The application discloses a memory device fault detection method and a test
system, and relates to the technical field of embedded testing. The method ensures the comprehensiveness of data of a to-be-tested memory device in a fault detection process by acquiring a multi-
source data stream of the to-be-tested memory device, the multi-
source data stream at least including physical signals, protocol data and hardware state data. Then, a space-time
graph model representing a real running state of the to-be-tested memory device can be accurately constructed according to multi-
modal feature vectors of the multi-
source data stream, the multi-
modal feature vectors at least including physical feature vectors, protocol feature vectors and hardware feature vectors. The to-be-tested memory device can be subjected to automatic fault detection and
root cause analysis by using the space-time
graph model, and a fault positioning result can be accurately obtained, thereby realizing efficient and automatic analysis of fault root causes and significantly improving the accuracy and efficiency of memory device fault diagnosis under a large-scale parallel test and rapid iteration development mode.