comparator and diagnosis logic is able to: - diagnose the correct position of 1 or 2 faulty channels affected by equivalent or non-equivalent faults; - diagnose the correct position of 3 faulty channels affected by non-equivalent faults; 1 - detect the presence of 4 and 5 faulty channels. Anyway, since a 5x5 switch affected by 4 or 5 faults has to be discarded, we don’t distinguish between these two scenarios. One might argue that when a communication channel fails, then the following testing phases have less inputs available and diagnosis capability reduces. In practice, this effect plays only a minor role, since a fault on a communication channel means that also (say) the arbiter of that channel should be considered faulty (unusable). So, the diagnosis capability reduces, but also the number of input ports to be checked reduces as well. 1The probability that more than two faulty channels produce the same output sequence in response to the same input stimuli is here neglected. inputs_N
