Abstract: Disclosed is a semiconductor fabrication method. The method includes forming a gate stack in an area previously occupied by a dummy gate structure; forming a first metal cap layer over the gate stack; forming a first dielectric cap layer over the first metal cap layer; selectively removing a portion of the gate stack and the first metal cap layer while leaving a sidewall portion of the first metal cap layer that extends along a sidewall of the first dielectric cap layer; forming a second metal cap layer over the gate stack and the first metal cap layer wherein a sidewall portion of the second metal cap layer extends further along a sidewall of the first dielectric cap layer; forming a second dielectric cap layer over the second metal cap layer; and flattening a top layer of the first dielectric cap layer and the second dielectric cap layer using planarization operations.

Abstract: A compute node, a failure detection method thereof and a cloud data processing system are provided. The method is adapted to the cloud data processing system having a plurality of compute nodes and at least one management node, and includes following steps: performing a self-inspection on operating statuses of services being provided and resource usage statuses, and reporting an inspection result to the management node by each compute node; dynamically adjusting a time interval of a next report and informing the management node of the time interval by the compute node; and checking a report condition of the inspection result according to the time interval by the management node, so as to determine whether the compute node fails.

Abstract: A compute node, a failure detection method thereof and a cloud data processing system are provided. The method is adapted to the cloud data processing system having a plurality of compute nodes and at least one management node, and includes following steps: performing a self-inspection on operating statuses of services being provided and resource usage statuses, and reporting an inspection result to the management node by each compute node; dynamically adjusting a time interval of a next report and informing the management node of the time interval by the compute node; and checking a report condition of the inspection result according to the time interval by the management node, so as to determine whether the compute node fails.