A data-driven method for error detection and fault diagnosis in proces
sor arrays is proposed under the assumption that data streams can only
be inserted and observed through the boundary processors. The method
consists of attaching tags to data streams, thereby allowing the data
items to carry their own control and error information. Our goal is to
detect the malfunction of a specific processor at a specific time ste
p. A tag, which initially contains control information to activate a t
esting process, is changed to indicate the occurrence of an error by a
checking processor detecting an inconsistency. To pinpoint a faulty p
rocessor, the front-end computer must go through the reverse process o
f identifying the processor that detected and signalled the inconsiste
ncy. Using two data streams, we can control every processor in the arr
ay and locate the faulty one. The resulting processor array is regular
in structure, and the number of bits used to encode the control and e
rror information is independent of the size of the array, thus leading
to efficiency and scalability. (C) 1997 Elsevier Science B.V. (C) 199
7 Elsevier Science B.V.